Intel is the most advanced U.S. semiconductor manufacturer and the largest by both revenue and number of employees, with revenues of $54 billion in 2023 and over 130,000 employees worldwide.1 “Fabless” semiconductor companies like Nvidia or AMD design but do not manufacture chips. “Foundries” like Taiwan’s TSMC or China’s SMIC manufacture chips on contract, but do not design them. Rather, like South Korea’s Samsung, Intel is an “integrated device manufacturer,” meaning it both designs and manufactures its own computer chips, including central processing units (CPUs), graphics processing units (GPUs), and more. Despite decades of preeminence in the computer industry strengthened by Intel’s proprietary x86 instruction set architecture (ISA) and long-term partnerships with personal computer companies like Apple and Microsoft, Intel became a dead player by the 2000s. In the face of sharply declining revenues and share prices (INTC) since 2021, the company is now attempting a radical pivot to contract manufacturing in order to become a globally competitive foundry.

An instruction set architecture is a standardized protocol or layer of abstraction for feeding software commands to a CPU, enabling compatibility of software across different CPUs. For decades, Intel’s business model has hinged on its x86 instruction set architecture, which it developed and which gained wide adoption during the personal computer revolution of the 1980s through partnerships with IBM and Microsoft. Intel helped to cement this predominance by heavily committing to backward compatibility of software with its ISA, a highly desirable feature for users but which greatly increased the hardware design complexity of making x86-compatible chips, thus making it harder to compete with Intel.2 Moreover, while other ISA designers like British rival Arm have widely licensed their ISAs to other chip designers, Intel never has, instead implicitly threatening legal action for any infringements on its ISA. The result was Intel having a worldwide quasi-monopoly on designing and manufacturing CPUs for personal computers and servers.

But Intel’s central position has significantly eroded. The mobile computing revolution since the introduction of smartphones in the 2010s is virtually entirely powered by Arm’s ARM instruction set architecture, which lacks x86’s degree of backward compatibility but is far more power-efficient. Intel also missed out on the revolution in parallel computing powered by Nvidia’s GPUs. Meanwhile, as a manufacturer, Intel has been outdone by competitors in East Asia, especially Taiwan’s TSMC, which pioneered the foundry model of chip manufacturing and is now the most technically advanced chip manufacturer in the world. Reforming Intel to become a foundry is now a task that falls to 63-year-old Pat Gelsinger, an Intel veteran who left in 2009 before returning in 2021 to become CEO.

Become a paid subscriber and get a new in-depth investigation of a key live player, institution, or industry in your inbox every Wednesday:

Subscribe now

Intel’s Products Are Becoming Outmoded

Intel categorizes its products into three overarching categories, roughly corresponding to chips for personal computers, for servers and data centers, and for network infrastructure. In the second quarter of 2024, these three categories brought in the vast majority of Intel’s revenue, 63% of which came from personal computers, 25% from servers and data centers, and 11% from network infrastructure.3 Over half of Intel’s revenue is from East Asian markets, with China making up 27% of total revenues in 2023.4 This is due to the majority of its products being chips used in personal computers and servers, which are overwhelmingly built and assembled in the region.5 Due to Intel’s reliance on selling chips for personal computers and servers, 40% of the company’s revenues is tied to just three computer manufacturers: U.S. companies Dell and HP, and Chinese manufacturer Lenovo.

The key to Intel’s core line of business is designing and manufacturing x86-based central processing units (CPUs), the central chips in any computer that execute instructions from human-written software. Intel would have a total monopoly on x86 chip design if not for the fact that, when Intel agreed with IBM to supply x86 processors for the first personal computers in 1981, it was stipulated that there should be a second supplier of such processors. The U.S. company AMD was the chosen second supplier and, despite vigorous legal action from Intel running to 1995, has maintained the right to design x86 chips. Unlike Intel, AMD is fabless and outsources manufacturing to foundries. Today, Intel has about 80% of the x86 processor market, while AMD has 20%.6 Notably, since 2019, AMD has somewhat increased its market share of x86 processors against Intel after over a decade of decline, indicating that even in its core business area Intel is facing serious competition.7

But Intel’s bigger problem is that it failed to take advantage of, let alone lead, the multiple fundamental transformations of the computing industry during the 21st century which are now obsoleting Intel’s technology. For example, in the 2000s Intel turned down an offer from Steve Jobs to manufacture processors for the then-brand-new iPhone, because Intel’s then-CEO Paul Otellini believed Apple would not sell enough phones to justify the initial development cost.8 Therefore, the initial iPhone processors were designed and manufactured instead by Samsung, eventually costing Intel untold billions in lost revenue.

Intel failed to later catch up in the smartphone revolution, in part because of its commitment to its x86 instruction set architecture, which was strictly technically inferior to rival Arm’s ARM architecture on mobile devices. Whereas x86 is a “complex instruction set computer” (CISC) where single instructions can execute multiple steps in a row, ARM is a “reduced instruction set computer” (RISC) where instructions are simplified in order to run faster and more power-efficiently, thus preserving operating time on a battery-powered device like a smartphone. ARM 9, the latest version, has 354 base instructions.9 In contrast, x86 has over 1500.10 Hardware constraints in the earlier and less advanced days of semiconductor manufacturing and design often necessitated more divergent CPU core and component designs to achieve different goals, which in turn necessitated more bespoke instructions in the ISA, that were then preserved by Intel and Microsoft’s commitments to backward compatibility, at the cost of greatly increasing the complexity of designing new x86-compatible CPUs.11

This dynamic served to entrench Intel’s leading position until the advent of the smartphone multiplied the number of valuable, battery-powered consumer devices demanding both high performance and efficiency that was only available with ARM-based chips. Unlike Intel, Arm always widely licensed the right to design chips with its ISA and, until 2006, Intel even owned XScale, a designer of ARM-based chips. But Intel doubled down on x86, developing an unsuccessful low-power mobile variant called Atom chips, and sold Xscale to semiconductor firm Marvell Technology.12 Today, Arm has an estimated 99% market share in mobile computing.13 

Losing out on the smartphone market was ignored in large part because Intel’s x86 processors saw massive growth in demand from data centers built by cloud computing companies like Amazon Web Services (AWS), but fifteen years later these too are now looking to transition to custom ARM-based chips.14 So are personal computer companies. Between 2020 and 2023, Apple fully transitioned off of x86 for its personal computers, switching to ARM. Even Microsoft is now switching to ARM for its upcoming line of “AI [personal computers].”15

Another significant error was Intel’s failure to build a discrete graphics processing unit (GPU) for high-performance computing early on. From 2008 to 2010, Project Larrabee was Intel’s attempt to make a discrete GPU, but it was shelved due to delays. Though Intel would eventually release dedicated GPUs, named Arc GPUs, in 2020, it lost the vast majority of the market to Nvidia, who pioneered the technology.16 Nvidia, as of 2024, has a 94% market share in GPUs being used in data centers for demanding AI workloads, with AMD having 4% and Intel under 2%.17 Current Intel CEO Pat Gelsinger has argued that the cancellation of Larrabee made redundant many of the acquisitions Intel later made in AI-related companies, as they all relied on Nvidia hardware for training their models.18

Intel also lost its technical lead in manufacturing the most advanced chips. As of mid-2024, the most advanced manufacturing process or “process node” for Intel is 7 nm, while both TSMC and Samsung are producing 3 nm chips.19 The failure was in part attributable to Intel deciding not to buy the Dutch semiconductor equipment manufacturer ASML’s first extreme ultraviolet lithography (EUV) machines in 2014, despite being a large investor in the company at the time.20 The decision was made by then-CEO Brian Krzanich, who believed it would not be economically justifiable.21 In 2018, Intel attempted to leapfrog TSMC and Samsung with a new and more advanced manufacturing process, but ended up failing and further delaying its planned chip releases.22

The two decades of strategic errors and failures on Intel’s part are the result of succession failure. Intel was originally founded in 1968 by luminaries like Robert Noyce—inventor of the silicon microchip—and Gordon Moore—who coined the famous “Moore’s Law”23—with alumni of the first major semiconductor company Fairchild Semiconductor, which was itself founded by former members of the laboratory of transistor pioneer William Shockley.24 Andy Grove, a junior executive at Fairchild, joined Intel on the company’s first day.25 This founding generation of the company led Intel until the early 2000s: Noyce was CEO from 1968 to 1975, followed by Moore from 1975 to 1987, followed by Grove, who remained CEO until 1998 then chairman until 2004. More than anyone, Andy Grove is credited with architecting Intel’s takeover of the global CPU market in the 1980s and 1990s. As soon as Grove retired, Intel began making mistakes. Paul Otellini, who took over as CEO in 2005, was the first Intel CEO not to be an engineer by training.26

In the absence of reform, Intel’s default trajectory would be managed decline as x86 loses adoption compared to ARM and its manufacturing operations continue to be outcompeted by TSMC and Samsung, perhaps eventually licensing x86 more broadly or even becoming a fabless chip designer and giving up on manufacturing entirely. Instead, Intel has chosen to pivot harder into manufacturing and become a contract semiconductor manufacturer for other chip designers, a foundry. This change is being led since 2021 by Pat Gelsinger, an engineer by profession who worked at Intel from 1979 to 2009, where he was mentored by Andy Grove and rose quickly as a chip designer to become the company’s chief technology officer, before leaving the company due to estrangement from Paul Otellini.27

The Plan to Expand Intel’s Manufacturing

Since 2021, Intel’s reform strategy has centered around building out the company’s foundry services, outlined in a strategy referred to as “IDM 2.0.” This involves a target to make Intel the second-largest foundry by revenue by 2030, surpassing Samsung but behind TSMC. This entails reaching $5 billion in foundry revenue for 2027 and $15 billion by 2030.28 The strategy has three parts. First, the company is spending significant sums to increase its internal manufacturing capacity for its own chips. Second, it is prepared to outsource the production of some of its chips to third parties, including TSMC, to speed up the design of its more advanced chips. Intel has, since 2021, already been relying on TSMC to produce its Arc GPUs.29 Third, Intel plans to become a significant foundry for fabless designers by 2030.30 While marketed as a novel strategy, it is a similar strategy to that of Samsung, which in 2017 split off its foundry business from its semiconductor design business.31

Gelsinger has emphasized a cultural shift at the company in his short tenure. He has promoted the adoption of “Grovian” execution. Andy Grove did much to popularize the use of “objectives and key results” (OKRs) to determine executive compensation.32 Gelsinger has also made clear his desire to have a stronger relationship with equipment suppliers, relying more heavily on them for system integration and being willing to pay higher costs for more technical assistance and preferential treatment in shipment orders. The director of Intel Foundry technology, Ann Kelleher, has been largely given free rein to spend on new technology.33 From 2021 to 2023, Intel’s research spending stayed roughly the same but increased its share of the revenue from 19% to 30%.34 Meanwhile, capital expenditure increased by $7 billion, going from 25% of revenue to 45%.35 Gelsinger began his tenure by halting share buybacks, which effectively financially reward stockholders using the company balance sheet.36

Intel currently has five major wafer fabrication facilities and five facilities dedicated to semiconductor assembly, testing, and packaging. Its wafer fabrication facilities are in the U.S., Europe, and Israel, and its assembly and testing facilities are in lower-wage jurisdictions in Costa Rica and East Asia. Intel’s largest current wafer fab operation is in Oregon, where it employs 22,000 people.37 Its second-largest is in Arizona, where it employs 12,000 people. Intel’s Ireland facility is based in Leixlip, Kildare County. The company chose Ireland as its European headquarters in 1989, with two factories today employing 5000.38 Intel also has close to 12,000 employees in Israel, beginning its operations in 1974 and building its first factory there in 1981.39 The Fab 28 facility at Kiryat Gat is capable of manufacturing 7 nm chips, currently Intel’s most advanced process node.40 Both Ireland and Israel have the requisite local workforce to meet Intel’s needs and have historically supported the company through either direct subsidies or generous tax incentives. 

In 2021, before Intel’s expansion announcements, it had the sixth-largest capacity for wafer fabrication globally, with the ability to produce nearly 900,000 wafers per month. For comparison, Samsung’s foundry had the capacity to produce 3.1 million wafers per month, followed by TSMC’s 2.7 million. The other significant manufacturers larger than Intel were Micron (1.9 million), SK Hynix (1.9 million), and Japanese memory foundry Kioxia (1.6 million).41 Intel needs to significantly expand its manufacturing capacity to become a globally competitive foundry, and it is.

In addition to expanding its existing wafer fabrication facilities in Arizona and elsewhere, Intel is building in two new locations—Ohio and Magdeburg, Germany—and an additional assembly and testing facility in Poland. The Ohio facility is expected to begin production in 2026, while the European facilities in 2027.42 Wafer fabrication plants have become enormously expensive to build due to the increasing demands of semiconductor manufacturing. A typical fab facility consists of three or four floors. The top floor is filled with fan and filter systems to keep the air in the facility free from particulates. The middle floor, the “clean room,” is where the fabrication process occurs and is filled with thousands of specialized machines. The ground and subterranean floors house pumps and electrical systems to carry electricity, gases, liquids, and waste to and from the clean room.43

The costs of the equipment used in a facility make up 80% of the overall expense.44 Based on estimates from the Institute for Progress, a fab’s cost is roughly 60% specialist manufacturing equipment from companies like Tokyo Electron, another 20% specifically for lithography machines from companies like ASML, and the remaining 20% spread across electrical systems, piping, heating, ventilation, air conditioning, concrete, steel, and site development.45 In 2022, Intel preemptively secured the 2024 stock of ASML’s most advanced extreme ultraviolet lithography machines. This totaled six machines, each costing $370 million.46 Intel’s investment in Ohio is expected to cost $20 billion.47 Because cost is tied to equipment spending, upgrading established plants is about as expensive as setting up new ones. Intel is spending $36 billion to upgrade its wafer fabs in Oregon and committed to spending $25 billion to upgrade its Israeli manufacturing base.48

In June 2024, Intel paused construction of its $25 billion fab project in Israel.49 At the same time, the Ohio plant, originally intended to open in 2025, has been delayed to 2027. The Magdeburg plant has also been delayed from starting construction in 2023 to at least 2025 due to $10 billion worth of subsidies not yet being approved by the EU Competition Authority.50 Given that Intel is not the only semiconductor manufacturer to see ballooning costs or delays in its recent fab construction projects, these problems seem not primarily related to the functionality of these companies in manufacturing semiconductors, but rather stem from construction, regulatory, logistical, and financial problems that are likely to be eventually resolved.51

While previously reporting revenues based only on product types, since 2024 Intel has split off the reporting of all of its manufacturing operations as a separate unit called Intel Foundry.52 In the second quarter of 2024, Intel Foundry reported $4.3 billion in revenue, but Intel also took off $4.3 billion in “intersegment eliminations.”53 This implies that $4.3 billion is the value assigned by Intel to the manufacturing of its own chip designs and that Intel Foundry so far has no customers besides Intel itself, which seems to comport with sparse publicly-available discussion of Intel’s foundry customers except statements like that Intel has $15 billion in “expected lifetime deal value with external customers” by 2030.54 In early August 2024, the company said it expected the “first external customer” to see production begin on its flagship process node, 18A, in the first half of 2025.55

Since 2021, Intel has variously announced that Microsoft, Qualcomm, and AWS would be future customers for its foundry services.56 Intel’s $4 billion in self-assessed and self-supplied quarterly manufacturing revenue pales in comparison to TSMC’s $18.8 billion in foundry revenue in the first quarter of 2024.57 It is higher than second-place Samsung’s estimated $3.4 billion, but this presumably counts only or mostly contract semiconductor manufacturing for external customers—rather than Samsung’s own chips—in which case Intel’s proper figure so far would presumably be $0.58

There is a logic to Intel focusing heavily on foundry capacity. First and foremost, long-term demand growth for foundry capacity is expected. One market research report estimates total foundry revenue for chips will grow from $106 billion in 2022 to $240 billion in 2032.59 Notably, 72% of current wafer fabrication capacity is in Taiwan or China.60 All of Intel’s expansions are in the U.S., Europe, or Israel, where governments perceive a need to increase capacity, meaning Intel receives significant subsidies. It can also market itself as a secure supply chain that is not vulnerable to hypothetical geopolitical shocks like a Chinese invasion of Taiwan. Financial subsidies and political pressure can push through the creation of new facilities and production lines equipped with expensive machines supplied by other functional companies.

The U.S. Government Backs Intel’s Foundry Pivot

Starting with the Trump administration in 2017 and continuing into the Biden administration since 2021, the U.S. government has taken a more protectionist stance on trade with China, especially trade in advanced, valuable technologies like semiconductors. Chinese companies like Huawei have been restricted from participating in U.S.-aligned markets while exports to China of cutting-edge semiconductor technology—like ASML’s extreme ultraviolet lithography machines—have been banned. This change is motivated partially by economic and industrial policy concerns, but more importantly by military and national security concerns. An authoritative report commissioned by the U.S. Congress and Defense Department in 2018 concluded that “if a potential adversary bests the United States in semiconductors over the long term or suddenly cuts off U.S. access to cutting-edge chips entirely, it could gain the upper hand in every domain of warfare.”61

Intel’s foundry pivot thus seems to be in large part a response to the U.S. government’s implicit and explicit offer of financial and political support for expanding semiconductor manufacturing capacity outside of East Asia, rather than an original live player strategy. While Pat Gelsinger became CEO of Intel in February 2021, joining from a decade at the software company VMware, TSMC had announced plans to build new fabs in the U.S. with U.S. government support as early as May 2020.62 It appears that Gelsinger was brought in due to his status as a successful executive and company veteran, rather than expertise in manufacturing.

In 2022, the U.S. Congress passed the CHIPS and Science Act, a sweeping piece of legislation that ordered hundreds of billions of dollars in spending on scientific and technological projects, including $39 billion in direct subsidies for semiconductor manufacturing in the U.S. The CHIPS Act was framed as a major cornerstone of U.S. industrial policy, and was unsurprisingly aggressively pushed by Gelsinger personally.63 While rivals TSMC and Samsung are receiving billions to build fabs on U.S. soil, Intel is the largest direct beneficiary.64 The bill was heavily lobbied for by the Semiconductor Industry Association (SIA), of which Gelsinger is a board member.

