A study of 21,166 profiles · data through August 2026

Where the frontier AI labs went to college

We looked at the education history of everyone working at OpenAI, Anthropic, Google DeepMind, xAI, and Thinking Machines Lab. Two schools dominate. After that, each lab recruits from a different world.

8,231
distinct colleges represented across the five labs
1 in 10
of all lab staff studied at Stanford or UC Berkeley
93%
of profiles carry education history, so coverage is near-complete

1 · The feeder schools

Stanford and Berkeley supply more people than the next three combined

Number of current employees who studied at each college, coloured by where they work now. Bars count people, not degrees: someone with two degrees from one school is counted once.

Anthropic OpenAI Google DeepMind xAI Thinking Machines
Stanford Stanford at anthropic: 238 Stanford at openai: 535 Stanford at deepmind: 345 Stanford at xai: 55 Stanford at tml: 15 1188 UC Berkeley UC Berkeley at anthropic: 216 UC Berkeley at openai: 466 UC Berkeley at deepmind: 284 UC Berkeley at xai: 72 UC Berkeley at tml: 22 1060 Carnegie Mellon Carnegie Mellon at anthropic: 81 Carnegie Mellon at openai: 265 Carnegie Mellon at deepmind: 264 Carnegie Mellon at xai: 46 Carnegie Mellon at tml: 10 666 Harvard Harvard at anthropic: 173 Harvard at openai: 296 Harvard at deepmind: 142 Harvard at xai: 34 Harvard at tml: 8 653 MIT MIT at anthropic: 105 MIT at openai: 267 MIT at deepmind: 202 MIT at xai: 26 MIT at tml: 14 614 UCLA UCLA at anthropic: 76 UCLA at openai: 177 UCLA at deepmind: 104 UCLA at xai: 36 UCLA at tml: 2 395 U Penn U Penn at anthropic: 93 U Penn at openai: 163 U Penn at deepmind: 72 U Penn at xai: 34 U Penn at tml: 6 368 Georgia Tech Georgia Tech at anthropic: 35 Georgia Tech at openai: 131 Georgia Tech at deepmind: 146 Georgia Tech at xai: 42 Georgia Tech at tml: 3 357 Cambridge Cambridge at anthropic: 52 Cambridge at openai: 66 Cambridge at deepmind: 221 Cambridge at xai: 9 348 Cornell Cornell at anthropic: 55 Cornell at openai: 135 Cornell at deepmind: 122 Cornell at xai: 25 Cornell at tml: 3 340 NYU NYU at anthropic: 74 NYU at openai: 142 NYU at deepmind: 92 NYU at xai: 28 NYU at tml: 3 339 Oxford Oxford at anthropic: 56 Oxford at openai: 96 Oxford at deepmind: 167 Oxford at xai: 15 Oxford at tml: 4 338 IIT (any campus) IIT (any campus) at anthropic: 29 IIT (any campus) at openai: 59 IIT (any campus) at deepmind: 215 IIT (any campus) at xai: 28 IIT (any campus) at tml: 2 333 USC USC at anthropic: 53 USC at openai: 137 USC at deepmind: 99 USC at xai: 31 USC at tml: 2 322 Columbia Columbia at anthropic: 66 Columbia at openai: 121 Columbia at deepmind: 110 Columbia at xai: 14 Columbia at tml: 3 314

Counts are absolute, so large universities have an advantage; we do not adjust for graduating class size. Cambridge, Oxford and IIT rank high largely on the strength of Google DeepMind, which accounts for half to two-thirds of each.

2 · The signature pipelines

Each lab over-recruits from a different corner of the world

Colleges that send more people to one lab than that lab’s overall size would predict. A 2.5x reading means a college is two and a half times as concentrated at that lab as it is across the five labs as a whole.

1.0x = no over-representation DEEPMIND École Polytechnique 2.52x 46 of 63 EPFL 2.49x 44 of 61 ETH Zurich 2.38x 85 of 123 Sharif Univ. of Technology 2.25x 30 of 46 IIT (any campus) 2.23x 215 of 333 ANTHROPIC Dartmouth College 1.99x 31 of 84 Trinity College Dublin 1.97x 23 of 63 Georgetown University 1.80x 27 of 81 UC Santa Barbara 1.78x 33 of 100 OPENAI University of San Francisco 1.60x 25 of 40 San Jose State University 1.57x 71 of 116 San Francisco State University 1.49x 54 of 93 XAI University of Maryland 1.39x 21 of 119 UT Austin 1.23x 37 of 237

Restricted to colleges with at least 40 people across the five labs and at least 20 at the lab in question, so these are pipelines rather than coincidences. Thinking Machines is too small to appear: its only qualifying entry is UC Berkeley at 2.60x.

