The Return of the Company College

The university will keep the gate for the many. The companies that own the machines have started building a school for the people who will run them.

Thomas Malthus spent almost thirty years, until his death in 1834, teaching political economy to teenagers whom a trading company had picked to govern India. The company was the East India Company, chartered in 1600 to buy and sell in Asia and, by the turn of the nineteenth century, the working government of millions of people on the Indian subcontinent. Its administrators had been chosen through the directors’ personal connections and trained by almost nobody, so in 1806 the directors opened a college in Hertfordshire and filled it with young men of sixteen to eighteen whom they had nominated for service themselves. The curriculum followed the jobs waiting in India: political economy, law, history, and the languages of the people the students would govern, among them Persian, Sanskrit, Bengali, Hindustani, and Arabic. The first chair of political economy in Britain belonged to a company’s school. Over five decades the East India College, known to everyone as Haileybury, trained more than two thousand of the Company’s civil servants. [1]

Two hundred and twenty years later, a venture capital firm announced a school of its own in San Francisco. The Horowitz Andreessen Academy, unveiled on September 22, will admit about fifty students, most of them straight out of high school or a year or two into an elite university, and some of them younger than eighteen, to a tuition-free founding fellowship that starts in fall 2027 and runs for a year. Students will build real projects in place of problem sets, tests, and homework, under instructors drawn from industry. A two-year program at elite-university prices is to follow in 2028 if regulators approve it. The school describes itself as co-created with ten founding partners: Anduril, Anthropic, Coinbase, Google, Meta, NVIDIA, OpenAI, Palantir, Replit, and Stripe. [2]

Most commentary has treated the launch as an education story, and the best of it as a parenting story. Two a16z writers, both fathers, asked whether they would send their own children and answered with warmth about raising kids at a moment when no career path can be called safe. [3] They are right about the uncertainty. Underneath the parenting question sits one about political economy. When private power begins doing work that looks like governing, it tends to build a school for the people who will do that work. The Academy is an early sign that the firms holding the AI stack have started down that road.

What Was Launched

The Academy was incubated by a16z and has raised $42 million from the firm and from individual investors including Tobi Lütke, Tony Xu, Garry Tan, and Palantir’s chief technology officer; Marc Andreessen and Erik Torenberg sit on its board. It is a for-profit company, separate from a16z but, in its own words, closely tied to it. Every founding fellow receives more than $50,000 in compute credits and technology resources from the partners, and more than forty further companies have signed on as hiring partners. Ben Horowitz called the school “an investment in the future of the country.” Gagan Biyani, its chief executive and a co-founder of Udemy and Maven, added that “College is still the right path for most students,” and the fellowship itself can serve as a gap year or a deferral from college. [4]

The partner list covers nearly the whole stack. Four of the ten build frontier models or the platforms those models run on. NVIDIA makes the chips. Palantir and Anduril are the companies most associated with selling software and autonomous systems to defense and intelligence agencies. Stripe and Coinbase run payment rails, one conventional and one crypto-native. Replit makes the tools on which a teenager now writes and ships code. According to the launch announcement, each partner offers advice on building the curriculum, access to its software, hardware, and compute, and time inside its operations. [5]

Taken one at a time, none of these features is unusual. Industry programs hand out cloud credits, bring in executives as lecturers, and invite recruiters to demo day. What sets the Academy apart is the combination, placed at the top of the talent distribution: a curriculum advised by the firms that own the frontier layers, instruction from their operators, and a first year spent building on their infrastructure. Most companies already build on AI systems they rent and do not control. The fellowship starts that arrangement at eighteen, and the student’s first workshop belongs to the landlord.

The Ladder, Priced

For most of the last century a degree worked as a claim on future training. The employer hired a graduate who could be taught, and the teaching happened on the job, in the junior work at the bottom of every profession. The worry now, voiced even in the fathers’ essay, is that AI will leave firms needing far fewer juniors to learn on; they ask whether law firms will still want armies of young associates drafting documents and finding precedent. [6] If that happens, the open question is where young people will gain their first real experience.

The Academy is one answer, for a very small number of people, and it answers a narrower need than the one the professions may be losing. A selective year for aspiring builders is no substitute for the paid junior work through which most graduates have learned a trade. What it offers is the part of that work the most ambitious valued most: early contact with real problems and with the people who run them. That contact now comes outside any employer’s payroll, organized by an investor and the firms at the center of the technology, and after the free founding year it will carry elite-university prices.