The CHIPS Act has given Intel an estimated $44.5 billion of total federal government funding commitments, including $8.5 billion in direct funding, $11 billion in loans, and $25 billion in tax credits.65 Separately, Intel has also received direct U.S. Defense Department funding, receiving $3.5 billion in 2024 to build military chips deemed too critical to manufacture through standard commercial deals.66 There are indications Intel could be the beneficiary of a second CHIPS act in the future, with U.S. Secretary of Commerce Gina Raimondo claiming at an Intel event that it will be necessary for the U.S. government to spend more money to competitively produce the most advanced chips.67 

Following the U.S. federal government’s lead, other governments and companies have also stepped in to provide Intel with generous funding. The Ohio state government and New Albany municipal government have committed to $2.1 billion in subsidies.68 Meanwhile, the German government has committed to $10.9 billion in subsidies for the Magdeburg facility. The Canadian financial firm Brookfield, meanwhile, has agreed to provide $15 billion in investment funding for the Arizona facility in exchange for 49% ownership.69 In total, Intel is expected to spend at least $116 billion in capital spending on its fab expansions.70 

Despite all this spending and support, there is little reason to expect Intel to surpass or even match TSMC in price or sophistication, although matching or surpassing TSMC in raw capacity may be eventually achievable with enough spending; not even Intel has set the goal of surpassing TSMC, only the goal of becoming the second-largest foundry after TSMC by 2030. The reason is that the U.S., Europe, and Israel all lack the workplace discipline, dense supplier ecosystems, lower salaries, skilled personnel, and traditions of knowledge in manufacturing that TSMC has in Taiwan. 

TSMC founder Morris Chang, now 93 years old, made headlines in 2023 for reportedly grilling high-ranking U.S. politician Nancy Pelosi at a private dinner on whether the U.S. really understood the difficulty of building a semiconductor manufacturing base.71 In Chang’s view, the push to “reshore” U.S. semiconductor manufacturing will be “a very expensive exercise in futility” because unit costs will never fall to Taiwanese levels; he cited the example that TSMC’s long-operating fab in Oregon remains 50% more costly per unit compared to Taiwanese fabs, although still profitable.72

Gelsinger has stacked his executive team at Intel with allies from VMware rather than manufacturing experts.73 Intel Foundry’s initial president was Randhir Thakur, a veteran of the semiconductor equipment manufacturing company Applied Materials and Intel’s former head of supply chains. In 2023, just two years into the job, Thakur resigned and moved to become the CEO of Tata Electronics, a subsidiary of the Indian Tata Group.74 He was eventually replaced by Kevin O’Buckley, a semiconductor industry veteran who spent seventeen years at IBM.75

Thakur was just one of several outside hires brought in by Gelsinger to improve Intel’s foundry capacity. A number of critical hires came directly from TSMC, Samsung’s foundry business, and from the Taiwanese foundry UMC.76 For example, Intel’s new vice president of customer enablement, Michael Chang, had been with TSMC for over thirty years.77 But although China’s SMIC and Samsung have in the past poached high-level manufacturing and R&D personnel from TSMC, it is unclear that Intel has poached personnel of comparable manufacturing expertise. If TSMC itself has only achieved manufacturing chips that are 50% more expensive than its Taiwanese-made chips in the U.S., Intel is unlikely to ever surpass this margin.

Gelsinger is attempting a radical reform of Intel. He has not shied away from difficult measures, including announcing the firing of 15% of the company’s workforce to cut costs in August 2024.78 So long as government support continues, it seems likely that Intel will ultimately succeed at expanding its production in the U.S. But it seems unlikely that Gelsinger will succeed at forging a manufacturing culture that goes against the general decline in U.S. manufacturing discipline visible at other companies like Boeing and matches, let alone exceeds, that of a company like TSMC. In recent years, it has taken exceptional live players like Elon Musk to found functional manufacturing organizations in the U.S., like Tesla. Although Intel may well succeed at increasing production, if its unit costs remain high regardless well into the future, Intel will not have been reformed into a functional and dynamic company, but rather a state-backed dead player not dissimilar to Boeing or defense contractors like Lockheed Martin, akin to a state-owned enterprise or “national champion” in countries smaller than the United States.

Become a paid subscriber and get a new in-depth investigation of a key live player, institution, or industry in your inbox every Wednesday:

Subscribe now

RTX Corporation, with a revenue of $68 billion in 2023, is the world’s second-largest defense contractor behind Lockheed Martin. As of August 2024, at $155 billion, it also has the highest market capitalization (RTX) of any defense company.1 The company is the result of a 2020 merger between the U.S. defense contractor Raytheon, which primarily manufactures radar and missiles, and United Technologies Corporation (UTC), an industrial conglomerate that owned businesses related to aircraft engines and aircraft electronics i.e. avionics. The resulting company was called Raytheon Technologies Corporation until its rebranding as RTX Corporation in June 2023. Today, it is organized into three totally distinct “business units.” Collins Aerospace manufactures components and avionics for the aerospace industry, Pratt & Whitney manufactures aircraft engines for both military and commercial aircraft, and Raytheon is a defense-oriented business that primarily manufactures sensors and missiles for the U.S. armed forces.

Whereas Lockheed Martin is almost entirely an arms manufacturer and three-quarters of its revenue come from the Pentagon, RTX Corporation is more of a civilian manufacturer of high-tech components that happens to sell to the military. As of 2023, 43% of the company’s revenue was dependent on international sales.2 Collins Aerospace is 50% dependent on foreign sales, Pratt & Whitney is 38% dependent, and Raytheon is 23% dependent. Of the three businesses, Raytheon is the only one that could be considered a fully-fledged defense contractor, with 75% of its revenue directly provided by the U.S. government and nearly all the rest supplied by foreign military sales. In contrast, over 60% of income for the other two divisions is based on commercial sales.

The three subsidiaries are roughly equal in terms of revenue. However, Pratt & Whitney represents the majority of the company’s backlog, with $114 billion in engine orders as of January 2024.3 With 185,000 employees, 70% of whom are located in the U.S. and of whom 57,000 are engineers, RTX Corporation is one of the key corporations of the U.S. aerospace manufacturing industrial base, alongside Boeing. But it is not an industrial conglomerate built by a live player, but rather an amalgamation of many heterogeneous organizations and separate traditions of knowledge, recently collected together in a single corporate entity by career financiers, accountants, and managers operating on the theory that the ideal firm is like a financial portfolio with stably growing financial metrics.

Become a paid subscriber and get a new in-depth investigation of a key live player, institution, or industry in your inbox every Wednesday:

Subscribe now

Raytheon Specializes in Radar, Lasers, and Guided Missiles

What primarily differentiates Raytheon from other major defense contractors is its focus on advanced radar and electronic systems. Today, the company produces the major radar systems allowing U.S. forces to detect aerial threats. This includes stationary radars at military installations, sea-based installations, and radars mounted on ships, trucks, and aircraft. Currently, the company’s largest contract is a $3.2 billion deal to manufacture 46 radar systems for the U.S. Navy.4 The second-largest is procuring transportable radars for the U.S. Army and the Saudi Arabian government.5

Radar is a foundational technology for modern military and commercial infrastructure. A radar emits electromagnetic radiation and detects the echo returned from the reflecting object. Modern radars can determine the trajectory and even the shape of the object.6 Radar systems have a frequency range between 30 megahertz and 300 gigahertz.7 Higher-frequency radars have less range and higher accuracy due to the relative shortness of their wavelengths. Different radar frequency bands are, therefore, tailored for various applications.8 Higher-frequency bands might be used for applications that demand greater accuracy, including high-resolution mapping or use in self-driving vehicles. If particular frequencies for radars can be matched by jamming systems, they can be disrupted.

The U.S. Space Force-run Global Positioning System (GPS) and equivalent systems like Russia’s GLONASS or Europe’s Galileo operate on similar technological principles to radars. GPS uses radio waves from a constellation of satellites and ground stations to accurately provide positioning information for a GPS receiver. Whereas radar is about detecting objects, GPS provides accurate locations. While Raytheon does not manufacture satellites, it does develop ground stations for GPS. Since 2010, it has been the primary contractor for the GPS Next Generation Operational Control Segment, or OCX, representing the large-scale replacement of GPS ground control systems for the entire U.S. military.

The progenitor to OCX, the Operational Control Segment (OCS), had been run by Boeing and Lockheed Martin from 1996 to 2012.9 It is a network of control stations and ground antennas, with the central control station located at Schriever Air Force Base in Colorado. With this ground-based infrastructure, modern GPS is possible. OCX was supposed to be fully operational by 2016, but due to delays, it is not expected to be deployed until mid-2025.10 The Pentagon threatened to terminate it in 2015 following significant delays and cost overruns but ultimately decided to continue funding it.11 The delay in rolling out OCX means that, of the 32 major GPS satellites used by the U.S. Space Force, only seventeen can use the latest signals or have access to the latest anti-jamming technology.12 The cost of the system, initially set at $4 billion, has increased to over $7 billion.13 In 2022, Raytheon paid $10 million to a former employee who was demoted for notifying management that his superiors asked him to falsify test data related to OCX.14 The delayed rollout of the updated infrastructure by Raytheon highlights significant failures and inefficiencies in the U.S. defense industrial base, which are not limited to individual components and weapons systems but to infrastructure essential for running virtually every electronic system in the military.

Raytheon is also the largest producer of guided missiles in the world. Raytheon jointly manufactures the Javelin anti-tank missile with Lockheed Martin; this missile is one of many the U.S. and its allies supply to Ukraine in its ongoing war with Russia. Both companies are also now the only major manufacturers of torpedoes, with Raytheon’s Mark 54 torpedo being designed for anti-submarine warfare and launched from aircraft and surface ships.15 As with most programs, these Raytheon systems rely on components manufactured from other defense contractors. For example, the Mark 54’s acoustic sensors are produced by Northrop Grumman.16 Raytheon also produces air-to-air missiles, with the capacity to produce around 1200 medium-range air-to-air missiles a year.17 The company is also the primary contractor for the GPS-guided Excalibur artillery shells.18 These shells have considerably higher accuracy and effective range but cost up to $100,000 per shell, compared to $3000 for a conventional shell.19 Raytheon has also received funding to develop hypersonic cruise missiles but so far has not fielded them.20

Much of Raytheon’s missile production is directed towards anti-air systems. It is the primary producer of portable anti-air Stinger missiles. Following the 2022 Russian invasion of Ukraine, demand from the Ukrainian armed forces led Raytheon to restart production of the system, bringing in retirees in the process. Raytheon’s most notable missile system is the Patriot air and missile defense system, which is the cornerstone for mid-range missile and air defense for the U.S. and its allies. The company manufactures the ground system, radar, and several missile variants, while Lockheed Martin manufactures other variants of the missiles used.21 Raytheon’s expertise in missile defense infrastructure has also made it a key partner to the Israeli government for its “Iron Dome” national air defense system.

These missile defense systems are technically impressive, but there are concerns that their cost per missile is becoming uneconomical, given the declining cost of airborne explosives through drone warfare. For instance, the Iranian Shahed “kamikaze drones”—which function more like guided missiles—while slow and easy to shoot down, cost only $50,000 per missile.22 An alternative technology to expensive interceptors for missile defense are “directed energy weapons” or powerful lasers. Raytheon is the most prominent corporate developer of directed energy weapons for the U.S. government. The nascent field is split roughly between high-energy lasers (HEL) and high-power microwave (HPM) systems. The former fire targeted laser beams to destroy single targets, while the latter emit high-frequency microwaves designed to hit multiple targets. The area of effect for a laser is measured in millimeters, while for a microwave weapon it is measured in tens of meters.23

Lasers have been touted for military applications for decades. The Earth's curvature limits them by only being able to fire in straight lines and, therefore, can only hit something within their line of sight; their ammunition is limited only by a consistent electric power source. Low-power lasers can be used for non-lethal purposes. They can induce nausea or burning sensations. Between 2002 and 2007, Raytheon was contracted to develop and test the Active Denial System (ADS), a vehicle-mounted microwave that could penetrate 1/64th of a human’s skin and cause intolerable burning sensations.24 The system was deployed to Afghanistan but was never used, and it was withdrawn in 2010.25

Raytheon has since developed more powerful microwave and laser weapons to destroy unmanned aerial vehicles (UAVs) i.e. drones. The U.S. Air Force is testing its High-Energy Laser Weapon System (HELWS) to destroy drones.26 The system is being mounted onto both U.S. and British military vehicles.27 While promising, major technological bottlenecks hold back directed energy systems. Pinpoint tracking of incoming threats at extended ranges for lasers is challenging due to clouds and rain and could be made more so by enemies deploying shrapnel or debris in the air.28 Beam control is another issue. A laser has to be pinpointed onto a single point of a projectile for a certain amount of time to destroy it, but when it hits, vibrations can cause jittering, which limits its penetrating power. Another major constraint is maximizing the power of the laser while limiting the size and expense of the system.29 

Pratt & Whitney is One of a Handful of Aircraft Engine Manufacturers

Pratt & Whitney serves both the commercial passenger and military aircraft market. In 2023, 48% of its revenue came from Airbus, with much of the rest likely coming from Boeing.30 The company is also the primary manufacturer of the F135 turbofan engine used by Lockheed Martin’s F-35 joint strike fighter program, and before that, it produced the F119 engine utilized by the F-22 Raptor.31 This makes the subsidiary a critical supplier to the program underpinning the Pentagon’s plans for future U.S. airpower. Pratt & Whitney is one of a handful of major aircraft engine manufacturers that can build and export the most significant engines for commercial aircraft; the others are General Electric Aerospace, British Rolls-Royce, and French Safran. While it nominally competes with these companies, it operated a joint venture with General Electric called Engine Alliance, which produced the GP7200 turbofan engine used for the Airbus A380 jumbo jets.32

From 2020 to 2029, Pratt & Whitney is expected to deliver 19% of military aircraft engines worldwide, with 70% of them being F135 engines for Lockheed Martin’s F-35.33 Pratt & Whitney and Collins Aerospace rely on Chinese suppliers and have a significant presence there. Pratt & Whitney has 2000 employees in China, primarily to maintain the engines of Airbus and Boeing aircraft operating there.34 The flagship Chinese aircraft manufacturer Commercial Aircraft Corporation of China (COMAC) buys systems from Collins Aerospace and Pratt & Whitney for its domestically-produced jetliners, though not engines, which it gets from General Electric and Safran instead. While the Chinese state-owned defense aerospace company AVIC is developing its own domestically-built turbofan engine, the CJ-1000A, it is not expected to be commercialized until at least 2030.35

Pratt & Whitney’s headquarters and historic manufacturing facilities are in East Hartford, Connecticut, but the company’s operations are increasingly dispersed today in the U.S. and abroad, with new facilities opening in recent years in Oklahoma, Florida, and even Singapore.36 The company has shown some similar manufacturing failings at its U.S. facilities as Boeing. The company lost close to $5 billion after having to recall and inspect hundreds of engines due to a manufacturing error.37 In 2015, the company expanded capacity at its subsidiary HMI in Clayville, New York, where the company makes powdered nickel to manufacture components for the PW100G engines. During the ramp-up, a contaminant was mistakenly introduced into the powder, which inspection failed to detect during that time.38 

Workers used contaminated powder between 2015 and 2021.39 This failure potentially degrades engine components and performance, resulting in potential safety issues and unplanned maintenance work across hundreds of aircraft. As a result, airlines have had to schedule sending hundreds of aircraft to Pratt & Whitney facilities, with every shop inspection taking up to 300 days.40 This is the most high-profile failure, but there have been others. In 2021, dozens of Boeing 777 aircraft were grounded following mid-air malfunctions from PW4000 engines.41 Should the company’s manufacturing discipline degrade further as Boeing’s has, this would put the functionality of Lockheed Martin’s F-35 jet fighter in question, as well as of both Boeing and Airbus jetliners, although the latter two could much more easily switch engine makers than Lockheed Martin.

Many Companies Bundled Into One Portfolio

Both RTX Corporation and its constituent parts are organizational patchworks with far-flung origins. Today, Collins Aerospace is the primary supplier of avionics, flight control systems, the nacelles that house aircraft engines, and components to the U.S. aerospace industry, alongside Honeywell. Its current incarnation was created in 2018 when UTC acquired the avionics contractor Rockwell Collins and merged it with its own aerospace division, UTC Aerospace Systems. Rockwell Collins was spun off as an independent company in 2001, from its parent Rockwell International, which originally acquired the Collins Radio Company in 1973, which itself was originally founded to make radios for the U.S. military in 1933.

Pratt & Whitney, meanwhile, spent nearly all its history since its founding in 1925 as a subsidiary of United Aircraft Corporation, which became UTC in 1975. Raytheon is the third major constituent part of RTX Corporation. It was cofounded in 1922 by the legendary American engineer and WWII-era science administrator Vannevar Bush in Cambridge, Massachusetts. Raytheon was the largest U.S. producer of magnetrons, a key component for radios, by the end of World War II, and in 1945 filed the first patent for the microwave.42 Raytheon subsequently built its own microwaves and licensed the technology to other suppliers.43 During the 1990s, Raytheon acquired other defense businesses, including the defense electronics business of Texas Instruments and the remaining assets of Hughes Aircraft, a legacy of the prominent billionaire and aviator Howard Hughes.44 During this time, Raytheon also divested its non-defense businesses, including Amana, a company focused on home appliances.45

RTX Corporation’s executive chairman is 63-year-old Greg Hayes, the former CEO of UTC from 2014 to 2020 and the CEO of RTX until May 2024.46 Hayes, an accountant by training, had spent two decades in management positions at UTC, including holding the position of chief financial officer.47 Upon the merger with Raytheon, Hayes became CEO, while Raytheon’s CEO, Thomas Kennedy, briefly became executive chairman before retiring. In 2022, Hayes moved the headquarters of the new company from Waltham, Massachusetts to Arlington, Virginia, in the suburbs of Washington, D.C.48 The move was intended to increase immediate access to government decision-makers and follows a recent trend of defense contractors including Boeing and Bechtel moving their executive leadership close to the U.S. capital, in these cases from Chicago and San Francisco, respectively.