3 · Two kinds of elite

Prestige and doctorates are different things

Placing each lab’s technical staff by the share from a QS Computer Science top-50 university against the share holding a doctorate.

40% 30% 20% 10% 50% 60% 70% 80% Share from a QS Computer Science top-50 university → Share holding a doctorate → research-institute profile product-company profile Thinking Machines Google DeepMind OpenAI Anthropic xAI 57–60%: a statistical tie Technical staff 3,900 700 120

Technical roles only: investors listed on company pages and annotation staff are excluded. Circle area shows the number of technical staff, so Thinking Machines’ position rests on 121 people against DeepMind’s 3,888.

4 · Beyond the famous names

A commuter school in San Jose sends more people to OpenAI than Cambridge does

Colleges outside the QS Computer Science top 50 that still put 70 to 120 people inside the frontier labs. San Jose State and San Francisco State both over-index at OpenAI, whose offices are a short train ride away.

University of Maryland 119 San Jose State University 116 Cal Poly San Luis Obispo 102 Northeastern University 96 San Francisco State University 93 Boston University 82 Purdue University 73 Arizona State University 72

Tier is the QS World University Rankings by Subject 2026, Computer Science and Information Systems. Ranking outside a subject top 50 is not a judgement of a school: it reflects QS’s indicators, which favour research output and international reputation.

How we did this

Source. Public professional-profile employment and education histories, aggregated by a commercial data provider; snapshot of August 2026. We resolved each lab’s company page by identity and checked it against the employer names people list on it.

Population. Every current employee of the five labs: 21,166 people, of whom 93% carry education history. We count a person once per college, so multiple degrees from one school do not inflate its total. Someone who attended two colleges is counted at each, so college totals cannot be added together to get a distinct headcount.

Exclusions. We removed 324 people listed on company pages as investors, advisors or board members, because they are not staff. We removed high schools, and bootcamps and accelerators such as Y Combinator (83 people), General Assembly (62) and Udacity (53), which appear in the education field but are not colleges. For the doctorate and technical-staff cuts we also set aside annotation and AI-tutor roles, a workforce of 710 at xAI.

Name handling. Colleges are grouped to the parent institution, so Wharton counts as Pennsylvania, Harvard Business School as Harvard, and Haas as Berkeley. UC campuses stay separate. IIT campuses are combined into one entry, which flatters IIT against schools counted individually.

Known limits. Absolute counts favour large universities; we do not normalise for class size, so this measures supply, not selectivity or per-graduate success. Counts of current staff run high: people are slower to remove an employer than to add one, so anyone who has left and not yet updated their profile is still counted. Comparing our totals against publicly reported headcounts suggests the effect is largest at OpenAI, on the order of 15 to 20 per cent. In the other direction, self-reported profiles under-report people who keep their education private. The five labs are not the whole industry, and rankings are one publisher’s view, not an objective measure of quality.

The counts behind every chart on this page are available as a CSV, under a CC BY 4.0 licence. Reuse it freely, with a link back to this page.

How to cite this

Trayo Research. “Where the frontier AI labs went to college.” September 2026. https://www.trayo.ai/research/ai-lab-colleges

Charts and figures may be reproduced or redrawn with attribution. If you would like a cut of the data we have not published, such as a single institution broken out by lab, email [email protected].

Download the data (CSV)

Get the next one

We publish original data on how teams are actually built. Drop your email and we will send the next report when it lands. No product pitch, no weekly newsletter.

Any address is fine, including a personal one. Unsubscribe in one click.

How this was built. Resolving 21,166 people to the right employer, collapsing 8,231 institutions so Wharton lands under Pennsylvania and Haas under Berkeley, and separating staff from investors and contractors is the same entity resolution Trayo runs to detect hiring and headcount signals inside target accounts. This time we pointed it at five companies instead of a customer’s ICP.

See the buying signals for your accounts

Drop in your work email and Trayo returns 8 real buying signals for your company — free, in about 30 seconds. No call, no credit card.

Get my 8 signals free Get my 8 signals free Start your 7-day trial Start your 7-day trial

Right signal. Right person. Right now.

$39/month • 7-day free trial • Cancel anytime

Start free trial Start free trial

Try Trayo

Drop in your work email, we'll spin up your account and email you when it's ready.

Already have an account? Sign in