In 2011 Peter Thiel paid about two dozen young people $100,000 each to leave college for two years and build. [7] The Academy’s later program proposes to charge roughly what a top university costs, for entry to a network. Thiel treated the university as a cost worth paying people to escape; the Academy treats the network as an asset worth paying to enter.

Biyani’s first company grew up in the online-course boom that the New York Times, in November 2012, crowned “the year of the MOOC,” when the argument ran that lectures would become free, scale would win, and the campus would empty. [8] Lectures did become nearly free, and a teenager can now learn from a model how to train a robotics system. [9] The campus did not empty. What stayed scarce was everything a lecture never carried: peers, mentors, introductions, and problems with real consequences attached. The Academy charges for exactly that part of education. The 2012 wave bet on reach; this one bets on proximity.

The Company College

The East India College was a company college, a school owned or organized by the power its graduates would serve, built once that power had begun performing public functions and needed people fluent in its world. The Company already had revenue and armies. What it lacked was a steady supply of administrators who could collect taxes, sit as judges, and read the languages of the people they ruled. The college supplied a curriculum shaped to those tasks and a cohort who shared an outlook before they shared a posting. The historian Callie Wilkinson has shown that the long argument over the college, in Company headquarters, in Parliament, and in the press, was at bottom an argument over what qualities an imperial official should have and what relationship he was meant to serve among the Company, Britain, and India. [10]

The differences from 2026 are large. The Company governed territory. The Academy’s partners sell models, chips, payments, developer tools, and, in two cases, defense and intelligence platforms, and nobody is being trained to collect a land tax. Haileybury’s students arrived with a posting promised by the director who nominated them; the Academy selects competitively, and nothing in its launch material promises anyone a job. The Company was a single employer, while the Academy’s partners compete with one another every day, and talent raids and platform fights among them could fracture a cohort as easily as bind it.

What carries over is narrower. Some decisions made inside the stack already work like policy, and the clearest case so far involved two of the Academy’s partners. In February 2026 Anthropic refused to drop contract terms barring the military from using its models for domestic mass surveillance or fully autonomous weapons. The Pentagon, which wanted the models available for all lawful purposes, said it would not let a vendor “insert itself into the chain of command,” designated the company a supply-chain risk, and set off litigation that is still unfolding; OpenAI said it had written similar principles into its own agreement with the department. [11] On the Pentagon’s account, a vendor’s usage policy reached into military command; on Anthropic’s, it drew narrow limits around high-level uses. On either account, a term written by a private firm became a question of national security policy. Quieter versions of the same thing happen constantly: who receives frontier access, which defaults an agency inherits when it buys a system. The people who will direct that capability matter in roughly the way Haileybury’s writers mattered, and a school co-designed by the firms that hold it is well placed to shape what those people come to regard as normal. A cohort divided by rivalry might still learn that the choices worth fighting over are choices among the same ten firms.

Selection and formation are different mechanisms, and the argument depends on the second. Selection would mean the Academy finds teenagers who already believe the frontier firms are where serious work happens and delivers them there, which would make it an efficient recruiting pipeline with an unusual guest list, the reading most critics have reached for. Formation would mean that a year of building on partner infrastructure, taught by partner operators, under a curriculum the partners advise, produces that belief or hardens it. The launch materials describe a design capable of formation. They cannot show that it will work, because the first class has not arrived.

Elites have always kept hard problems and real stakes for their own children. The parents’ essay praises the Academy for taking young people seriously by “adding real stakes early.” [12] The change is in who supplies those stakes. For roughly fifty students a year, the firms that own the stack supply them, for a fee after the first year. For everyone else, the university keeps doing what it has always done at scale: it sorts, certifies, and holds the gate.

The Arc Runs Backward

For about a century, professional formation in the United States moved in one direction, into the university. Medicine is the standard case. In 1910 Abraham Flexner’s report for the Carnegie Foundation condemned the fee-dependent proprietary medical schools, many of them owned by the doctors who taught in them, and argued for placing medical education inside universities, joined to laboratory science and teaching hospitals. [13] Consolidation had begun before the report, and over the following decade licensing boards, professional ratings, and university policy did most of the closing and merging. [14] The settled preference was for training that no single employer owned.

At the top of the distribution, the university gave a gifted student three protections at once. No employer owned it. It set its own curriculum. It issued a credential that any employer could read, so the graduate’s standing did not depend on one network’s regard. The Academy’s students remain free to work anywhere, but two of those protections loosen: the curriculum is advised by the firms that will do much of the hiring, and the only credential is recognition by the network itself, which is most legible inside that network. For a talented eighteen-year-old in 1990, the three came bundled. For fifty students a year from 2027, they come apart.