The 2020 merger between UTC and Raytheon was driven by Greg Hayes, from his position as CEO of UTC, and represented the endpoint of a long-running reorganization of UTC from a diversified conglomerate to an aerospace and defense company. As a separate company, UTC was a disparate industrial conglomerate that was roughly split between the aerospace division, which included Pratt & Whitney, and two commercial companies: Otis, the well-known manufacturer of elevators, and Carrier, a manufacturer of refrigeration and air conditioning systems. The latter two were spun off due to activist pressure from Bill Ackman, head of Pershing Square Capital Management, and Dan Loeb of the hedge fund Third Point. In 2018, both investors called for breaking up UTC into its component businesses, arguing that these functional divisions were trading at a discount due to the perceived inefficiencies of conglomerates.49 

Meanwhile, the same year, Hayes bought the avionics company Rockwell Collins for $30 billion, which became Collins Aerospace.50 Shortly thereafter, UTC and Raytheon announced they would merge. The merger was explicitly one of convenience and risk mitigation through diversification. According to the two companies’ CEOs, the “resilience” of creating a company whose business was 50% domestic, 50% international, 50% defense, and 50% commercial was a driving factor.51 This would help the new corporation avoid cyclical financial downturns from either defense or commercial contracts through sheer size and redundancy. Since UTC was focused on the civilian side and Raytheon on the defense side, and neither built “platforms” i.e. finished high-profile products, but rather just components, the companies would be purely complementary.52 There would be no merging of operational units, just corporate leadership and administration, in an “integration lite.”53

The company’s new president and CEO since May 2024, 50-year-old Christopher Calio, is a lawyer who served as Hayes’ chief of staff at UTC.54 Calio joined UTC in 2005, presumably not long after acquiring his law degree and MBA from the University of Connecticut, and served in multiple legal positions at the company before eventually becoming chief operating officer of RTX Corporation. The corporation’s chief digital officer and head of the supply chains similarly came from United Technologies. The chief technology officer previously spent his career at General Electric, while the head of government relations, Jeff Shockey, spent most of his career at Boeing.55

Much like other major defense contractors, RTX Corporation’s central business units have their own presidents. Shane Geddy, the head of Pratt & Whitney, initially joined from General Electric Aerospace, the company’s chief competitor. Stephen Trimm, the head of Collins Aerospace, had been with the business for years. Both Trimm and Geddy are non-technical executives who graduated from business schools. Phil Jasper, the current head of Raytheon, had previously spent his entire career at Collins Aerospace. The leadership of RTX Corporation is, therefore, almost entirely made up of UTC veterans, the larger of the two groups, despite the 2020 merger being nominally one of equals.56

The reorganization of such unlike companies into one based primarily if not solely on top-level financial metrics is reminiscent of constant corporate restructurings in other industries where dead players predominate. For example, pharmaceutical companies like AstraZeneca are effectively holding companies for rapidly-changing portfolios of drugs determined by CEOs who are trying to stitch together the ideal basket of profitable assets at any given time, and whose primary tool to do so is acquiring, divesting, or merging with other companies, not long-term strategy based on technical innovation, scientific traditions of knowledge, or industrial efficiency. This theory of the firm as a portfolio of potentially if not even preferably unrelated assets turns firms into financial products aimed at appealing to passive investors, rather than mission-driven organizations seeking to raise production and lower prices in order to reap profit at great risk but potentially great reward.

Raytheon is Emblematic of the Centrally Unplanned U.S. Defense Industrial Base

The merger between UTC and Raytheon in 2020 invited scrutiny from the U.S. Department of Justice (DoJ), which has extensive powers to enforce antitrust laws on potential mergers. When mergers or acquisitions in the defense industry are proposed, the Justice Department is responsible for scrutinizing the merger, and the Department of Defense (DoD) is invited to give its own analysis. The Undersecretary for Acquisition and Sustainment, an advisor to the Secretary of Defense, can scrutinize a potential deal if it is believed to leave a particular market with only a single supplier. They can subsequently call for divestments of specific businesses or oppose the merger altogether. Ultimately, the DoJ accepted the merger, and the DoD did not oppose it.57

The DoJ review did, however, conclude that the overlap between the businesses regarding airborne radios, military GPS, and electro-optical or infrared sensors could lead to a single supplier. As a result, Raytheon had to divest its airborne radios business, and UTC was required to divest its GPS and optics business. The British BAE Systems bought Raytheon’s airborne radios business and the GPS business for $2 billion, while UTC’s electro-optics division was also divested and acquired by the Arka Group.58 These small-scale divestments are common in any defense merger. As part of UTC’s acquisition of Rockwell Collins in 2018, the DoJ demanded it divest its actuation and flight control businesses to the French engine maker Safran, a notable competitor to Pratt & Whitney.59

Personnel at the Department of Justice have a theory about preventing monopolies. Activist investors and executives trained in the legal and academic business traditions share a theory about maximizing shareholder value and targeting other financial metrics. But the Department of Defense, which is the institution charged with planning the U.S.’ military posture, seems to offer little to no input on the specific corporate structures and reorganizations of its defense contractors, despite these companies manufacturing all of the equipment needed for the U.S. military to function at all. Only after the Defense Authorization Act of 2024 was it made necessary for companies and the DoJ to provide the DoD with information linked to major mergers.60 A Government Accountability Office (GAO) report stated in 2023 that, from 2018 to 2022, of the roughly 400 mergers and acquisitions occurring each year, only around 40 even received assessment from Pentagon officials.61 

This is in large part because the intrinsically subservient relationship between defense contractors and the Pentagon means that, as long as specific “programs” for specific pieces of equipment are retained across mergers and divestments, the Pentagon does not really care who the CEO or shareholders of a company technically are or how often they change. The power imbalance between the Pentagon and its contractors is intended to ensure that weapons programs, any number of which could be considered essential to national security, are not cut or endangered unilaterally by business executives for business reasons. In this way, there is a natural complementarity between the Pentagon’s preferences and the portfolio theory of the firm subscribed to by financiers and accountants.

This dynamic takes the planning and decision-making power in the defense industrial base out of the hands of defense contractors and puts it exclusively in the hands of the Pentagon. But, having taken this power, if the Pentagon does not engage in planning, then there is simply no planning whatsoever. The load-bearing parts of the defense industry, like specific manufacturing plants or traditions of knowledge in engineering, continue on autopilot even while financiers rearrange titles and administrative divisions to create new companies, on paper, with better financial metrics. This does not result in immediate catastrophe but also eliminates the possibility for live players to successfully plan the defense industrial base to respond to likely long-term scenarios and changes in technological, industrial, or geopolitical conditions.

The Justice Department’s acceptance of consolidation with the caveat of rejecting any possible narrow monopoly means business consolidation, but specifically not in any domain where consolidation would improve efficiency or accelerate research. Even when not forced to divest, major defense companies increasingly sell assets that do not neatly fit their new portfolio or have lower margins. For example, Rocketdyne is one of only two solid rocket motor manufacturers, meaning its failures impact the entire U.S. armed forces. North American Aviation founded it, after which it was bought by Rockwell International, sold to Boeing, subsequently incorporated into Pratt & Whitney, sold again to GenCorp, merged with another supplier, Aerojet, and sold again to the military communications company L3 Harris. In 2022, it failed to deliver rocket motors for a Raytheon naval defense missile on time.62 Rocketdyne was owned by Rockwell Collins and Pratt & Whitney and is now an unreliable supplier to their successor.

The defense industry thus enjoys neither the benefits of vertical integration and scale nor the benefits of Darwinian free market competition. The resulting inefficiencies are excused and continually enabled by generous financing from Congress and little to no effective auditing of the Defense Department.63 Meanwhile, the officials in charge of various businesses are interchangeable managers who move seamlessly between rival companies. Shane Geddy, the head of Pratt & Whitney, spent most of his career at the company’s chief rival, GE Aerospace. Mark Mills, a significant program developer for Raytheon’s SPY-6 radar systems, had previously spent nearly sixteen years at Northrop Grumman, the company’s chief competitor in military radar production.64 

Official DoD policy, based on an internal 2022 report, is that it needs to strengthen merger oversight and support new entrants and small businesses in the defense supply chain.65 This shows that the Pentagon understands it has a problem with its defense industrial planning, but it is still drawing the wrong conclusions about how to solve it. Should RTX Corporation encounter financial difficulties in the future, it is likely to be pressured by the Pentagon to divest Raytheon or other defense-related businesses, or otherwise reorganize itself.

Become a paid subscriber and get a new in-depth investigation of a key live player, institution, or industry in your inbox every Wednesday:

Subscribe now

Since 2006, the Mexican military has participated in domestic law enforcement duties against Mexico’s drug cartels, large criminal organizations whose primary source of profit is the trafficking of illegal narcotics to the United States. Violence between the cartels over territorial and business disputes, exacerbated by the Mexican government’s more vigorous persecution of cartel leaders, has caused Mexico’s homicide rate to more than triple since 2007, reversing a previous long-term decline.1 The U.S. military now estimates that the cartels directly control around 30-35% of Mexican territory.2 Over eighty politicians or candidates for political office were killed in Mexico during the country’s 2021 midterm elections.3

You can listen to this Brief in full with the audio player below:

As of early 2024, despite the incarceration of leading cartel figures such as Joaquín "El Chapo" Guzmán Loera, the organization he headed, the Sinaloa Cartel, remains the dominant cartel in Mexico and is also an increasingly powerful force in drug networks across the world. Its main competitor is the Jalisco New Generation Cartel (CJNG) and the two often engage in violent competition, alongside smaller cartels like the Gulf Cartel, the Juarez Cartel, the La Familia cartel, and many more local criminal organizations. In 2017, Americans consumed $153 billion worth of banned narcotics.4 The cartels satisfy a large fraction of this demand. There are no precise estimates of cartel revenues and profits, but it is likely that annual revenues are in the low tens of billions of dollars and profits total several billion after the costs of business, including bribes. The cartels also generate revenue from other criminal activities like human trafficking, extortion, and even illegal logging.

Around the world, such criminal activities have shown to be lucrative enough and resilient enough to state persecution to fund rebellions that could topple governments. For example, the Marxist FARC guerillas in Colombia, as well as multiple generations of Taliban rebels in Afghanistan—first fighting the Soviets, then the U.S.—were funded in this way. Because of the drug war, ongoing violence, and continued influence of cartels in Mexican society, Mexico has sometimes been described as a failed state and some U.S. politicians, such as former President Donald Trump and Republican Senator Tom Cotton, have even called for taking unilateral military action against the cartels, as was done against ISIS, the short-lived Islamist statelet in Iraq and Syria.5

But Mexico’s cartels are not ideologically or politically-motivated groups making the jump to crime to fund their activities. They are rather amorphous criminal groups motivated by profit-seeking, usually relying on familial and regional ties. From a business perspective, it is preferable to collaborate with the government when possible, rather than invite anarchy. Since, through bribery, the cartels represent an important source of revenue for Mexico’s elites, this interest is mutual.

As a result, the cartels are far more like junior partners to corrupt government officials rather than an independent and competing force of their own, though their allegiances have ebbed and flowed from the state level to the federal level—Mexico is a federation of united states—and seemingly back over the last sixty years. This makes Mexico’s cartels clients of the Mexican state, not its competitors, and, in turn, Mexico’s status as a client of the U.S. explains why the cartels continue to flourish and why there is unlikely to be any U.S. intervention in the near future.

Become a paid subscriber and get a new in-depth investigation of a key live player, institution, or industry in your inbox every Wednesday:

Subscribe now

The Mexican Drug Industry

The drug trade is a notable, but not large, portion of Mexico’s economy. U.S. government estimates from the last fifteen years have settled on figures ranging from $6 billion to $29 billion annually for the amount of money going from the U.S. to Mexico for illegal drugs.6 According to the Department of Homeland Security, up to three-quarters of cartel cash revenue might never even be laundered into a financial institution by a cartel, but just stored indefinitely in cash form or presumably used to pay off others in cash.7 In 2023, a statistical estimate of the number of Mexicans working in the entire drug industry—including armed members, farmers, and chemists—reached a figure of 175,000.8

This would be just 0.3% of Mexico’s labor force of sixty million people as of 2024, while even if the drug trade brought in revenues of $50 billion annually, this would still be well below 5% of Mexico’s GDP of $1.47 trillion as of 2022.9 For reference, Mexico’s largest company, the state-owned oil company PEMEX, brought in $74 billion in revenue in 2019.10 The flagship telecommunications company of Carlos Slim’s business empire, América Móvil, brought in $45 billion in 2023.11 While the cartels derive revenue from other rackets such as domestic extortion, these are unlikely to be as profitable, on the whole, as drug trafficking to the U.S. The drug trade in its entirety is about as large as Mexico’s largest company, but the two largest cartels together are believed to employ just 45,000 members, on the high end.12 The drug trade is powerful in Mexico not because of its size, but because of its liquidity, anonymity, and informality, which makes it easy to enrich particular individuals.

Before the 2010s, the cartels primarily grew their own marijuana and heroin for export to the U.S., while also trafficking cocaine from Colombia. Legalization of marijuana in the U.S. since 2012, as well as the emergence of powerful synthetic opioids like fentanyl, has caused the cartels to shift from farming to chemistry to maintain their drug revenues. The primary exports now are apparently methamphetamine and synthetic opioids made in large cartel laboratories. From 2013 to 2020, marijuana seizures at the U.S. border fell by 81% and cartel members have described marijuana trafficking as “barely profitable now.”13 Seizures of meth have more than doubled, meanwhile, and seizures of fentanyl have grown by orders of magnitude.14

All the Mexican cartels have regional strongholds where they originate, but just the Sinaloa Cartel and the CJNG have regular national reach. According to a January 2024 assessment by the British journalist of Mexico Ioan Grillo, the Sinaloa Cartel today controls most of the northwest, while the CJNG controls much of central and southern Mexico.15 The two dominant cartels contest much of the rest of the country, while elsewhere local organizations predominate, including the major Gulf Cartel, Juarez Cartel, and La Familia. Cartel control of territory does not necessarily mean the Mexican government has been chased out, just that there is a preponderance of criminal activity associated with one cartel rather than others. Multiple cartels can be active in many areas simultaneously.

The Mexican Cartels Are Networks Not Hierarchies

The major Mexican cartels operating today nearly all descend from the Guadalajara Cartel, the dominant force in the 1980s Mexican drug trade. Like most of the so-called “cartels,” however, the Guadalajara Cartel had no clear leader nor a set hierarchy. It consisted of a shifting set of allegiances formed between a number of drug traffickers from the Mexican state of Sinaloa, their political connections in the PRI—Mexico’s dominant political party for most of the twentieth century—the federal police services, and their personal bodyguards. These traffickers had familial roots in the drug trade going back at least one generation. The leadership of today’s cartels turns over rapidly and allegiances between different organizations come and go, but the major bodies are relatively stable despite periodic inter-cartel wars.

The Sinaloa Cartel is today the leading Mexican drug trafficking organization. It is controlled by two competing factions, one led by the four sons of El Chapo—known as the “Chapitos”—the other by Ismael Zambada García, known as “El Mayo.” Now aged 76, El Mayo is the last remaining free man among the cartel old guard and has never been arrested in a lifetime of drug trafficking. His faction includes Aureliano Guzmán Loera, El Chapo’s brother. The Chapitos were largely responsible for the cartel’s move into the fentanyl market, which has opened up a new and profitable revenue stream.16 

They have also proved willing to engage in direct gun battles with the Mexican military, which the Sinaloa Cartel had historically avoided whenever possible. When Ovidio Guzmán López, one of the brothers, was briefly arrested in 2019, over seven hundred cartel gunmen dramatically deployed in the city of Culiacán, setting vehicles on fire, blocking streets, orchestrating a prison break, and taking soldiers hostage, threatening mass civilian death until President Andrés Manuel López Obrador ordered Ovidio’s release.17 Ovidio was later re-arrested in a complex military operation in 2023—which also led to cartel unrest—then extradited to the U.S., but his brothers remain in control of the cartel. Reportedly fewer than fifty people died in both incidents combined, mostly cartel members.

The Chapitos have also repeatedly engaged in violent conflict with El Mayo and his allies. José Rodrigo Aréchiga Gamboa, an El Mayo ally and head of a cartel-affiliated murder squad, was himself murdered by the Chapitos in May 2020.18 Even before the recapture and extradition of El Chapo, the Sinaloa Cartel has been an unstable entity. The death of Ignacio Coronel Villareal, a senior Sinaloa Cartel leader killed by Mexican armed forces in 2010, left a power vacuum in the Jalisco region that he managed. From the resulting power struggles emerged the Jalisco New Generation Cartel. Led by Nemesio Rubén Oseguera Cervantes, “El Mencho,” the cartel grew rapidly through extreme violence. Its forces ambushed and killed fifteen Mexican police officers in April 2015, at the time one of the single most deadly incidents in the history of the Mexican security services.19 

The CJNG also defeated the Los Zetas cartel in Veracruz, and has become the only other cartel to achieve a national presence, at least outside of Sinaloa itself and the “Golden Triangle” region of opium cultivation. In 2016, it kidnapped two of the Chapitos from a restaurant in Puerto Vallarta. After a week of negotiations between the cartels, the Chapitos were released. It is not clear what price the CJNG extracted in return.20 The CJNG was also a very early mover into the fentanyl market, but was quickly imitated by the Sinaloa Cartel.

The willingness to murder each other of members of even the same nominal cartel underscores the lack of organizational discipline and hierarchy among Mexican cartels that is typically assumed to exist for most other kinds of organizations. The process of the Guadalajara Cartel evolving into multiple competing cartels, or the CJNG arising out of the Sinaloa Cartel, is not an intentional series of organizational design decisions, but the outcome of local schemes and power struggles between criminals who fundamentally do not trust one another nor have any impersonal organizational backbone to fall back on during disputes or succession crises. This is why familial and local ties are so important to cartel activity. In the views of a major Colombian cartel boss and a Colombian attorney who represented Colombia’s drug cartels, “cartels do not exist,” but are labels applied by police and prosecutors to disparate and fractious groups of drug traffickers to more easily make prosecutorial cases.21 While a self-serving argument, it has some truth to it.

