Companies supply the most prior-work hires at 7 of the 9 frontier AI labs
Data as of
- Prior-work transitions to frontier labs
- 620
- From companies
- 48%
- From universities
- 41%
- Feeder organizations
- 248
Which organizations do researchers and builders leave directly before joining a frontier AI lab, and how do the hiring pipelines of the frontier labs differ? Across the 9 frontier labs with organization records — Google DeepMind, OpenAI, Anthropic, Mistral AI, DeepSeek, Thinking Machines Lab, Meta Superintelligence Labs, xAI and Safe Superintelligence, counting Google Brain and Google DeepMind together — we trace 620 career transitions by 550 people from an immediately preceding work role at 248 parent organizations. Companies supply 296 of those 620 prior-work transitions (48%), universities supply 255 (41%), and research institutes, nonprofits and government bodies supply 69 (11%). At the lab level, companies supply the largest share of prior-work transitions at 7 of the 9 frontier labs. Universities supply the majority at Google DeepMind and DeepSeek.
Key findings
- The 9 frontier labs divide into 7 company-led pipelines and 2 university-led pipelines: Companies supply a majority or plurality of prior-work transitions at OpenAI, Anthropic, Mistral AI, Thinking Machines Lab, Meta Superintelligence Labs, xAI and Safe Superintelligence, reaching 100% (2 of 2) at Safe Superintelligence, 87% (20 of 23) at Thinking Machines Lab and 73% (11 of 15) at Meta Superintelligence Labs and xAI. Universities supply 67% (22 of 33) of prior-work transitions at DeepSeek. Universities supply 60% (128 of 214) of prior-work transitions at Google DeepMind.
- Google, Meta and OpenAI supply 20% of all prior-work transitions into the frontier labs: Google leads all feeder organizations with 56 prior-work transitions, followed by Meta with 36 and OpenAI with 34. Together, those 3 companies supply 126 transitions (43% of the 296 company-origin transitions), and 42 of the 56 transitions from Google (75%) come from people who held roles in Google DeepMind or Google Brain.
- The University of California, Berkeley and the University of Oxford lead the 102 university feeders: The largest academic feeders are University of California, Berkeley (20), University of Oxford (19), Stanford University (18), University of Cambridge (11), Harvard University (10) and Massachusetts Institute of Technology (10). Universities in the United States supply 47 of Google DeepMind's 128 university transitions (37%) and universities in the United Kingdom supply 35 (27%); universities in the United States supply 38 of OpenAI's 51 (75%) and 20 of Anthropic's 31 (65%), and universities in China supply 18 of DeepSeek's 22 (82%).
- The largest single feeder-to-lab corridors run into OpenAI and Anthropic: Google to OpenAI leads all organization-to-lab pathways with 29 prior-work transitions, ahead of Google to Anthropic and OpenAI to Anthropic (16 each), Meta to Google DeepMind and University of Oxford to Google DeepMind (11 each), and University of California, Berkeley to OpenAI (10).
- Company feeders overtook university feeders in 2023: Universities supplied at least as many prior-work transitions into the frontier labs as companies in every year from 2010 through 2022 (141 university transitions and 87 from companies across that span). From 2023 through 2026, companies led every year, supplying 209 of 371 transitions (56%) and 114 (31%) from universities.
Company-fed and university-fed frontier labs
Of the 999 people recorded at the 9 frontier labs from each lab's founding year on, 550 (55%) entered from a work role at another parent organization across 620 prior-work transitions.
Stacked horizontal bars for 9 frontier AI labs showing prior-work transitions from universities (255), companies (296) and research institutes, nonprofits and government (69).
Source: People of Superintelligence career records from biographies and OpenAlex, as of 22 October 2026.