History also supplies the strongest counter-reading. At least one well-known company college ended up ordinary. General Motors took over a Flint technical school in 1926 and for decades used it to train engineers and managers for its own production system; after GM reduced its operations in Flint, it divested the school in 1982, and the school became an independent private university, later renamed Kettering, that places co-op students across many industries. [15] The school’s fate followed the firm’s local power, much as Haileybury’s followed the Company’s. If the Academy follows that path, it is a talent funnel with a famous guest list, and the comparison with Haileybury is too large. Many earlier alternatives to college faded, fifty students is a rounding error against American higher education, and the chief executive himself says college remains the right path for most. That colder reading is fully available today, and the next few years will decide between the two.

The Charter Question

Haileybury’s end came through the Company’s charter. In 1853 Parliament renewed that charter for the last time and opened the Company’s civil service to competitive examination, so that appointees no longer had to pass through the college. In 1855 Parliament relieved the Company of its obligation to maintain Haileybury, and the last students were admitted in 1856. The college closed in January 1858. That August, in the aftermath of the rebellion of 1857, Parliament transferred the government of India from the Company to the Crown. [16] The college’s fate was decided inside the settlement of the Company’s power, and the quality of its teaching had little to do with it.

The Academy faces a far smaller test. Its two-year program needs regulatory approval, and the school says it will apply. [17] A state regulator will decide whether a program co-created with ten firms, with no grades and no degree, qualifies as a school. That decision is one limited point where public authority touches the arrangement, and it reaches only the school. The larger question, on what terms the firms that hold the AI stack exercise functions that look like governing, is being fought elsewhere, in procurement offices, courtrooms, and legislatures, and no one has settled it. The Academy will live under whatever settlement eventually arrives there.

The first class arrives next fall, and four things will show whether this reading holds. If the published curriculum gives the partners no design role and fellows spend real time with institutions outside the stack, the school will look more like a talent filter than a place of formation. If hospitals, agencies, or universities join the founding table on equal terms, it stops being organized around the owners of the stack. If the first cohorts scatter into work with no tie to the partners, the claim that it supplies the stack’s future operators weakens. And if by 2030 the frontier labs still hire their young staff mainly from selective universities, formation at the top has not moved. A failed startup, a small first class, steady university enrollment, or a regulator insisting on a degree would settle none of it.

Two fathers asked whether they would send their children to the Academy, and it was the right question for a parent. The older question is the one the directors of the East India Company answered for themselves in 1806: who forms the people who will run the new machinery, and in whose house. The founding class has not yet arrived, and the answer remains open. The university will keep the gate for the many. The companies that own the machines have started building a school for the people who will run them.

Notes

[1] “East India Company College,” Wikipedia (secondary source, relied on for the founding, the nomination system, the age range, the curriculum, and the figure of more than two thousand writers trained; consulted September 23, 2026), https://en.wikipedia.org/wiki/East_India_Company_College; Keith Tribe, “Professors Malthus and Jones: Political Economy at the East India College 1806-1858,” European Journal of the History of Economic Thought 2, no. 2 (1995): 327-354, https://doi.org/10.1080/09672569508538573 (Malthus as the first professor of political economy in Britain, appointed 1805 and succeeded on his death in 1834).

[2] “Horowitz Andreessen Academy Launches in San Francisco for the World’s Most Ambitious Young Builders,” press release, GlobeNewswire, September 22, 2026, https://www.globenewswire.com/news-release/2026/09/22/3366392/0/en/horowitz-andreessen-academy-launches-in-san-francisco-for-the-world-s-most-ambitious-young-builders.html (cohort size, fellowship length and start date, the planned two-year program and its tuition, co-creation with the founding partners and their curriculum role, compute credits, hiring partners, funding and board, for-profit status, and the Horowitz and Biyani statements); “a16z Is Challenging Silicon Valley’s Love for Drop-Outs by Launching a School,” TechCrunch, September 22, 2026, https://techcrunch.com/2026/09/22/a16z-is-challenging-silicon-valleys-love-for-drop-outs-by-launching-a-school/ (admission open to high school graduates younger than eighteen).

[3] Alex Danco and Packy McCormick, “Would We Send Our Kids to HAA?,” a16z, September 2026, https://a16z.news/p/should-we-send-our-kids-to-haa.

[4] Launch press release, note [2] (funding, board, for-profit status, compute credits, hiring partners, and the Horowitz and Biyani statements).