The Sinaloa Cartel leadership has historically been somewhat more averse to extreme violence than that of the CJNG, which helps it retain tacit political support. While extorting legitimate businesses in the regions it controls, the cartel also helps protect them from extortion attempts from other cartels, and from government officials trying to extract either bribes or legitimate taxes. It also polices its territories, solving some kidnapping cases and handing the culprits over to the police. During the COVID-19 pandemic, it handed out cash payments to local residents in its core territories, part of a long-running strategy to retain local support by providing social benefits in the form of cash or grants for schools and churches.22

This activity is near the limits of systematic political action or policy undertaken by the cartels that are unrelated to drug trafficking. Both cartel bosses and footsoldiers tend to come from poor and rural families—El Chapo is illiterate—and their personal views reflect this background.23 In his short video interview recorded for the Hollywood actor Sean Penn, El Chapo revealed he hadn’t done drugs in decades, said he got into trafficking because it was the only way to make money as a youth in rural Mexico, explained that the drug trade wouldn’t go away so long as there was demand for illegal drugs, and finally described himself as “a person who’s not looking for problems in any way.”24

Nazario Moreno González, called “El Más Loco” lit. “The Craziest One,” was another and much more flamboyant cartel leader who provided a glimpse into the thinking of cartel members. A leader in the previous iterations of the La Familia cartel, Moreno González converted to Protestant Christianity and made his underlings read “pseudo-Christian” self-improvement books he wrote about cartel values, as well as a book about Christian masculinity written by an American pastor.25 American-style evangelical Christianity has surged in popularity in Latin America in recent decades, especially among the poor, and in some countries is set to overtake Roman Catholicism as the dominant religion.26 

The code of conduct Moreno González promulgated banned selling drugs in Mexico, drinking alcohol, or harming women and children. Like El Chapo, he said he only trafficked drugs because of the lack of financial opportunity in rural Mexico. He was killed by Mexican security forces in 2014. Like the Sinaloa Cartel, his cartel gave money to the poor in his home region of Michoacán and he has reportedly been venerated as a saint by some in the region.27

Fundamentally, the cartels are profit-seeking rather than political operations. They do not have any source of legitimacy to rule that can compete with that of the Mexican government; insofar as they promulgate a strain of rural populism, it is to defend their own reputations as good God-fearing men among their rural peers, not to legitimize cartel activity in the abstract. Even when totally dominant on the ground in a region like Sinaloa or Michoacán, the cartels do not try to supplant existing state structures and offices, but just bribe and cooperate with them so that their illegal activities in the service of maximizing profits go unhindered. For a cartel boss, the ideal is not to replace the Mexican government, but to be left alone by it.

While the Mexican cartels are powered at all levels by direct financial patronage—not dissimilar to the Nigerian or Saudi Arabian governments—it would be impossible for any cartel to seize formal power from the Mexican government and legitimize exporting drugs to the United States as the basis of regime revenue to preserve itself. A formally separatist Sinaloa drug state would invite U.S. military intervention, including on the behalf of the Mexican government. A cartel that gave up drug trafficking in favor of formal separatism would not just lose its most profitable source of revenue but also incur all the expenses and organizational complexities of local government. It is doubtful whether the financial patronage-based structures of the cartels could survive on such narrowed profit margins. Exclusion from formal power is then a feature, not a bug, for the cartels, and it is very unlikely, though possible, for a live player to rise up among the cartels and attempt to make the jump to formal competition for power.

The Cartels and the Modern Mexican State Evolved Together

In Mexico, drug cartels have never been able to operate without clandestinely cooperating with various arms of the Mexican state to prevent their arrest or assist the elimination of rivals. This relationship is often described as cartel “infiltration” or “corruption” of the state through bribery, but it is equally a political client-patron relationship where the cartels pay tribute to Mexican officials in the form of kickbacks from drug profits. While cartels can and do run protection rackets on legitimate businesses, government officials can and do run protection rackets on cartels. Mexico is a relatively decentralized federal state with low state capacity and low salaries for public servants, so this is an appealing way for government officials to benefit from their positions. Depending on one’s rank, cartel kickbacks can be anything from a second salary to an early retirement plan, if not more.

This relationship extends to the top of the Mexican government. Genaro García Luna, formerly head of the Federal Investigation Agency—Mexico’s equivalent of the FBI—was appointed Secretary of Public Security, one of the most senior posts in the Mexican government, by President Felipe Calderón in 2006 and served until 2012. In 2019, he was arrested in the United States on charges of taking at least $6 million in bribes from the Sinaloa Cartel. García Luna is currently serving a twenty-year sentence in a U.S. prison.28 General Jesús Gutiérrez Rebollo, Mexico’s top-ranking drug interdiction officer in the 1990s, was on the cartel’s payroll and routinely passed its leadership intelligence received from the U.S.29 His arrest led to revelations of 34 other senior officers who had taken bribes. In the 1970s, testimony from informants alleged that the president, attorney general, and defense secretary of Mexico were all “aware and supportive” of widespread bribery from drug cartels at all levels of government.30

Multiple Mexican presidents have been implicated in taking money from the cartels. Raúl Salinas de Gortari, brother of the president that signed NAFTA, was acquitted in 1998 on charges of laundering money he had received from the cartels. A 1998 report by the U.S. Government Accountability Office, however, found that he had colluded with Citibank to move approximately $100 million to accounts he held in London and Switzerland.31 Investigators would eventually find over $250 million in accounts linked to Salinas de Gortari.32 Sitting President Andrés Manuel López Obrador appears to have received cartel funds for his unsuccessful 2006 presidential campaign, while former president Enrique Peña Nieto, who was in office from 2012 to 2018, was accused in U.S. court by a cartel witness of asking for a $250 million bribe from El Chapo, before settling on and receiving $100 million in October 2012, shortly before he became president.33

Government involvement in the drug trade extends as far back as the 1940s, when the trade brought in substantial foreign currency and provided rural employment through marijuana and opium poppy cultivation. Prior to the late 1960s, the drug trade was institutionalized at a state level under the auspices of men such as Governor of Sinaloa Leopoldo Sánchez Celis, who served from 1963 to 1968, and Óscar Flores Sánchez, Governor of Chihuahua from 1968 to 1974 and later national attorney-general from 1976 to 1982.34 

These state governors effectively supervised the trafficking in their areas via the protection rackets that they ran, siphoning off a portion of the drug revenues to fund the police and public infrastructure such as schools and sports fields. This bargain came with real benefits for ordinary residents: Sinaloa, despite its position as a key state for drug trafficking, had the fourth-lowest homicide rate among all Mexican states in the 1960s. By 1974, the Sinaloan state police numbered five thousand, more than ten times the number of federal police officers employed at that time in the whole country.35 The drug traffickers were effectively arms of the state governments.

In October 1969, however, the U.S. and Mexico signed their “Operation Cooperation” agreement, a framework for drug eradication programs and anti-narcotics police training in return for U.S. aid. On the Mexican side, an anti-narcotics drive offered an opportunity to reassert the federal government’s power over the increasingly independent state governors. The Federal Judicial Police (PJF), at the time Mexico’s closest equivalent to the FBI or DEA, grew accordingly in manpower and technology. By the mid-1970s it operated dozens of planes and helicopters and hosted a specialist anti-narcotics unit.36

Over the years, increased federal power stripped control of the protection rackets away from state police—who had themselves taken it decades earlier from local municipal police—and escalated them to the national level. The PJF itself, operating alongside the leaders of the first proto-cartels, became the dominant force in the drug trade. In Ciudad Juárez, for instance, the local PJF commander collaborated with Pedro Avilés Pérez, a pioneer in the use of light aircraft and narcotics warehouses in the drug trade, to extort or kill smaller traffickers. From November 1975 onwards, the PJF took a leading role in Operation Condor, a U.S.-funded crop eradication program that aimed to cut off the flow of heroin at its source in Mexico’s poppy fields.37

In the early 1980s, however, the PJF’s role in the drug trade was largely supplanted by another bureaucracy, the Federal Security Directorate (DFS), an elite secret service and intelligence agency. The DFS ran the drug trade on a territorial basis, with regional commanders taking monthly payments from traffickers in exchange for protection, and most of the money ultimately went back to the DFS head office. The protection was not just immunity from arrest: DFS agents would serve as supervisors at marijuana cultivation fields, ensure the safe passage of drugs through regional airports, and guard stash houses. One of these DFS regional commanders was Rafael Aguilar Guajardo, who led the DFS in Ciudad Juárez in the 1980s, and co-founded the Juárez Cartel only a few years later.38 Julián Slim, the elder brother of telecommunications billionaire Carlos Slim, was a senior DFS officer in the 1970s.39

The three leading personalities in the Guadalajara Cartel, at its founding in the 1980s, were Ernesto Fonseca Carillo (“Don Neto”), Rafael Caro Quintero, and Miguel Ángel Félix Gallardo.40 Carillo and Quintero were both arrested shortly after the 1985 murder of the American DEA agent Kiki Camarena, and served long sentences. This left Félix Gallardo as the most powerful player in the cartel’s operations. His success, and that of the group he led, depended on his connections to the government. Félix Gallardo had worked as a personal bodyguard to Leopoldo Sánchez Celis, the Governor of Sinaloa, and later stood as best man at the wedding of one of the governor’s sons.41 He also served on the Board of Directors of the northern regional department of SOMEX, a state-backed investment bank.42 

Félix Gallardo was a former policeman who joined the federal PJF at the age of 17. Similarly, the Los Zetas cartel was cofounded by a member of the Mexican Army’s special forces. The cartels and the security services have not infrequently been the same people at different stages of their careers. Even in recent years, the same is often true of local and state government officials. In 2009, dozens of state and local government officials, including twelve mayors, were arrested in the state of Michoacán for allegedly protecting the La Familia cartel.43

Although cartels frequently assassinate low-level government or political figures, especially local candidates who speak out against cartel violence, the rates of attrition for cartel leaders are far higher. In 2009, the Mexican government published a list of 37 wanted drug lords.44 El Mayo, now aged 76, is the last remaining free man of those on the list. The other 36 are in prison or dead. In contrast, assassinations of high-level and even mid-level Mexican officials or politicians are relatively rare. The last major presidential candidate to be assassinated was in 1994, but it is unclear whether it was a cartel-related killing.45 Unlike in Brazil, presidents of Mexico leave office safely and without being criminally investigated or otherwise persecuted afterwards.

Although apparently a few sitting members of the national lower house of the legislature have been assassinated in the last twenty years, apparently no sitting members of the upper house, the Senate, have been. No sitting governors have been killed, though the former governor of Jalisco was assassinated in 2020 and a gubernatorial candidate for Tamaulipas state was killed in 2010. Seemingly just a handful of judges have been killed since 2006, perhaps only four; Mexico had nearly 1400 federal magistrates in 2016.46 As of April 2024, reportedly at least 28 Mexican political candidates for the year’s elections had been attacked and 16 killed, but this was out of an estimated 70,000 candidates competing to fill over 20,000 offices.47

The level of violence from Mexico’s cartels against the legitimate government and even general population is still far below that of Colombia’s in the 1980s, during the heyday of drug lord Pablo Escobar, who went so far as to commit outright acts of terrorism, like bombing airplanes, despite harboring legitimate political ambitions. Mexico’s homicide rate today is less than a third of Colombia’s peak homicide rate in 1991; Colombia and Mexico in fact have had roughly the same homicide rate since around 2017.48

Although Mexico’s national homicide rate is very high by international standards, it is still lower than that of the Bahamas and far lower than South Africa’s or Jamaica’s, all three of which are popular tourist destinations.49 Within Mexico, a majority of Mexican states are apparently safer than Brazil is nationally.50 The patterns of killings and cooperation between the cartels and arms of the Mexican government suggest that the cartels and the government are not competitors for power, but on-and-off partners, with competition for power instead occurring between Mexican state governments and the Mexican federal government, or between other cliques of Mexican elites, and the cartels often being a resource that can be extracted for money.

The United States Prefers a Weak Mexico

In 2007, the Bush administration and President Felipe Calderón agreed to the Mérida Initiative, a security cooperation agreement. Under this agreement, the U.S., by 2023, had delivered around $3 billion of assistance to Mexico, mostly in the form of helicopters and other equipment for the Mexican police and military, as well as training and maintenance.51 The Initiative was replaced in 2021 by the U.S.-Mexico Bicentennial Framework for Security, Public Health, and Safe Communities, which continued much of the security assistance but also added new commitments from the U.S. side to reduce drug demand and cut the flow of American firearms to Mexico.

This spending, however, is very small in the context of a U.S. military budget of over $800 billion per year and a foreign aid budget of approximately another $50 billion. The DEA’s budget is another $2.5 billion annually, and since the agency states that its primary focus is defeating the Sinaloa and CJNG cartels, most of this sum is presumably spent on anti-cartel action.52 Over twenty years of war in Afghanistan, the U.S. spent a total of $2.3 trillion, while its closest neighbor became the primary manufacturing and transshipment point for fentanyl and other synthetic opioids.53

Although on paper the U.S. government is deeply opposed to the drug cartels, in practice it just does not care very much. American elites are not morally opposed to using drugs, but rather the opposite; whether drugs are procured legally or illegally is ultimately a quibble and American elites have led a slow-rolling, piecemeal legalization of drugs over the last ten years. The Mexican cartels are careful to prevent violence from spilling over into the United States or affecting U.S. citizens. When members of the Gulf Cartel kidnapped and accidentally killed two Americans in March 2023 in a case of mistaken identity, the cartel publicly apologized, disavowed the five hitmen responsible, and voluntarily handed them over to Mexican police.54 The statement began with the “Grupo Escorpiones” lit. “Scorpions Group” of the cartel apologizing for harming a “working mother.” The difference from how ISIS would treat Americans couldn’t be more stark. Despite the high homicide rate and tens of millions of yearly tourist arrivals in Mexico, murders of foreign tourists are rare enough to garner international media attention when they happen.

The Mexican cartels are clients of Mexican government elites, providing one of the best—if not the best—sources of remuneration during their time in office or after, which is why the Mexican government cannot and will not eradicate the cartels. Broadly speaking, Mexico’s elites, both within and without government, are formed not through industrial entrepreneurship or careers in meritocratic institutions, but through opaque political patronage and social proximity to powerful officials. The drug cartels are not a serious impediment to the personal empires of Mexico’s elites, but rather another opportunity for extraction.

The Mexican sociologist Luis Astorga has argued that the 2000 electoral defeat of Mexico’s ruling party for most of the 20th century, the PRI, caused a breakdown in patronage networks that resulted in decentralization of the drug networks in the government.55 This is plausible. The federal government’s introduction of the Mexican military to the drug war in late 2006, and the resultant rise in homicides since then, is then perhaps best thought of as a war between Mexico City and the states, or some other such intra-elite conflict, as opposed to a war between Mexico’s government and Mexico’s cartels. The PRI was never an ideologically-motivated political party, but more of a big-tent corporatist machine for dispensing patronage. Mexican political parties today still seem to be organized along these lines, as coalitions of elites held together by opaque personal economic interests.

Despite popular expectations to the contrary, even well-armed non-state actors have repeatedly proven to be weaker than ordinary states in recent years. The violence caused by the Sinaloa Cartel over the arrest of one of El Chapo’s sons is perhaps comparable to the Russian mercenary group Wagner Group’s ill-fated “coup attempt” in June 2023. Both actions were dramatic, but ultimately failed in their ostensible goals. Rather than genuine power grabs, these were perhaps just the violent protestations of a well-armed client to a well-armed patron. Although the decorum of modern bureaucratic institutions was violated, these power relations are actually based not on automated bureaucratic relationships but flexible personalistic ones.

The Mexican government is, in turn, a client of the U.S. government, which is by far the dominant economic and military superpower in the Western Hemisphere. Since the cartels do not threaten U.S. power or geopolitical interests, the U.S. government has little reason to care about them; any negative externalities are just costs the U.S. pays for maintaining its relationship with Mexico. For Mexico’s drug cartels to be definitively eradicated, either illegal drug consumption in the U.S. would have to totally evaporate, the U.S. military would need to invade and occupy Mexico, or Mexico itself would have to undergo what amounts to a political revolution and become a strong centralized state.

Blanket drug legalization in the United States notwithstanding, only the latter option is ideologically compatible with the preferences of U.S. elites. Such a strong Mexican state would, however, have interests often at odds with those of the U.S., and thus would likely not receive support in a serious drug war. With over 120 million people, its own oil reserves, and access to two oceans, a strong and centralized Mexican government would pose a unique land-based military risk to the continental United States; the same strength and centralization that would make it possible for the Mexican government to eradicate the cartels would also make it possible to develop its own economic and security preferences that diverge from those of the U.S. It might further develop and promote ideologies that challenged U.S. norms or found root among the large population of Mexican Americans. As a result, insofar as cartel influence and violence contributes to Mexico’s backwater status, this is a feature, not a bug, for U.S. policy towards Mexico and the cartels, and is unlikely to change in the near future. Insofar as American elites tacitly support the continued legalization of drugs, the days of Mexico’s cartels are numbered anyway.

Become a paid subscriber and get a new in-depth investigation of a key live player, institution, or industry in your inbox every Wednesday:

Subscribe now

The population of India is thought to have surpassed China’s in April 2023, making India the world’s most populous country with a population of over 1.4 billion people.1 This is more than the combined populations of North America and Europe. This huge population, together with the country’s arsenal of nuclear weapons and significant regional military influence, has led many, over the decades, to predict that India will become a global superpower rivaling China and the United States. Despite long-term economic underperformance relative to China and other high-growth East Asian countries, India is often heralded as a potential global economic powerhouse due to its relatively youthful population and lowered fertility, resulting in a lower dependency ratio and an expected economic “demographic dividend” from the higher ratio of productive working-age people.2 The median age as of 2023 is 29.5 years old, compared to 38.5 in the United States and 39.8 in China.3 

You can listen to this Brief in full with the audio player below:

The total fertility rate, at 2.07 children per woman, is very close to the replacement rate of 2.1 children per woman—much higher than China’s rate of 1.09 or the United States’ rate of 1.66.4 The highest projections predict over 2 billion Indians by 2100, with India remaining the most populous country in the world, perhaps followed in the rankings by African countries like Nigeria. But these figures bely a much less certain future. India’s total fertility rate has fallen rapidly over the last decade and is likely to be well below replacement in the next decade. Indian elites have historically favored population control, and still do, with outright sterilization long being one of the most widespread forms of contraception. Rather than reaching 2 billion people by the end of the century, the population is more likely to peak at 1.5 billion by 2050 and begin falling thereafter, and will plausibly be below 1 billion by 2100. 