Data table
| Frontier lab | Prior-work transitions | People | Feeder organizations | From universities | University share | From companies | Company share | From institutes, nonprofits and government | Largest prior-work feeder(s) |
|---|---|---|---|---|---|---|---|---|---|
| Google DeepMind | 214 | 212 | 116 | 128 | 60% | 63 | 29% | 23 | Meta and University of Oxford (11 each) |
| OpenAI | 161 | 158 | 86 | 51 | 32% | 98 | 61% | 12 | Google (29) |
| Anthropic | 109 | 106 | 54 | 31 | 28% | 57 | 52% | 21 | Google and OpenAI (16 each) |
| Mistral AI | 48 | 48 | 34 | 15 | 31% | 23 | 48% | 10 | Meta (5) |
| DeepSeek | 33 | 33 | 21 | 22 | 67% | 11 | 33% | 0 | Microsoft (6) |
| Thinking Machines Lab | 23 | 23 | 9 | 2 | 9% | 20 | 87% | 1 | Meta (8) |
| Meta Superintelligence Labs | 15 | 15 | 10 | 3 | 20% | 11 | 73% | 1 | OpenAI (5) |
| xAI | 15 | 15 | 10 | 3 | 20% | 11 | 73% | 1 | Google (5) |
| Safe Superintelligence | 2 | 2 | 2 | 0 | 0% | 2 | 100% | 0 | OpenAI and Pioneer (1 each) |
Across the 9 labs, 2 draw more than half of their prior-work transitions from universities and 6 draw more than half from companies, and Mistral AI draws a plurality (48%) from companies. At DeepSeek, universities supply 22 of 33 prior-work transitions (67%), compared with 11 from companies (33%). At Google DeepMind, universities supply 128 of 214 prior-work transitions (60%), more than twice the 63 from companies (29%), with 23 (11%) from research institutes, nonprofits and government bodies. Both halves of Google DeepMind share that academic tilt: of the 128 university transitions, 107 entered Google DeepMind proper and 21 entered Google Brain.
Companies lead prior-work transitions at the other 7 frontier labs. At OpenAI, companies supply 98 of 161 prior-work transitions (61%), ahead of the 51 from universities (32%). Among the five other company-led labs, company feeders supply 2 of 2 prior-work transitions (100%) at Safe Superintelligence, 20 of 23 (87%) at Thinking Machines Lab, 11 of 15 (73%) at Meta Superintelligence Labs, 11 of 15 (73%) at xAI and 23 of 48 (48%) at Mistral AI. At Anthropic, companies supply 57 of 109 prior-work transitions (52%), universities supply 31 (28%), and research institutes, nonprofits and government bodies supply 21 (19%). Google DeepMind has the most third-sector transitions (23 of 214, 11%). Mistral AI has the highest third-sector share (10 of 48, 21%).
The largest feeder organizations overall
The 620 prior-work transitions come from 248 parent organizations: 102 universities, 106 companies and 40 research institutes, nonprofits and government bodies. Across those 248 organizations, 79 (32%) send 2 or more prior-work transitions into the frontier labs, and 169 send a single transition each. At the top of the distribution, 13 organizations send at least 8 transitions each and supply 260 of the 620 total transitions (42%).
Horizontal bar chart of the 13 largest feeder organizations into the 9 frontier labs, led by Google with 56 transitions.
Source: People of Superintelligence career records from biographies and OpenAlex, as of 22 October 2026.
Data table
Google is the largest feeder overall, supplying 56 prior-work transitions (55 people) across 5 frontier labs, followed by Meta with 36 transitions (36 people) across 5 labs and OpenAI with 34 transitions (32 people) across 6 labs. Together, Google, Meta and OpenAI supply 126 prior-work transitions by 109 people — 20% of all prior-work transitions into the frontier labs and 43% of the 296 transitions from companies. Across all feeders, Microsoft reaches the most frontier labs, sending prior-work recruits to 8 of the 9 labs (22 transitions), ahead of OpenAI and Allen Institute for AI (Ai2) at 6 labs each.
Of the 56 prior-work transitions from Google into other frontier labs, 42 (75%) involve people who held roles in Google DeepMind (30) or Google Brain (21). At Microsoft, 9 of the 22 prior-work transitions (41%) come from Microsoft Research, and at Meta, 16 of the 36 transitions (44%) come from FAIR (Meta).
Among the 102 university feeders, University of California, Berkeley leads with 20 prior-work transitions across 3 frontier labs, ahead of University of Oxford (19 across 4 labs), Stanford University (18 across 5 labs), University of Cambridge (11 across 2 labs), Harvard University (10 across 3 labs) and Massachusetts Institute of Technology (10 across 4 labs). Among the 40 research institutes, nonprofits and government bodies, Allen Institute for AI (Ai2) leads with 8 prior-work transitions across 6 frontier labs, followed by Inria with 5 and Mila, Machine Intelligence Research Institute and Open Philanthropy with 4 each.