[5] Launch press release, note [2] (the partners’ curriculum-advice, access, and operations roles).

[6] Danco and McCormick, note [3] (the question whether law firms will still need young associates drafting documents and finding precedent).

[7] “New Fellowship Pays for College Kids to Drop Out,” All Things Considered, NPR, May 26, 2011, https://www.npr.org/2011/05/26/136690533/new-fellowship-pays-for-college-kids-to-drop-out. Reports of the first class’s size vary between twenty and twenty-four.

[8] Laura Pappano, “The Year of the MOOC,” New York Times, November 4, 2012, https://www.nytimes.com/2012/11/04/education/edlife/massive-open-online-courses-are-multiplying-at-a-rapid-pace.html.

[9] Gagan Biyani, “Introducing the Horowitz Andreessen Academy,” X article, September 22, 2026, https://x.com/i/article/2102396821008273618.

[10] Callie Wilkinson, “The East India College Debate and the Fashioning of Imperial Officials, 1806-1858,” Historical Journal 60, no. 4 (2017): 943-969, https://doi.org/10.1017/S0018246X16000492.

[11] Congressional Research Service, “Pentagon-Anthropic Dispute over Autonomous Weapon Systems: Potential Issues for Congress,” IN12669, April 2026, https://www.congress.gov/crs-product/IN12669 (the February 27, 2026 directive and the supply-chain-risk designation); “Pentagon Says It Is Labeling AI Company Anthropic a Supply Chain Risk ‘Effective Immediately,’” Yahoo News, March 2026, https://www.yahoo.com/news/articles/pentagon-says-labeling-ai-company-234341906.html (the Pentagon’s “chain of command” statement; Amodei’s description of the limits as relating to high-level usage areas rather than operational decisions); “Hegseth Declares Anthropic a ‘Supply Chain Risk to National Security,’” Yahoo News, February 2026, https://www.yahoo.com/news/articles/hegseth-declares-anthropic-supply-chain-000300350.html (the “all lawful purposes” demand; Altman’s statement that OpenAI put its prohibitions into its agreement); “A Timeline of the Anthropic-Pentagon Dispute,” Tech Policy Press, April 10, 2026, https://www.techpolicy.press/a-timeline-of-the-anthropic-pentagon-dispute/ (the course of the litigation through April). On September 25, 2026, the U.S. Court of Appeals for the D.C. Circuit ruled 2-1 in the Pentagon’s favor on the second of its two designations, delaying the decision’s effect to allow Anthropic to seek rehearing; a separate ruling by a federal judge in San Francisco, which found the parallel designation unlawful, remains in place. “U.S. Appeals Court Upholds Pentagon Designation of Anthropic as Supply Chain Risk,” CNBC, September 25, 2026, https://www.cnbc.com/2026/09/25/pentagon-anthropic-ai-risk-appeals-court.html.

[12] Danco and McCormick, note [3] (“adding real stakes early”).

[13] Abraham Flexner, Medical Education in the United States and Canada, Bulletin No. 4 (New York: Carnegie Foundation for the Advancement of Teaching, 1910); “The Flexner Report,” A History of UCSF, https://history.library.ucsf.edu/flexner_report.html (Flexner’s case for placing medical education within universities).

[14] “Flexner Report and Medical Education Reform, 1910,” Historia Medica, https://historiamedica.org/timeline/flexner-report/. A secondary timeline, relied on only for the direction and causes of consolidation before and after 1910, not for school counts, which vary across sources.

[15] “Our Story,” Kettering University, https://online.kettering.edu/about (GM’s 1926 takeover; service to GM engineers and managers exclusively; independence in 1982; renamed 1998); “Kettering University,” Wikipedia, https://en.wikipedia.org/wiki/Kettering_University (separation from GM after it reduced operations in Flint; expansion of co-op employers).

[16] Wikipedia, note [1] (the 1855 act relieving the Company of the obligation to maintain the college; closure in January 1858); Government of India Act 1858, 21 & 22 Vict. c. 106 (royal assent August 2, 1858), transferring the government of India to the Crown; “Haileybury,” FIBIS Wiki, Families in British India Society, https://wiki.fibis.org/w/Haileybury (secondary source: the 1853 act opening the service to competitive examination, last admissions in 1856, and the location of the college’s records at the British Library, IOR/J/1-4); Tribe, note [1] (closure following the introduction of open competitive examination).

[17] Launch press release, note [2] (the planned regulatory application).