Despite India’s projected rapid aging and fertility crash, its demographics should not be a barrier to industrialization. Over 70% of the workforce is still rural and over 40% continues to work in low-value agricultural jobs.5 Underemployment is the norm rather than the exception across the Indian workforce, and female labor force participation is still only around 30%.6 This labor reserve numbering in the hundreds of millions is sufficient to drive industrialization if the institutional factors blocking Indian development are resolved. There is also plenty of room for catch-up growth: per capita incomes, adjusted for purchasing power parity, were a little under half of those of China as of 2019.7 But should India successfully mobilize this workforce, industrialize, and become wealthier, all previous cases point to even a further reduction in fertility and as a consequence a much-accelerated graying of the country.

Become a paid subscriber and get a new in-depth investigation of a key live player, institution, or industry in your inbox every Wednesday:

Subscribe now

India’s Population is Likely to Peak Soon

India is something of a demographic anomaly. It has the high marriage rate and early age at first birth of a low-income high-fertility country—the average Indian mother gives birth to her first child at just over 21 years of age—but the lower fertility rates characteristic of much richer nations. Fertility rates in the richest Indian states, such as Goa, are very similar to those in notoriously low-fertility countries such as Japan. India’s low-fertility states are disproportionately in the country’s richer and more industrialized southern regions, while the high-fertility states are typically poorer, more rural, and located in the north.

Bihar, the highest-fertility state, with a total fertility rate of 3.0, fits this description, as does Meghalaya, which has the second-highest at 2.9. Uttar Pradesh, India’s most populous but second-poorest state, still has a total fertility rate of 2.4.8 Nevertheless, fertility rates across all states have fallen rapidly over the last three decades. In 1993, the fertility rate in Uttar Pradesh was 4.8, 4.0 in Bihar, and even 3.0 in Delhi. Today, Delhi has a total fertility rate of 1.6, comfortably below the replacement rate and close to the fertility rates of the United Kingdom and the United States.9

Indian women have come to accept the idea that the ideal number of children is no more than two. In one survey, 72% of women born after the year 2000 report that their desired number of children is two, and another 19% want either one child or zero. Just 9% of women report wanting three or more children.10 The current “wanted fertility rate” is just 1.6, and is consistently lower than the actual fertility rate across every religious group, every level of education, level of wealth, and caste.11 Ideal fertility was still above 2.5 for women born in the late 1960s, but has fallen steadily since and has consistently been lower than India’s actual total fertility rate, probably due to limited access to contraception, especially in poorer rural areas. In the United States and many European countries, the situation is completely reversed: women consistently report desired fertility that is both above replacement and higher than the actual total fertility rate. In the U.S, for instance, the desired fertility rate is 2.7.12

The social shift to older motherhood and a rising age at first birth that has been characteristic of Western fertility transitions is nowhere to be found in India. In a sense, this should not be surprising: Indian women still have low labor force participation rates and the 2011 census recorded a substantially lower literacy rate among women (65.5%) as compared to men (81.3%). Economic underdevelopment has meant that Indian women have not adopted delayed childbearing in response to rising education levels, greater labor market opportunities, and the availability of fertility assistance technologies such as IVF. In fact, sterilization, rather than purely economic factors, is instead doing most of the work in lowering Indian fertility rates.

Around 3.5 million Indians are sterilized every year, the overwhelming majority being women.13 Vasectomies today are just 1.1% of sterilizations in India, as of 2020-21.14 The most common procedure for sterilizing women, tubal ligation, or having one’s “tubes tied,” is generally considered permanent and irreversible. State governments organize makeshift sterilization camps and also run incentive programs for sterilization, funded in part by the national government. The state of Madhya Pradesh, for instance, pays women undergoing post-birth tubectomy 3000 rupees, about 12 days’ pay for a rural laborer.15 Incentives are higher still in districts with higher fertility. In 2014, sixteen women died at a government camp in Bilaspur after undergoing sterilizations, and such camps continue to operate today despite court rulings ordering the government to cease operating them.16

In the last year before the COVID-19 pandemic, the number of sterilizations performed in the year fell to 3.3 million, but the state attributes this decline largely to greater uptake of long-acting reversible contraception methods like intrauterine devices (IUDs).17 These are becoming more popular, especially among more educated women.18 In 2020-21, the last year for which data are available, India conducted 2.4 million sterilizations, a -25% decline on the year before due to the effect of lockdowns.19 Despite this, the number of IUD insertions actually rose slightly in this year. 

Declining sterilizations are unlikely to be evidence of a resurgence in high-fertility ideals, but just evidence of adoption of more modern Western contraceptive techniques, which have proven just as if not more effective at preventing births, though they are notably reversible. As India becomes richer, sterilization activity might move away from camps towards hospitals, but efforts to enforce a 2016 Supreme Court ruling that ordered the government to phase out operation of sterilization camps seem lax at best. The conditions in these camps remain low-quality. While national guidance limits the number of sterilizations per camp to thirty a day, there are recorded occasions of over one hundred sterilizations taking place per day.20 

Local officials periodically come under intense pressure to ensure that sterilization rates remain high. In 2020, an official circular from the Madhya Pradesh state government threatened the members of its Health Mission with financial penalties or firing if they failed to meet male sterilization targets.21 Financial incentives offered for vasectomies are actually higher now than those offered for tubal ligation, despite the lower risk involved in the former procedure, but uptake remains low. The now-widespread presence of Accredited Social Health Activists (ASHAs) across rural villages has been a key tool for promoting female sterilizations. ASHAs are local women who are not paid a regular salary but are themselves financially rewarded per sterilization that occurs in their village. Notably, payments for sterilizations are three times higher than those that ASHAs receive for delivering maternal healthcare services.22

The UN’s “medium variant” population projection projects that India’s total fertility rate stabilizes at around 1.8 by 2050 and falls only slightly thereafter, to 1.7 by 2100. In this scenario, the population would peak at around 1.7 billion before falling back to a little over 1.5 billion by 2100.23 The percentage of Indians aged 65 or older would reach 30% by the end of the century, and older Indians would outnumber those under 25.24 This scenario, however, seems unrealistic in the face of the low fertility desires of Indian women and the potential for further downward pressures on fertility to come into play as the economy industrializes, such as increased secondary and tertiary education access and high female labor force participation rates. 

It is much more likely that fertility rates will continue to decline rapidly towards 1. In Singapore and Malaysia, both much higher-income countries, Indians have very low fertility rates—around 1.2—that are very comparable to the fertility rates of ethnic Chinese in these nations.25 In this scenario, India would age much faster and older Indians might outnumber those under 25 much sooner than the UN projects. India’s window to get rich before it gets older is likely to be shorter than currently thought.

The UN’s “low variant” population projections, therefore, are likely more credible. India’s population in this scenario would peak at a little over 1.5 billion by 2050, and fall to under a billion by 2100.26 Even this scenario may be too optimistic. All Indian demographic data is outdated since the last census was undertaken in 2011. The planned 2021 census was canceled due to the COVID-19 pandemic and has been repeatedly delayed since. Updated accurate estimates of fertility therefore rest on the National Family Health Surveys, which may not accurately sample the population. Previous censuses have been affected by inflated population estimates in at least some states due to local attempts to extract more resources from regional and national governments, a similar issue to that affecting Nigeria’s demographic data.27 

Moreover, the total fertility rate is a synthetic measure that extrapolates from present birth rates and previous cohort fertility trends to the projected lifetime fertility of women currently still in the childbearing years. It is not, therefore, robust to changes in the timing of fertility: in a country where the fertility window is expanding and women are shifting towards later motherhood, the total fertility rate will underestimate actual completed cohort fertility. This effect is visible in U.S. data, where despite years of below-replacement total fertility rates, actual completed cohort fertility has been above replacement until very recently.28 In India, however, due to the effect of sterilization, fertility windows are actually declining, so Indian total fertility rates may actually overestimate the completed cohort fertility of Indian women still in their childbearing years. Women born in 1980-84 bore their last child at an average age of just 25.7.29 It appears difficult if not impossible for India’s population to ever reach 2 billion outside of a complete policy volte-face.

One final factor that may serve to keep India’s population growth below even the lower end of mainstream demographic projections is rising involuntary infertility. Secondary infertility is defined as the percentage of reproductive-age women who have given birth before, express a desire for another child, are not using contraception, are in a stable union, and have had no children in the last five years. It has risen from just under 20% in 1993 to over 28% by 2016.30 In contrast, the U.S rate of secondary infertility, by this definition, is around 6%.31 Obesity, which lowers fertility, has increased dramatically in India over the last several decades: 13% of the adult population aged 18-54 were classified as obese by body-mass index in the fifth wave of the National Family Health Survey, and over half were classified as abdominally obese.32

Moreover, rising obesity is likely to interact with other India-specific factors that lower fertility, such as the high prevalence of vegetarian diets. Around 40% of Indians abstain entirely from meat and over 80% consciously limit their meat intake, in accordance with the traditional practices of Hinduism.33 Vitamin B12 deficiency is both extremely common among vegetarians and is also linked to fertility issues in men and women, decreasing sperm quantity and motility in men while increasing miscarriage risk in women.34 Absent a national program for B12 supplementation, natural infertility is likely to increase, in part due to the effects of environmental factors that affect fertility, such as pesticides and industrial air pollution, both serious problems in India across rural and urban areas respectively. On the other hand, meat consumption levels are still very low at just 15% of the global average, but if they rise to match global norms, the health of Indians can be expected to improve considerably.35

The ubiquity of sterilization makes forecasting India’s future fertility rates difficult. In theory, India perhaps has the most room for policy maneuvering to increase fertility rates of any large country: by simply ending government support for mass sterilization, the state could vastly increase the reproductive window of Indian women. Such a policy volte-face, however, seems very unlikely in the face of the popular acceptance of low-fertility ideals and a secure elite consensus in favor of population control.

Indian Elites Have Favored Low Fertility For Decades

The decline of fertility in India has been accelerated by an elite population control consensus dating back to the 1950s. This consensus owed its origins both to the influence of foreign institutions such as the World Bank and the U.S. government, but also to domestic upper-caste elite concerns about lower-caste and Muslim fertility. As far back as 1940, the dominant, socialist-leaning Congress Party’s National Planning Committee, chaired by Jawaharlal Nehru—a high-caste Brahmin, and later India’s first prime minister upon independence—had sponsored a report by Radhakam Mukherjee, another Brahmin, calling for the sterilization of “the entire group of hereditary defectives” and the propagation of anti-natalist propaganda aimed at the lower castes.36 

Mukherjee also endorsed the removal of barriers to intermarriage between different elite castes as part of an explicit eugenics program.37 The post-independence National Planning Commission continued this focus, as when in 1951 it received a report from its population sub-committee advocating free sterilization and contraception. The post-independence elite was not immediately uniformly in favor of population control: Mahatma Gandhi opposed birth control, as did Rajkumari Amrit Kaur and Sushila Nayar, who were both close allies of Gandhi. Nayar further led the Ministry of Health until 1957. Kura was a Christian and did not share the Brahmin concern with controlling the fertility of the lower castes. Between these divisions and the weak state capacity of the new Indian state, no immediate tangible national commitment was made on the question of population control.

Frustrated by the lack of progress, some local elites took matters into their own hands. In 1959, the chief secretary then-Madras state, R. Gopalaswami, was paying almost a tenth of per capita GDP to individuals for sterilization.38 By 1960, Maharashtra state was hosting camps for mass vasectomies, which were preferred over female sterilization due to the ease of the operation. Explicit demographic targets at a national level, however, did not enter the picture until the Fourth Five-Year Plan from 1969 to 1974, which targeted a reduction in the birth rate from 39 to 32 per thousand population by 1974. Explicit targets for sterilization and copper coil insertions were set, and a dedicated Department of Family Planning was established.

Further impetus towards population control policies was provided by Western elites. The Rockefeller and Ford Foundations both backed The Population Council, a non-governmental organization set up by John D. Rockefeller III, which in turn founded the first UN population research center in Bombay in the early 1950s.39 The Population Council’s concerns were more straightforwardly economic than eugenic: it viewed uncontrolled population growth as a threat to the productivity of developing nations. The foundations themselves also worked directly in India: Ford employed seventeen full-time consultants advising the Indian government by 1966.40 Although U.S. President Dwight Eisenhower would not support developing world population control policies openly, fearing a domestic political backlash from social conservatives, President Lyndon Johnson was persuaded of its merits by Robert Komer, his national security advisor, who in turn had been influenced by the work of Stephen Enke, a Ford Foundation consultant.41

Johnson, therefore, initiated a policy of making food aid conditional on domestic political change, including population control. To his aides, Johnson defended the policy, arguing he was “not going to piss away foreign aid in nations where they refuse to deal with their own population problems.”42 The conditionality of aid remained in the face of a severe famine throughout Bihar and Uttar Pradesh in 1966. As famine grew in these states, so did contraception uptake, as women grew increasingly desperate to claim the incentive payments available, despite the risks of perforated uteruses and infection from the dirty conditions in which the procedures were often undertaken.43

In this, the United States was assisted by other actors, many of them dependent on U.S. funding, such as the World Bank, USAID, and the United Nations. All employed full-time staff that worked closely with the health ministry on incentive programs for sterilization and IUD uptake. The domestic political appetite for population control, however, also remained high, especially due to the interest of Indira Gandhi, Nehru’s daughter and Prime Minister from 1966-77 and 1980-84. Gandhi and her son Sanjay implemented a crash mass sterilization program during “The Emergency,” a period from 1975-77 in which civil liberties were suspended and Gandhi was legally authorized to rule by decree. About 8.3 million sterilizations were performed in a year, an increase from 2.7 million in the year prior to the proclamation of the Emergency, most of them vasectomies and many of them coerced.44 

The popular backlash against the Emergency-era coercive sterilization program forced subsequent governments to abandon the most violent elements of the scheme, but mass sterilization in camps continued, both of a voluntary and quasi-coercive nature. Over time the focus has shifted towards women, in part a legacy of the Emergency’s focus on male vasectomies. Despite the eugenic and upper-caste elite origins of India’s drive towards population control, there seems to be very little remaining emphasis on raising upper-caste fertility, which has declined along with those of less economically-successful social groups, such as Muslims and lower-caste Hindus. Amidst the sub-sample of Brahmins in successive waves of National Family Health Survey, the total fertility rate has declined from over 3.0 in 1993 to below replacement today.45

The rate of decline, however, is actually lower than that observed in some lower-caste groups like Marathas, a group that draws heavily from the rural peasantry and lower-class tradesmen in Maharashtra, India’s second-most populous state. The Maratha fertility rate was over 5.0 in 1993, but was only just above replacement in the fourth wave of the National Family Health Survey, conducted in 2016.46 A similar pattern is observed across other lower-caste groups: although their fertility rates were and are higher than those of upper castes, the rate of decline is somewhat faster and all groups are quickly converging to sub-replacement fertility. While Hindu nationalist politicians have occasionally raised the alarm about the slowly-falling percentage of the population that is Hindu, there seems to be virtually no modern pro-natalist movement amongst India’s upper castes.

Current Prime Minister Narendra Modi, who is 73 years old and leads the Hindu nationalist Bharatiya Janata Party (BJP), has publicly endorsed small families and population control.47 He is himself childless. Modi’s eagerness to secure visa deals in trade negotiations with foreign nations such as the United Kingdom is likely also motivated by his underlying belief that India is overpopulated and this overpopulation serves to constrain its development. The Hindu leadership in BJP-voting states such as Uttar Pradesh have proposed barring persons with more than three children from public sector jobs and limiting the access of such families to welfare.48 Across a number of states, including Modi’s home state of Gujarat, persons with more than two children are legally barred from contesting local government elections. Similar legislation exists in Rajasthan, Andhra Pradesh, Odisha, and Telangana.49

Modi’s support for population control, and that of the BJP, is likely accentuated by the fact that the Muslim fertility rate remains elevated and is still above replacement at 2.4, although it has fallen dramatically from 1993, when each Indian Muslim woman had on average 4.4 children.50 Indian Muslims have also had lower child mortality rates than Hindus, a pattern observed over many decades despite their lower average socioeconomic status.51 Perhaps rather than achieving high Hindu fertility to maintain a demographic balance, the standard Hindu nationalist preference has been to achieve low Muslim fertility. Given the difficulty of raising fertility rates, this is likely to remain the preference. Muslims today are estimated to be about 15% of India’s population, up from around 10% in the 1950s.

A Vast Labor Force Remains Despite Falling Fertility

The “demographic dividend” is the boost in economic productivity in a country with a high proportion of working-age people relative to both young and old dependents. If other factors are favorable, the demographic dividend can be one of the forces that contribute to the industrialization of a poor country like India. Sustained low fertility eventually leads to a high dependency ratio as formerly working-age people become old dependents, and the few young dependents become a smaller cohort of working-age people. 

Every major developed country has seen its working-age proportion of population decline for years if not decades: according to the OECD, which defines working-age as between the ages of 15 to 64, the working share in Germany peaked in 1986, in Japan in 1992, and in the U.S. in 2007.52 China and South Korea peaked in 2009 and 2016, respectively. The only major country that is still rising rather than declining is India. While every other major developed country has an elderly share of the population between 15%-30%, India’s share is only at 6.9% as of 2022.53 India’s demographic dividend is expected to peak in 2041 but last until 2055 or so.54

Between the predominantly rural workforce and the low rates of female labor force participation, a large majority of Indians today are either unemployed, not looking for work, or stuck in casual labor with no social security benefits or employment rights. Around 90% of the employed workforce was in this kind of casual labor as of 2018.55 In an economic sense, many of these persons may as well not exist given how unproductive they currently are. Internal migration will therefore ensure, for many decades to come, a large flow of additions to the portion of the labor force doing useful work, a phenomenon that has also defined China’s economic rise. 

For comparison the internal migrant population in China is estimated to have reached 376 million.56 With well over a billion people, India most likely has plenty of talent to fill out globally-competitive industrial organizations, even should it turn out not to have as much such talent as, for example, China. India’s enrollment ratio in tertiary education is 32%.57 The country has far more graduates than it has graduate-level jobs for. 

The main obstacle to Indian industrialization is rather India’s political economy, which impedes live players and thus slows down the emergence of functional economic institutions and industrial empires. In China, South Korea, and Japan, governments worked closely with industrial leaders and new entrepreneurs to help build domestic companies that could eventually compete globally through better economies of scale, preferential regulatory treatment, technology transfer, and, eventually, clearly superior products. This is what distinguishes these industrial success stories from poorer countries like Mexico, Brazil, or Vietnam, where manufacturing growth has come on the back of foreign industrial empires moving production facilities to those countries, but there has been no concerted effort to support and build domestic industrial empires led by domestic live players, while economic elites are minted through political patronage rather than industrial entrepreneurship.

As a result, India’s industrial trajectory remains uncertain. India has much of the labor market regulation more typical of upper-middle-income and rich countries, such as minimum wages and limits on the maximum number of hours per shift. Legislation such as the Industrial Disputes Act requires employers with more than 100 employees to seek government permission for any layoffs, although enforcement of this and other restrictive regulations is lax at best and varies substantially by state.58 Legacies of the socialist “permit Raj” post-independence era, these regulations are a disincentive to formal labor market participation and firm expansion. There are however some recent moves away from this default: Tamil Nadu’s state government expanded the maximum permitted number of hours per shift from 8 to 12, in response to pressure from Foxconn and other manufacturers.59 

The manufacturing share of the Indian economy has grown over time, and now stands at around 30% of GDP by some measures. The overall growth rate has however been fairly weak and employment growth in the sector has been sluggish.60 Indicators such as electricity consumption and power demand reaffirm the impression of real but relatively slow development: the twenty-year average of power demand increase is around 5%, but 2023 saw much higher power demand growth of around 10% for the year.61 This kind of growth is much closer to the historic growth in power demand that China saw during its takeoff years of industrialization, when power demand grew by 13-15% a year. Although it remains unclear if India can shift towards this kind of growth trajectory, demographics will not fundamentally shape the path of industrialization. 

Instead, if it happens, industrialization will further reshape the country’s people. The “demographic transition” towards lower fertility rates and older societies as countries become more affluent is unlikely to see an exception in India, which is already seeing declining fertility rates despite being relatively poor. The strong emphasis on education and schooling is likely to further suppress fertility in its own right.62 The country’s vast labor force means that any external correction is unlikely to come anytime soon. Fewer than one billion Indians by 2100 isn’t just plausible, it’s likely. Together with lower-than-previously-expected growth in African countries like Nigeria, and still crashing U.S. and Chinese fertility rates, humanity might have a far less numerous future than has been commonly assumed.

Become a paid subscriber and get a new in-depth investigation of a key live player, institution, or industry in your inbox every Wednesday:

Subscribe now

Y Combinator (YC) invests in, educates, and promotes early-stage companies, especially in software. Since its founding in 2005, the startup accelerator’s educational program has helped birth companies including Airbnb, Reddit, DoorDash, Stripe, Twitch, Coinbase, Dropbox, Instacart, and FlexPort, and over a thousand others. YC estimates the total valuation of these companies at over $600 billion, meaning the organization’s profits on these investments are likely in the tens of billions of dollars. Through its mentoring and prestige, the accelerator has set much of the professional culture and institutional practices of software startups. Through the essays of YC founder Paul Graham, as well as through its influential online discussion forum Hacker News, it has also exerted and continues to exert a wider cultural and intellectual influence on Silicon Valley, which is the core driver of U.S. economic growth and technological dynamism in the present day.

You can listen to this Brief in full with the audio player below:

The core of Y Combinator is a three-month semistructured program for the founders of young companies. Entrepreneurs work intensively on their companies while immersed in a social environment that instills the explicit lessons and illegible norms of YC’s tradition of knowledge.1 Entrepreneurs are sorted into “groups” and smaller “sections” to form deeper bonds. The instructors, themselves former entrepreneurs and alumni, provide mentorship and advice during “office hours.” Participants attend talks by founders of successful companies. This experience inducts participants into being “YC alumni,” serving as a vetted network of people with similar outlook, skills, and shared experiences, much like a 20th century American college fraternity.

A company finishes the program with its sales pitch to investors at “Demo Day,” where the graduate is all but guaranteed offers of large investments at favorable terms. Software culture interprets acceptance into Y Combinator as a strong credential, so YC companies have a much easier time wooing investors, hiring employees, and sometimes even acquiring customers on the basis of this endorsement. There are now other “startup incubators” and “startup accelerators,” but none have even a fraction of YC’s prestige, nor of its track record of successful graduates.

Venture Capital and the Valuation Game

Y Combinator is a combination of a venture capital fund on the one hand and an education and accreditation program on the other. As an early-stage venture fund, its business is investing money into young and untested companies in return for a substantial ownership share. The theory is that most of an early-stage venture capital firm’s portfolio companies will go bankrupt and return nothing, some will do reasonably well, while a tiny fraction will become wildly successful and return most of the investor’s profits.

Startups normally take investment in a series of “rounds.” Unlike publicly-traded companies whose stock is bought and sold on exchanges like the New York Stock Exchange, ownership of startups is privately held and is traded only rarely. In a given round of financing, a startup’s owners and potential investors will negotiate to determine what fraction of the company ownership to sell in return for how much money, and this will be used to calculate the company’s valuation i.e. the implied worth of the entire company in dollars.

Currently, Y Combinator invests $125,000 in return for 7% ownership in the companies that go through its program, sets aside $375,000 to automatically invest in the next round of financing, and also “gets a right to continue to invest in subsequent rounds” at the same price as other investors.2 For most of its history YC has bought 7% of the companies that go through its program, although the amount of money it offers in return for this share has grown massively over time. 

Like any investor, Y Combinator’s ownership percentage will drop in later financing rounds as it is “diluted” by the creation of new stock, and also if it sells off its ownership in its portfolio companies before they reach their peaks, but its ownership may also rise if it invests more in later rounds, as it regularly does. The total value of YC’s investments is not publicly known, but if the 7% figure is taken as a very rough starting point, it would suggest that its stake in the companies it graduated—which are valued at over $600 billion in total, according to the organization itself—is worth somewhere around $40 billion. Such investments take a long time to mature, and the eventual value of YC’s portfolio may well grow further. Its future depends on whether it can continue picking or creating the founders who will go on to create unicorn companies and earn outsized returns for their early investors.

The goal is to eventually “exit,” which comes in two forms. The first is to hold an initial public offering (IPO) and be listed on a public stock exchange so that investors ranging from individual households to giant asset managers like BlackRock can buy the company’s stock. The other is to be acquired by a large company such as Meta or Google, which buys all the stock at once. In either case, the early investors—and other stockholders like the founders and early employees—can sell their ownership for a profit. 

Before an exit, these companies often operate at a loss for a decade or more, meaning they have to raise more rounds of money from investors to pay their growing bills. A successful company will have a larger valuation in each round of financing, as its customer base grows and it shows greater promise. A drop in valuation is referred to as a “down round,” and is generally considered very negative. In principle these valuations are based on expectations of future profits. In practice, as in many other forms of contemporary finance, the link can often be tenuous. 

Valuations are prone to large swings for opaque reasons, such as when payment processing company Stripe, once YC’s highest-valued portfolio company at $95 billion in 2021, cut its valuation down to $50 billion at a stroke in 2023. Stripe cofounders Patrick and John Collison attended YC in 2009. DoorDash, the largest food delivery company in the U.S., went through YC in 2013. Today DoorDash has a valuation of over $30 billion but still operated at a loss of over $1 billion in 2022.3 It remains to be seen whether DoorDash will become profitable, but this is only incidentally related to whether investors themselves profit. SoftBank invested $680 million into DoorDash starting in 2018, then sold much of its stake for $5 billion within a year after DoorDash’s 2020 IPO.4

Charismatic startup founders like WeWork’s Adam Neumann or FTX’s Sam Bankman-Fried have infamously conjured vast but ultimately fleeting valuations. No major YC company has suffered such a total collapse, although it is possible that OpenSea, a non-fungible token (NFT) marketplace valued at $13 billion in 2022, might. OpenSea’s trading volumes have dropped 99% as the recent NFT craze has faded, reportedly prompting one investor to internally mark down its investment by 90% so far.5 Such extreme cases are rare, but nevertheless demonstrate that the wider market for startup equity is in large part a “Keynesian beauty contest”—market speculation where choosing the same company to invest in as other investors can outweigh company fundamentals, since most investors lack special knowledge of the company’s circumstances.6 

Because the core staff of Y Combinator are mostly former software entrepreneurs themselves, and therefore have at least some special knowledge of the companies they are investing in, they are somewhat insulated from the dynamics of such a Keynesian beauty contest. However, Paul Graham has argued that most major buyers and sellers of equity in software companies are “making big decisions about things they don't understand,” and it is these decisions which set the prices in the market YC relies on.7 

Graham further writes elsewhere that “valuations are made up” on the basis of strategic considerations, custom, and outright collusion, and so are “far [...] from reflecting any kind of value of the company.”8 Even if YC is better able to evaluate substance, they can in practice make money simply by reselling their investment to less grounded investors at an eventual IPO, or to large foreign investors like SoftBank or Saudi Arabia’s Public Investment Fund.

The Power of the Alumni Network

In 2005, Paul Graham cofounded YC with Trevor Blackwell and Robert Morris, two of his core colleagues at his earlier startup Viaweb, as well as Jessica Livingston, his now-wife. It began with a summer fellowship in Cambridge, Massachusetts, pitched to college students as an alternative to summer jobs, with YC’s investment consisting of about $6000 per person for living expenses, unless “you've told us about specific hardware you need.”9 

When Y Combinator’s staff speak of “graduates,” “alumni,” and “office hours,” it is not just turns of phrase. Like the elite universities of Graham’s youth, YC is meant to provide an education appropriate for top businesspeople, socialization into an upwardly-mobile class, an induction into a high-trust professional network, and a strong credential of generalist competence. As universities have declined over the decades, YC now ironically achieves these goals better than the elite universities themselves.

This success was partially built on the reputation of the dozen or so young entrepreneurs of the now legendary “first batch,” including later OpenAI CEO Sam Altman; Steve Huffman and Alexis Ohanian, the founders of social media giant Reddit; Justin Kan and Emmett Shear, who would later found the streaming video platform Twitch; and the internet activist Aaron Swartz, widely perceived as a martyr for his cause. While it would take time before these companies earned high valuations, Graham and his colleagues were encouraged and began scaling up YC. They ran the program twice per year, alternating between Cambridge and Mountain View, California, until permanently settling in the San Francisco Bay Area in 2008.

Afterwards, applicants were drawn by Graham’s personal charisma and the worldview he articulated in his essays. At this stage, Y Combinator attracted people willing to eschew the traditional paths to success in order to take a chance on an unproven career track. By the early 2010s, the widespread usage and high valuation of early YC companies like Reddit and Dropbox also drew in people motivated to follow the trail blazed by those early alumni. As the startup founder became a more familiar cultural archetype and no longer appeared like a deviation that could end in ruin, ambitious strivers arrived in growing numbers. The twice-annual batches grew from about ten companies each in the early years, to 74 in the winter 2014 batch before Graham stepped down.10

With an intake process that emphasizes past accomplishments and interviews rather than bureaucratic credentials, and with an ethos of taking chances on those who show promise rather than weeding out those who might fail, YC has been able to select charismatic and energetic candidates for its program. The three-month program serves as the foundation of the alumni’s later bond. YC provides some mentorship and guidance, but the company founders are mostly self-taught on the job rather than in classes, and spend the vast bulk of their time in the program working on their companies. 

Entrepreneurs often describe building a company as a harrowing experience and doing this alongside a peer group serves as an emotionally powerful initiation for many founders, similar in some ways to the initiation and bonding rituals for groups ranging from college fraternities to military recruits. Graham has said that bringing cohorts of entrepreneurs together like this is YC’s “most important idea.”11 The process has been further standardized as the organization’s early investments became successes and it became a respected institution itself.

In 2010, YC began adding more staff to accommodate its larger class sizes, hiring alumni and veteran startup founders to teach.12 These staff are compensated partly in “deal flow,” as many of them—including all of Graham’s successors as President—also take the opportunity to invest their personal fortunes in promising companies. Over time, the growing accelerator was reorganized to rely more and more on these second-generation instructors rather than the original founders.13 

Meanwhile, investors began to perceive graduates as pre-vetted winners, and in 2011, investors Yuri Milner and Ron Conway offered an automatic $150,000 investment to every YC graduate.14 Since then, an ecosystem of other venture capital funds has grown which invest mainly or entirely in YC companies.15 As YC’s reputation and its financial ecosystem have grown, founders have been able to raise more money on better terms.

The alumni network kicks in to offer connections, private information, and other assistance to founders, a practice which was largely started by Sam Altman during YC’s infancy and has since become ingrained.16 Fellow founders may take a meeting, make an introduction, offer a discount, or buy an untested product out of solidarity. They are also given access to a private social network, Bookface, which they retain as alumni. YC maintains a private blacklist of investors who have unethically screwed over its companies and ensures its portfolio companies are forewarned. Many staff also use the program as an opportunity to scout promising companies and make investments, either in their personal capacity or on behalf of funds which they manage. This level of access and support by no means assures that a company will succeed, but it undoubtedly provides a very useful advantage.

Paul Graham’s Theory of Economic Growth

Paul Graham is the live player who designed the Y Combinator program and built its economic engine. He is also an influential intellectual who, more than anyone else, created and popularized the idea of what a “startup founder” is. Graham constructed a legitimizing ideology that cast startup founders as economic heroes responsible for society’s prosperity. As he thought through “all the things they should change about the [venture capital] business,” Graham was inspired to start YC and transform venture capital in line with his vision.17 

Paul Graham was born in 1964 in England and moved to the U.S. at the age of four, growing up in Pennsylvania. He graduated from Cornell University and then acquired a PhD in computer science from Harvard University. After briefly pursuing a career as a painter, in 1995 he cofounded Viaweb, a software company that let clients build online storefronts, which was then a cutting-edge business. The company proved successful and in 1998 it was bought by Yahoo, at the time the world’s most formidable internet software conglomerate, for $49 million worth of stock.18 Soon after, Graham left Yahoo and used his newfound leisure to work on personal projects like designing a new programming language and publishing essays on his blog. In these essays, Graham thought through an economic theory which would serve as the basis for YC’s strategy.

Software and computer entrepreneurs predate Y Combinator. Microsoft’s Bill Gates, Apple’s Steve Jobs, and Amazon’s Jeff Bezos had built highly successful companies in the decades before Graham’s influential essays, as did Graham himself. However, at the time, the public perception of Gates, Jobs, and other such tech entrepreneurs was not of visionary businessmen, but of nerds and engineers whose obsession with computers had allowed them to strike it rich.19

Software entrepreneurs who instead marketed themselves as businessmen, like the Bloomberg Terminal’s Michael Bloomberg, were not perceived as software entrepreneurs at all. For their part, the culture of programmers—or “hackers,” as Graham called them—generally viewed business as exploitative at best and dishonest at worst, and denigrated the practices of salesmen, MBAs, and other “suits” in favor of the purity of writing code.

Graham changed all this by painting a picture where software entrepreneurs were virtuous pioneers responsible for creating tremendous value and wealth for the world. Going into business was no longer an abandonment of the higher principles of the craft for mere worldly concerns, but was now nobly taking up a burden on behalf of society, and being justly rewarded for hard work and courage if the work should succeed. He articulated paths to professional and social success for startup founders and later engineered the software industry’s culture and institutions to be more favorable to founders. Graham laid out these ideas in essays on his blog, and later the Hacker News forum catalyzed an influential community which popularized his views throughout programmer culture. Notably, all this also guarantees deal flow for YC, since those compelled by Graham’s writing will possibly decide to apply to YC.20

The core of Graham’s economic analysis that underwrote his archetype is the productivity of individuals, as he wrote the year before he started YC.21 Graham argues that some people produce far more than their peers, or can in the right circumstances, but this does not happen in large companies because people who work harder and better see relatively little reward. In his words, “I think the single biggest problem afflicting large companies is the difficulty of assigning a value to each person's work. [...] You can't go to your boss and say, I'd like to start working ten times as hard, so will you please pay me ten times as much?” 

The only way to incentivize such productivity, and the great personal sacrifices that go with it, is by starting a small company where the benefits accrue directly to the owner-operators, the founders. This belief in the outsized value produced by the best people is also the reason for YC’s obsessive focus on funding the most promising individuals even if their company seems bad. In Graham’s words “If we want to get the most out of [the most productive individuals], we need to understand these especially productive people.”22 

In Graham’s view, the advent of software since the 1990s means that these people can now start major companies with no capital investment beyond their own living expenses, changing the balance of negotiating leverage from investors to founders and releasing a flood of untapped entrepreneurial energy.23 If these founders can set aside office politics and complex plans as secondary in order to simply “make something people want,” then the rest will fall into place.

Graham further argues that wealth acquired through business is the most moral and legitimate type of wealth.24 “There are plenty of other ways to get money, including chance, speculation, marriage, inheritance, theft, extortion, fraud, monopoly, graft, lobbying, counterfeiting, and prospecting. [...] For most of the world's history, if you did somehow accumulate a fortune, the ruler or his henchmen would find a way to steal it. But in medieval Europe something new happened. A new class of merchants and manufacturers began to collect in towns. Together they were able to withstand the local feudal lord. So for the first time in our history, the bullies stopped stealing the nerds' lunch money. This was naturally a great incentive, and possibly indeed the main cause of the second big change, industrialization.”

Through his writing and creating the institution which educated so many prominent startup founders, Paul Graham has influenced the ideal of what a startup founder should be more than any other individual. The second-most influential individual is fellow investor-intellectual Peter Thiel, who achieved comparable impact through his bestseller book Zero to One and the Thiel Fellowship, which gives young people $100,000 to drop out of college and work on their own more impactful projects. The popularity of Graham’s writings also contributed to books such as Eric Ries’ The Lean Startup that further developed the software entrepreneurs’ tradition of knowledge and as well as the heroic self-image of founders.

In 2007, Y Combinator launched Hacker News, a discussion forum Graham coded in his homebrew programming language Arc.25 Hacker News was largely inspired by Reddit, itself a YC graduate, and was intended to “recreate the way Reddit felt back in 2006.”26 This soon became a major hub for the growing startup culture, centered around YC alumni, and for hacker culture more generally. The forum is designed to preserve its niche culture, with active and aggressive moderation, limited privileges for new users, and algorithmic tweaks to discourage unwanted posts. Graham has continued publishing essays—albeit more slowly—developing and popularizing his thoughts on software startups, and making the case that ambitious young people should found companies.27

Sam Altman’s Leadership

As a company made up largely of startup founders, YC also began organizing and advocating for the interests of startup founders as a distinct social class. By example and by advocacy, it played a role in raising the investment amounts early funders receive, and in marginalizing the once-common practice of early investors receiving a seat on the board of their portfolio companies. For example, in late 2013 YC released the SAFE (Simple Agreement for Future Equity) and eventually established it as the industry standard for investment agreements.28

Graham retired a few months later in early 2014, succeeded in his role as President by his protege Sam Altman, a live player with a very different skillset and character than Graham. Whereas Graham is an intellectual first and foremost, who uses analysis and theory to understand the economic landscape and deduce the correct moves, Altman is mainly a dealmaker, networker, and marketer. Today Altman is best known as the CEO of OpenAI.

At the age of 19, Altman attended YC’s first-ever cohort and left school to continue building his company. His startup, Loopt, began as a phone app that let users share their location with friends, and over time added more and more “social” features in an ultimately unsuccessful attempt to achieve massive user growth. Altman nevertheless made a strong impression on Graham, who in 2006 wrote “Loopt is probably the most promising of all the startups we've funded so far. But Sam Altman is a very unusual guy. Within about three minutes of meeting him, I remember thinking ‘Ah, so this is what Bill Gates must have been like when he was 19.’”29

Graham took Altman under his wing and Altman began aggressively networking on behalf of Y Combinator and its companies. Graham later recalled that Altman “set the standard for how much the alumni help one another” and “did most of the initial intros in Silicon Valley for YC itself.”30 In 2011, Altman officially joined as a part-time partner instructing new companies.31 In 2012, he sold Loopt for $43 million to Green Dot, a prepaid debit card company. This was an apparent “acquihire,” a term for when an established company buys a startup in order to acquire its employees. Shortly after, Altman started Hydrazine Capital, a venture fund which invested in early-stage companies, including many YC companies.

As President, Altman oversaw the continued steady growth of the core startup accelerator program, which soon had well over one hundred companies going through each batch. Altman increased the size of initial investments about sevenfold, to $120,000, in return for the same 7% of the company.32 He also launched a number of initiatives that later petered out, including a Chinese branch of the organization and a fellowship for companies not yet ready for the core program. But most significantly, in hindsight, was what would later become OpenAI. In October 2015, Altman launched YC Research, a nonprofit lab, with $10 million of his own money.33 YC Research’s first program was on artificial intelligence, and within months it also began its second program, a basic income experiment, which has since spun off into a separate organization and changed its name to Open Research.34 

In December 2015, YC Research’s AI project joined forces with Elon Musk and leading AI researchers, who were interested in—and ostensibly concerned about threats from—artificial general intelligence (AGI). Together they formed OpenAI, a nonprofit research organization. Musk drifted away and formally left in 2018, shortly before OpenAI released GPT. In March 2019, Altman stepped down as President of Y Combinator to focus on OpenAI, and simultaneously reorganized OpenAI from a nonprofit to a for-profit company, which has since grown tremendously and developed GPT further.

Altman believes technological progress driven primarily by Artificial Intelligence will create material abundance that will upend modern civilization.35 He wants to build and steer this technological revolution, and redistribute the material gains of technology to society at large through policies like universal basic income (UBI). Some of Altman’s recent efforts are aimed at both bringing this transition about and smoothing over the expected socio-economic crisis. An example is a company he co-founded: Worldcoin, launched in 2019, which seeks to create a global cryptocurrency based on biometric identification—this would allow global verification of who is and isn’t human online in an era where the internet might become dominated by artificial intelligence. Altman further states he hopes this infrastructure might help lead to UBI to address expected disempowerment of labor compared to capital driven by technological advances.36 

Partly to this end, partly following a renewed mid-2010s Silicon Valley interest in “hard tech,” he pushed YC to accept more companies building physical technology rather than just software.37 These included the fusion energy research company Helion Energy and supersonic airplane designer Boom Supersonic. Altman co-founds, invests in, and promotes companies that explicitly aim to solve global problems of historic consequence that are in vogue in tech circles. This strategy may have influenced other YC alumni, such as John and Patrick Collison, who have used their online payments company Stripe to fund the development and rollout of carbon-capture technology to address climate change.38

After Altman’s departure in 2019, he was replaced as President of Y Combinator by Geoff Ralston, a longtime software executive who had been a YC partner since 2012. Ralston led the organization through the turbulent years of the COVID-19 pandemic, making few changes to the organization itself. Ralston made the second-round investments automatic rather than discretionary.39 In 2023, Ralston stepped down and was replaced by Garry Tan.

Tan had previously founded Posterous, a YC-backed blogging platform which was acquihired by Twitter in 2012, and he was the first outside angel investor in Coinbase.40 He joined YC as a partner in 2011. In 2013 he left the organization, though not the network, to pursue Initialized Capital, a venture capital fund he cofounded with other YC alumni to invest in YC companies.41 Initialized Capital raised and invested billions of dollars before Tan returned to Y Combinator as President. While it is still early in Tan’s tenure, he is notably pursuing local political reform in San Francisco; he is involved in multiple “moderate” organizations like GrowSF, frequently comments on the city online, and has donated at least $278,000 to political campaigns since 2021.42

The Next Google is Not a Web App

The household names among YC’s portfolio companies all come from Paul Graham’s tenure, and this is not purely a matter of age, since some of the early companies like Reddit and Airbnb became widely-used and well-known in well under a decade. In the nine years since Graham’s departure, YC’s portfolio has continued to perform extremely well if not better as measured by valuation.43 

However, the value they provide to consumers does not seem to measure up. Early successes like Reddit, Dropbox, and Airbnb, which went through YC in 2005, 2007, and 2009 respectively, made genuinely new things possible for their users. Even the less flashy successes like payroll company Gusto, which went through YC in 2012 and is valued at $9.5 billion as of 2021, participated in the phenomenon of “software eating the world,” making it notably easier to automate administrative tasks which had formerly involved more human labor.44

In contrast, the successes from after Graham’s tenure are lesser known and less transformative for their users. For example Faire, an online wholesale marketplace using a model similar to Amazon, Alibaba, or Etsy, was valued at $12.5 billion in 2022, and has undergone two rounds of layoffs since then.45 Brex, which provides credit cards and other financial services to businesses, was valued at $12.3 billion in 2022.46 Both Faire and Brex were part of YC’s winter 2017 batch.

Early-stage venture capital investments take a long time to come to fruition. These or other YC companies from after Graham’s tenure may yet prove to be economically and culturally important like their forerunners. The evidence so far, however, suggests that their impact is well below that of earlier YC companies of similar age, in spite of their higher valuations. It might seem, then, that the golden age of the internet startup has passed. Decades ago, a handful of hackers in a dorm room could upend major parts of economic and civic infrastructure. Over time, however, the internet has centralized.47 Major internet companies have entrenched themselves politically, and upstarts have a far harder challenge.

Any new industry goes through an initial period where it is a wide-open field suitable for building new independent empires and power centers, but by the time this opportunity becomes widely known to ambitious strivers, it has usually been fully exploited and tapped out by the trailblazers who got there first.48 The raw economic opportunities for transformation may not be exhausted, but the political economy makes them far harder to exploit.

Many Silicon Valley technologists hope the recent boom in artificial intelligence startups will prove to be a continuation of the previous golden age of startups. Notably, this AI boom follows a quickly-forgotten boom in cryptocurrency and “Web3” startups that was also marketed as such. According to Paul Graham, “in the last batch [of YC] half the companies weren’t AI startups.”49 This implies YC is now betting a full half of its capacity on AI. Unlike previous waves of startups, even including cryptocurrency, the AI boom faces a clear and imminent threat of heavy regulation from governments that would prevent large new empires from forming. Silicon Valley may prevail nevertheless, but these are new and formidable challenges.

Some of YC’s more recent companies have been economically important, but are not internet software startups so much as hardware startups. For example, Cruise, a self-driving car company that in 2014 went through the final YC batch before Graham stepped down, operates its self-driving software in the physical world and will continue to do so assuming they can overcome the problems behind a 2023 safety recall.50 Flexport, from the same batch, has been “eating the world” of international freight logistics, requiring both digital and material expertise. These companies have not just achieved high valuations, but aim at transforming economic life. Altman’s attempt to diversify away from the internet and software may prove prescient.

While companies like Faire or Brex are unlikely to carve out empires comparable to earlier companies like Stripe or Twitch, let alone conglomerates like Google or Meta, YC’s pipeline is in little immediate danger. Venture capital is not necessarily quick to react to such changes, and it took over half a decade before venture capitalists fully realized the potential of YC’s approach in the first place. YC’s network and access means their companies will have preferential access to funding streams for some time to come. Even if successes at the level of Stripe or Airbnb peter out, then YC would lose out on the unprecedented profits it has enjoyed so far, but more modest software unicorns and acquihires would likely keep it profitable by any reasonable measure.

YC’s business model will very likely remain lucrative for as long as it remains the favored destination of the most ambitious software entrepreneurs, as long as the financial system continues investing in the software industry, and as long as the world continues to use software. None of these facts are likely to change any time soon. But as the contemporary political economy of the United States has changed, such software successes will be less likely to greatly grow the economy and less likely to result in new, independent personal empires run by live players that can make further transformative bets in business and technology.

ASML is a publicly-traded Dutch technology company with a market capitalization of $236 billion as of October 2023, making it Europe’s largest tech company.1 Headquartered in a suburb of the small city of Eindhoven, the semiconductor company is one of the largest global suppliers of photolithography machines, which are used by semiconductor manufacturers to create the required patterns on silicon wafers used in computer chips. ASML also holds a monopoly on the development and manufacturing of extreme ultraviolet lithography (EUV) machines, which are necessary to produce the most advanced chips available today, such as the H100 Tensor Core graphics processing unit (GPU) designed by Nvidia and manufactured in Taiwan by TSMC.

You can listen to this Brief in full with the audio player below:

Entrepreneurs, investors, technologists, and even government officials in the developed world have come to expect that increasing the total amount of available computing power will result in economically transformative artificial intelligence that will offer technological solutions for a wide range of economic and political challenges arising over the last twenty years from the repeatedly disappointing performance of dominant institutions. This “scaling hypothesis,” as opposed to building artificial intelligence through novel architecture or data sets for training, necessarily relies on vastly increasing manufacturing of the most advanced chips, which so far can only be done by expanding the operations of a handful of companies like TSMC, Intel, or Samsung. 

These companies, in turn, all rely on ASML alone to supply the EUV machines necessary for expansion. Through its EUV monopoly, ASML is an irreplaceable supplier for all of the largest chip manufacturers. The company is arguably the single hardest layer to recreate in the chip manufacturing stack, not only because of the vast capital costs that would be necessary to build the complex machines, but also because its success rests on a unique Dutch tradition of knowledge in photolithography that goes back to the 1960s, if not earlier.

As a result, no competitor to ASML is likely to emerge. ASML’s close relationships with the other major semiconductor manufacturers and its key component suppliers, several of whom are themselves monopolies on various other parts of the supply chain, are further obstacles. Likely only extensive subsidies and targeted government policy could end ASML’s monopoly on EUV machines, but the only industrialized country with both the incentive and ability to follow through on this is, for now, China.

ASML is a Monopoly Through Technical Superiority

ASML has about a 90% market share in photolithography machines and a 100% market share in the most advanced form of photolithography, extreme ultraviolet lithography.2 This effective monopoly is straightforwardly the result of technical superiority. Photolithography is the process of forming patterns on a silicon wafer using light, in this case at an extremely small scale. To build machines capable of reliably doing this is an extremely difficult feat of engineering. Light is passed through a photomask that contains the desired circuit pattern. The wafer that the light is aimed at is coated with chemically sensitive photoresist, which reacts to the light to form the required patterns. Modern photolithographic techniques use light beyond the visible spectrum, since shorter-wavelength light allows for higher transistor density on the wafer, which means more computing power per chip.

The process is conceptually similar to using a sharper pencil to write smaller letters on paper. “Deep ultraviolet” (DUV) techniques use wavelengths as short as 193 nanometers, while extreme ultraviolet (EUV) techniques use even shorter wavelengths of 13.5 nanometers. To manufacture and control these extreme ultraviolet wavelengths requires the heating of tin to a temperature of 220,000 degrees Celsius, forty times hotter than the surface of the Sun, in order to transform the tin into plasma. In the machines, the tin is heated by a laser that travels through vacuum, striking each speck of tin twice at frequencies of 50,000 impacts per second. The resulting light is then directed towards the photomask by mirrors composed of over one hundred layers, each layer several nanometers thick. The correct positioning of the wafer is critical to achieving maximum yield. These machines keep the wafers positioned with a quarter of a nanometer’s precision.

There are no how-to guides or written instructions for reliably reproducing such a difficult technical feat. Achieving it consistently relies on intellectual dark matter: the tacit knowledge, idiosyncratic skill, and unwritten rules that are necessary for success but cannot be made legible to outsiders or non-experts except through training with the current expert practitioners, in whose heads the knowledge actually lives.3 ASML staff are deployed along with its EUV machines to oversee their operation and maintenance: they remain on site throughout the lifetime of a given machine. The chip manufacturers are therefore dependent on the company for tacit technical knowledge as well the machines themselves.4 Much of this tacit knowledge relates to how light refracts in different contexts. The maintenance of this knowledge in an organized manner by the company means it carries a unique tradition of knowledge.5

Advancing this tradition is very expensive. ASML employs over 14,000 people in its R&D departments, out of over 39,000 employees worldwide, and spends almost $1 billion on R&D every quarter.6 To build a working EUV machine took thirteen years from the first prototype shipped in 2006 to the first EUV-enabled products released in 2019, and cost roughly $10 billion in R&D and capital expenditure costs, according to some external estimates.7 The company itself estimates that it spent $6 billion in R&D alone over 17 years.8 Including the costs borne by component suppliers, the total EUV development costs likely ran into the tens of billions.

Given the costs ASML paid to develop EUV, its prices are unsurprisingly high. The machines cost around $200 million each. In 2022, the company shipped just 54 of them, though it also profits from maintenance contracts on installed machines.9 Next-generation “High-NA” EUV machines are currently in production; the “high numerical aperture” refers to a greater ability to collect and focus light. Over five prototypes have been ordered by Intel and other customers. These are selling for $340 million per unit and forthcoming production models will be priced still higher.10

ASML generated almost $22 billion of revenue in 2022 at a 35% profit margin before interest and taxes (EBIT) and is growing revenue at around 20% annually.11 The company forecasts that by 2030 it will be generating between $45-60 billion of revenue annually.12 The company usually trades on the stock market at a high multiple of its current earnings, indicating that investors have high expectations of the company’s future growth and profitability. While its margins are substantially lower than those of TSMC, at around 65%, and Intel, they are comparable to many of the largest companies using the chips it helps to make.

In 2022, ASML operated at a higher gross profit margin than Apple, despite the latter’s pricing power due to its luxury branding. Smartphone manufacturers and cutting-edge AI companies alike are ultimately dependent on the semiconductor industry and its key suppliers. This means that the semiconductor firms and their key suppliers are able to capture a large portion of the windfall from the growing demand for high-performance computing, new models of smartphones and laptops, and more exotic use cases like augmented reality or virtual reality devices. This windfall goes in large part to the semiconductor design and fabrication companies, who are both dependent on ASML.

In turn, ASML inevitably shares some of its part of the windfall with its own suppliers. By one count, only 15% of the components of an EUV machine are manufactured by the company.13 The other 85% are manufactured by external suppliers. ASML assembles its machines itself at a single factory in Veldhoven, a suburb of the Dutch city of Eindhoven.14 ASML maintains close relationships with its most important suppliers, especially the privately-owned German optics manufacturing firm Carl Zeiss. Zeiss supplies the layered mirrors that reflect extreme ultraviolet light through the photomask onto silicon wafers. It is the world’s leading manufacturer of lithography optics equipment and has been a key supplier to ASML since the latter’s founding. ASML has described the relationship as “two companies, one business,” although throughout the 1990s the relationship was less close.15

In 1996, Zeiss attempted to sell its new optics technology to ASML’s then-competitor Silicon Valley Group (SVG), which caused a crisis in relations between the firms. SVG had grown out of the U.S. optics firm Perkin-Elmer and become one of Intel’s main suppliers. ASML management made it clear to Zeiss that they regarded this as unacceptable, in part because the optics had been developed due to ASML’s investments. Then-Zeiss CEO Peter Grassmann was invited to join ASML’s board of directors and the two companies signed an exclusivity deal the year after. In 2016, ASML paid €1 billion for a 25% stake in Zeiss, with its money going towards Zeiss’ R&D for optics that will constitute the core of the next generation of EUV equipment.16

Other key suppliers have been bought outright, partly in order to ensure quality control and align the supplier more closely with the company’s key priorities. Current CEO Peter Wennink reportedly warned one supplier that “if you don’t behave, we’re going to buy you.”17 Light source manufacturer Cymer was acquired in 2012 for $2.5 billion, explicitly in order to speed up the development of EUV and keep control of the EUV light source technology.18 ASML has also purchased Hermes Microvision, a manufacturer of inspection equipment, acquired in 2016 for $3.1 billion. Going back further, the company even bought out key competitors, including SVG, which it acquired in 2001. ASML’s business model and its pattern of acquisitions show a company that is entirely focused on advancing the cutting edge of technical superiority in a very narrow domain, rather than trying to increase margins for profitability or expand into new domains to build a conglomerate. This is unsurprising given that the company is built around a tradition of knowledge passed down by Dutch scientists.

The Dutch Tradition of Knowledge That Birthed ASML

ASML’s research and product development efforts have been led largely by 66-year-old Martin van den Brink, who today serves as co-president and chief technology officer of the company, roles he has held since 2013. He is paid equally to the company’s CEO and other co-president, Peter Wennink, who first joined the company in 1999 as chief financial officer, having previously worked at Deloitte as an accounting partner.19 Wennink and van den Brink are two of the six members of the ASML management board, but van den Brink is the only one whose history at the company dates back to its founding in 1984. The other management board members, as well as the supervisory board members, are relatively short-term appointees typically drawn either from the company’s executives or European industry broadly.

As a young man, van den Brink earned degrees in electrical engineering and physics from the HAN University of Applied Sciences in Arnhem and the University of Twente respectively. These are both Dutch universities in small cities. He joined ASML as an engineer at the age of 27 years old and has worked there ever since. Van den Brink led the technical development of ASML’s first successful lithography machine, the PAS 2500, which secured the company its first sales to TSMC and AMD and allowed it to turn its first profit in 1989.20

Van den Brink was also closely involved in the engineering of the PAS 5500, which secured ASML its first sales to IBM. Refurbished PAS 5500s remain in operation today and ASML continues to offer support for the model.21 By 1994, at 37 years of age, van den Brink was promoted to lead the company’s R&D department, although he was first sent on a six-week management course to learn how to handle the egos of the technical managers he had bruised in previous years.22 Subsequently, van den Brink also led the development of ASML’s first “step-and-scan” model, introduced as the deep ultraviolet lens PAS 5500/500, the machine that powered the company to market leadership when released in 1997. This model was a technical design advance on previous machines called “steppers.” Steppers, among other things, combined optical alignment and projection systems i.e. the wafer-positioning laser was sent through the same lens that projected the light wave onto the wafer.

Van den Brink was also pivotal to the decision to take a seemingly less ambitious path to product development after 1986. ASML’s main competitors at the time, the Japanese companies Nikon and Canon, sought to jump directly from 436 nm “g-line” steppers to 248 nm deep ultraviolet steppers, then considered the cutting edge. Van den Brink was aware of the technological challenges that the shift to DUV would bring and ensured that ASML instead first put its main efforts behind a reliable and fast intermediate 365 nm “i-line” stepper.23

This product came to market in 1991 and its sales allowed the company to take market share from its Japanese rivals throughout the 1990s. ASML did not introduce a deep ultraviolet product until 1997. Under van den Brink, the company was also the first to commercialize immersion lithography from 2003 onwards, a generational shift in production techniques that replaced the traditional air gap between lens and wafer with water. These are both examples of van den Brink’s deep technical expertise in photolithography guiding business decision-making and demonstrate the centrality of this expertise to the company’s commercial success.

Van den Brink subsequently oversaw the lengthy development of commercially-viable extreme ultraviolet lithography machines. In 2006 he gave a speech at the International Symposium on Extreme Ultraviolet Lithography that outlined his belief that EUV was the only cost-effective option for increasing the number of chips per silicon wafer.24 Although commercial development overran by a decade longer than planned per van den Brink, the trust ASML built with its customers had attracted large-scale investments to finance the final stages of EUV’s development, including from TSMC, Samsung, and Intel, who collectively bought a 21% stake in ASML in 2012, with Intel buying the largest piece, up to 15% for $4 billion.25

Van den Brink has had long-term allies within ASML, most notably Frits van Hout, a fellow physicist who studied at Oxford and in Switzerland then also joined the company in 1984. He served as project manager for the PAS 2500’s development alongside van den Brink, who led the systems engineering. Van Hout left the company in 1992 but rejoined in 2001, two years after van den Brink’s promotion to the company board, and was himself added to the company’s board in 2009. From 2013 to 2021, he oversaw the last stages of the development of EUV as Chief Program Officer and managed relationships with ASML’s key customers and suppliers as Chief Strategy Officer. He retired in 2021.26

Van den Brink himself is expected to retire in 2024, having persuaded the board and Carl Zeiss to make the required investments in High-NA EUV against, according to him, significant opposition.27 It is unclear, however, if he will actually do so, especially given that he has already explicitly set the company’s main strategic priorities for the next decade and may be tempted to stay and drive through his plans to maximize the efficiency of the company’s current products and push EUV to the limit of its potential.28

Van den Brink did not found the tradition of knowledge that has underpinned the company’s success, but inherited it and extended it. His first manager at ASML was the Dutch engineer Herman van Heek, co-inventor of the prototype of the world's first wafer stepper. Van Heek co-developed it with his colleague Gijs Bouwhuis at the physics and materials research laboratory of the Dutch electronics conglomerate Philips, called NatLab, which was comparable in size and scope to, for example, AT&T’s famous Bell Labs. Throughout the 1970s it employed over 2000 people. NatLab’s greatest successes were technologies key to the invention of the cassette tape and the CD, both consumer electronics innovations that Philips successfully commercialized.

In 1971, van Heek and Bouwhuis developed the world’s first design for a “stepper,” which was called internally the Silicon Repeater. This technology was patented and the intellectual property later transferred to ASML. While a working prototype was built by 1973, the NatLab scientists failed to persuade Philips’ Science and Industry Division to commercialize it, although development of the stepper continued throughout the 1970s inside NatLab. The Repeater’s eventual commercialization at Philips was secured by the intervention of Wim Troost, then director of a unit of the Science and Industry division, but later ASML’s second CEO. Progress towards making a working model ready for commercial use was slow. Philips lacked the scientific expertise to develop a commercially viable model and the NatLab scientists were eventually called back in to develop a working control system. This shows how the tradition of knowledge was essential to commercial viability before ASML even existed.

By 1982, Philips had sold its first stepper, known as the PAS 2000, to IBM, its first customer beyond Philips’ own subsidiaries. But, at the time, the conglomerate’s senior management were looking to trim its size. An unprofitable cutting-edge semiconductor equipment business was an obvious target for divestment. Wim Troost, however, believed in the Repeater’s potential and so reached out to another Dutch semiconductor firm, ASM, to form a joint venture, after three other potential joint ventures fell through. ASM contributed $2.1 million in cash to the joint venture and Philips contributed seventeen PAS 2000 machines and 47 staff. The nascent company began operations in 1984 as “ASM Lithography,” now officially just ASML.

Van den Brink was one of the early external hires of the new company, but was discouraged by the initially lethargic corporate atmosphere and announced his intention to quit. In response, van Heek offered him the chance to lead the design of the lens projection and alignment systems of the PAS 2500. Van den Brink also insisted on spending a day a week at NatLab itself, learning from the men who had designed the stepper and continued improving it throughout the 1970s, including Gijs Bouwhuis and Joseph Braat, the Repeater’s optical designer.

ASML is thus a success of Dutch scientists and engineers working on photolithography at Philips and later ASML passing down this unique knowledge from one generation to the next. It is unclear whether van den Brink has a successor in mind who can replace him as the technical head of the company. The other management board members are not technical. From 2018 to 2021, the head of research at ASML was the 56-year-old Dutchman Hein Otto Folkerts, who is now the head of development and engineering for sensors and mechatronics.29 The vice president of R&D since 2022 has been Rafael Howell, a 47-year-old American.30 Both men have spent all or almost all of their careers at Philips or ASML. NatLab was disbanded in 2001 and its facilities became part of the so-called High Tech Campus Eindhoven, a corporate industrial park.

Through NatLab, ASML’s tradition of knowledge originated under the patronage of the Philips industrial empire. NatLab was founded in 1914 by the founders of Philips themselves, the brothers Gerard and Anton Philips, who had founded the company with their father in the 1890s to manufacture lightbulbs. The Philips brothers came from an illustrious family: their father was a first cousin of Karl Marx and their grandfather Lion Philips, a wealthy tobacco merchant, was Marx’s main financial sponsor. 

The company grew rapidly and was given royal Dutch patronage as early as 1916; the company officially renamed itself to Royal Philips in 1998.31 The Philips family were major patrons of the city of Eindhoven and family members served as company CEO until 1977. In European social democracies like Sweden and the Netherlands, governments formed broad alliances with industrial family dynasties to support their business empires in exchange for financing generous welfare states. ASML is ultimately a product of this system and the low social mobility of social democracies likely still helps the company keep much of its talented workforce in the Netherlands rather than Silicon Valley.

Favored by Dutch Industrial Policy and U.S. National Security

Throughout its history, ASML has benefited from European industrial policy. In its early years, the company survived on grants from the Dutch Ministry of Economic Affairs, which wanted to preserve and develop the domestic semiconductor industry. It also won subsidies from the European Economic Community (EEC), the predecessor of the European Union.32 Nederlandsche Middenstandsbank and ABN Amro, two of the largest Dutch banks, also served as patrons, repeatedly providing ASML with cheap loans. Prior to its initial public offering in 1995, the company still owed large debts to Philips and the banks, which it used to argue that the Dutch tax authorities should estimate the company’s value at zero and consequently levy no tax on the 5% shareholding awarded to a group of forty company insiders. The tax authorities agreed on a $5 million valuation, considerably below the almost $600 million valuation the company later listed at.33

In 1988, when ASM wanted to sell its stake in ASML, Philips assumed its stake, including the debts ASML owed to ASM. At the time, ASML was losing money, but Philips kept the company alive and continued to lend it money. This was partly due to the intervention of German board member Gerd Lorenz, who viewed chip production and lithography as strategic assets for Europe. Against a background of growing concern over the economic rise of Asian countries and Japan in particular, both European and U.S. elites wanted to retain control of cutting-edge technologies, much as anxieties over the rise of China have driven sanctions and restrictions on Chinese companies today.34

ASML further benefited from U.S. technology policy. Much of the early scientific work on EUV was, in addition to Intel, funded by the U.S. Department of Energy. While the relevant scientific discoveries took place in Japan in the early 1980s, the wider Japanese research community was at the time extremely skeptical of its commercial viability. American researchers were similarly unconvinced, but Bill Brinkman of Bell Labs was able to persuade the Department of Energy to fund further research at Livermore Labs.35 Brinkman had a long career in the U.S. scientific bureaucracy; he subsequently served in the Obama administration as head of the Department of Energy’s Office of Science.36 

From the U.S. government’s perspective, the technical leap forward to EUV offered an opportunity to break the dependence of the U.S. semiconductor industry on Nikon and Canon, and help to secure the domestic industry’s long-term future by reasserting its technical superiority over the main Japanese firms.37 By 1997 it was sufficiently clear to leading U.S. semiconductor manufacturers that EUV was worth their investment, despite the potential of alternative techniques like electron beam or X-ray lithography. Intel, AMD, and Motorola created a joint venture, “EUV LLC” to fund development. They in turn brought Silicon Valley Group on board as a partner to develop the products stemming from their collective research efforts. ASML, which had already been pursuing its own EUV efforts in partnership with Zeiss and Oxford Instruments, began working with EUV LLC in 1999 and became a fully-fledged partner upon its 2001 acquisition of SVG.38 

While the Japanese companies Nikon and Canon were not part of the coalition, some information from the program was shared with them, and the Japanese government funded their own EUV development efforts. Nikon built a working prototype in 2008 but delayed and eventually abandoned its EUV program after the 2008 global financial crisis and Intel’s 2012 decision to invest in ASML, signaling its commitment to buying their machines.39 Though U.S. policymakers could have blocked ASML’s takeover of SVG on antitrust grounds, they chose to let it go through, requiring only that ASML spin out SVG subsidiary Tinsley, a key supplier of lens-polishing technology for U.S. satellites. Intel’s own advocacy for the merger appears to have been a key factor in assuaging the Bush administration’s concerns. The U.S. Semiconductor Industry Association also backed ASML’s takeover of SVG. 

Intel preferred SVG survive under foreign ownership—albeit the ownership of another Western firm—rather than fold due to its inability to compete at scale with Nikon and Canon. Moreover, Intel already viewed EUV as vital for its long-term future, and while SVG held key optics technology for EUV’s development, ASML was thought of as a better platform manufacturer.40 The rising capital intensity of the lithographic industry made corporate consolidation inevitable and ASML became the de facto Western champion by the late 1990s. The fact a Dutch rather than U.S. company became the champion suggests U.S. policy was not driven by a live player interested in building up U.S. manufacturing, but reactive to narrow concerns like preventing Japanese monopolies.

For the Netherlands, ASML is today a national economic champion. Both manufacturing and R&D largely take place at the company’s facilities in and around Eindhoven, which has become a growing technology hub largely thanks to the company.41 ASML has invested heavily in the local Eindhoven University of Technology, recently announcing a new on-campus research facility and the funding of 40 PhD positions a year in plasma physics, AI, and lithography.42 Across Europe, ASML sustains a network of suppliers who depend on its success, such as Zeiss and German laser manufacturer Trumpf. For the U.S., ASML has become an effective means of restricting China’s access to the most cutting-edge chips, fulfilling a national security objective through “friendshoring.” From an economic perspective, access to EUV has helped Intel’s fabs stay competitive with TSMC and Samsung. This together means the U.S. government is unlikely to expend substantial effort to try and unseat ASML.

The Future of ASML

Demand for computer chips has been rising for decades and this is unlikely to change even if the current sharp upswing in demand due to tech companies pursuing artificial intelligence development fades away. This is because the global proliferation of information technology and the internet over the last thirty years, in everything from the workplace to the military, has rested on the development of ever more powerful computer chips that can do more useful work on smaller devices at lower costs. In the short-to-medium term, demand for GPUs driven by artificial intelligence is a major tailwind for ASML. All the largest tech companies in both the U.S. and China are seeking to rapidly increase their AI capabilities primarily through adding more computing power, rather than through improvements in model architecture or novel sources of training data.

OpenAI’s GPT-4, for example, was trained using 66 times the compute that the company deployed to train GPT-3. Google has reportedly trained their Gemini model, due for release later in 2023, on five times the compute used to train GPT-4. Meta, Amazon, Google, Microsoft, and Tesla are all buying up as many Nvidia GPUs as TSMC can make, while also developing their own chips designed in-house specifically for AI model training and inference, such as Microsoft’s Athena or Amazon’s Trainium. The chips are intended for their own in-house model development, but surplus capacity will also be rented to other players in the ecosystem, including well-funded startups such as OpenAI, Anthropic, Inflection AI, and xAI.

ASML’s financial future seems assured so long as it doesn’t face a competitor on the cutting edge of photolithography, which itself seems highly unlikely due to the capital costs that would be necessary and the difficulty of reproducing such a unique tradition of knowledge. TSMC, for example, was incubated by the Taiwanese government in the 1980s and led by Morris Chang, by then a veteran of the semiconductor industry with a long career in the U.S. at Texas Instruments, then the incumbent. On the face of it, any competitor to ASML would most likely need to both find a willing government backer and poach high-level members of ASML’s technical team. The only country that plausibly has the ability and incentive to attempt this, or anything similar, is China.

Since 2019, ASML’s expansion in China has been restricted by U.S. export controls aimed at limiting Chinese technological advances in AI. The company is already banned from selling EUV machines to China, although for now there are no sufficiently advanced Chinese fabs that could actually use them. Until recently, ASML generated around 14% of its revenue from China and at least some of its own suppliers manufacture their equipment there.43 In June 2023, further export controls were announced as part of a mutual agreement between the U.S., the Netherlands, and Japan. These restricted the export of ASML’s most advanced DUV machines.44 

ASML’s senior executives have repeatedly argued against the global trend towards regionalization and reshoring of semiconductor supply chains, arguing that in the long run this will simply force China to develop its own end-to-end domestic semiconductor supply chain and that, in the words of CEO Peter Wennink, “the laws of physics in China are the same as here.”45 If the Chinese government pursues such a policy, ASML is likely to be the target of intensive Chinese industrial espionage, especially given its close links to its local university and its decision to open up a R&D facility in Taiwan.46 In all likelihood this is already happening, with, for example, at least one report of an ex-ASML employee stealing IP and now being employed by Huawei.47 Industrial espionage will not, however, be able to recreate the whole tradition of knowledge underpinning ASML’s technical superiority. If China succeeds in matching ASML in the future, it will be due to the exceptional achievement of recreating its own tradition of knowledge in “the laws of physics.”

Despite its arguments for keeping the semiconductor industry globalized, ASML and its domestic position as a Dutch national champion mean the company will ultimately comply with U.S. efforts to choke off Chinese access to cutting-edge chip manufacturing technology. The Netherlands are a close U.S. ally and the company’s compliance at the behest of Dutch authorities illustrates Europe’s fundamental strategic dependence on the U.S. and inability to pursue an independent foreign policy.

ASML’s technological future is less assured than its financials or politics. Martin van den Brink has “mapped out” the course of the next ten years of ASML’s strategy, focused on deploying the High-NA EUV machines and reducing the costs of both EUV and High-NA patterning.48 He believes there is still room for improvement in making EUV machines more sophisticated and economical. Van den Brink has reiterated his suspicion that there will not be another generational breakthrough in EUV after High-NA, the so-called “Hyper-NA,” because he believes the costs of manufacturing still smaller chip structures might be insurmountably high.49

The company has a research group working on “Hyper-NA,” however, just in case. So long as van den Brink is leading ASML, there is good reason to think the company will continue making incremental improvements in photolithography and maintain its monopoly, perhaps even making one more big breakthrough. But the key juncture for the company will be the day when van den Brink inevitably retires or dies, when technical leadership will fall to an as-yet unknown successor. While the company would likely stay immensely profitable for a long time, similarly to Apple, that might ultimately leave an opening for technological disruption from a new competitor. Such a competitor would have to bet on a new technological paradigm of lithography and is today more likely to emerge in China or even South Korea rather than the U.S., let alone anywhere in Europe.