How the main feeders differ lab by lab
Each frontier lab draws from a distinct set of primary feeder organizations:
- Google DeepMind (214 prior-work transitions from 116 organizations): Meta and University of Oxford lead with 11 transitions each, followed by University of Cambridge and University of California, Berkeley (9 each). Google DeepMind's company transitions come from Meta (11), OpenAI (5), Microsoft (4), Twitter (3), Amazon (2), Apple (2), Baidu (2), Codeium (2) and IBM (2). Across its 128 university transitions, universities in the United States supply 47 (37%), universities in the United Kingdom supply 35 (27%) and universities in Canada supply 17 (13%), with 35 of 35 transitions from universities in the United Kingdom and 15 of 17 from universities in Canada entering Google DeepMind proper; 13 of Google Brain's 21 university transitions came from universities in the United States.
- OpenAI (161 prior-work transitions from 86 organizations): Google leads with 29 transitions (18% of OpenAI's prior-work arrivals), more than twice runner-up University of California, Berkeley (10). Other leading feeders are Meta (9), Stanford University (7), Microsoft (5), Harvard University (4), Tesla (4), Baidu (3), Massachusetts Institute of Technology (3) and Uber (3). Universities in the United States supply 38 of OpenAI's 51 academic transitions (75%).
- Anthropic (109 prior-work transitions from 54 organizations): Google and OpenAI (16 each) and Machine Intelligence Research Institute, University of Oxford and Stanford University (4 each) supply 44 of Anthropic's 109 prior-work transitions (40%). Its next largest feeders are Allen Institute for AI (Ai2) (3), Meta (3), Microsoft (3), Open Philanthropy (3), Vercept (3), ETH Zurich (2), Government of the United Kingdom (2), Harvard University (2), King's College London (2), New York University (2) and Vector Institute (2). Universities in the United States and the United Kingdom supply 20 and 6 of its 31 academic transitions.
- Mistral AI (48 prior-work transitions from 34 organizations, plus 15 education-only transitions): Meta leads with 5 prior-work transitions (Baptiste Rozière, Devendra Singh Chaplot, Guillaume Lample, Olivier Duchenne and Thibaut Lavril), followed by Google with 3. Universities in the United Kingdom and France supply 10 of its 15 university work transitions (67%).
- DeepSeek (33 prior-work transitions from 21 organizations): Microsoft leads with 6 transitions (Daya Guo, Jiawei Wang, Yixuan Wei, Yu Wu, Zhenda Xie and Zhibin Gou), ahead of Peking University with 4. Universities in China supply 18 of its 22 academic transitions (82%).
- Thinking Machines Lab (23 prior-work transitions from 9 organizations): Meta (8: Andrea Madotto, Horace He, Myle Ott, Naman Goyal, Piotr Dollár, Sam Gross, Sam Shleifer and Soumith Chintala) and OpenAI (5: Andrew Tulloch, Barret Zoph, John Schulman, Lilian Weng and Rowan Zellers) together supply 13 of its 23 transitions (57%).
- Meta Superintelligence Labs (15 prior-work transitions from 10 organizations): OpenAI leads with 5 transitions (Huiwen Chang, Hyung Won Chung, Jack Rae, Shuchao Bi and Trapit Bansal), followed by Stanford University with 2.
- xAI (15 transitions) and Safe Superintelligence (2 transitions): Google leads xAI with 5 transitions (Christian Szegedy, Hieu Pham, Igor Babuschkin, Yuhuai (Tony) Wu and Zihang Dai), and OpenAI and Pioneer lead Safe Superintelligence with 1 each (Daniel Gross and Ilya Sutskever).
From university pipelines to inter-lab recruiting
Grouping prior-work transitions by year tracks how the feeder mix shifted over time. From 2010 through 2020, when Google DeepMind (including Google Brain) and OpenAI were the active frontier labs in this group, universities supplied 108 of the 185 prior-work transitions (58%) and companies supplied 61 (33%). From 2021 through 2026, as 7 additional frontier labs entered, the company share rose to 54% (235 of 435 prior-work transitions) and the university share fell to 34% (147 transitions).
Line chart of annual prior-work transitions into the 9 frontier labs from 2014 to 2026, comparing companies, universities, and institutes, nonprofits and government.
Source: People of Superintelligence career records from biographies and OpenAlex, as of 22 October 2026.
Data table
| Year | All prior-work transitions | From universities | University share | From companies | Company share | From institutes, nonprofits and government | From Google, Meta and OpenAI |
|---|---|---|---|---|---|---|---|
| 2010 | 3 | 3 | 100% | 0 | 0% | 0 | 0 |
| 2011 | 3 | 2 | 67% | 1 | 33% | 0 | 0 |
| 2012 | 2 | 2 | 100% | 0 | 0% | 0 | 0 |
| 2013 | 8 | 7 | 88% | 1 | 13% | 0 | 0 |
| 2014 | 13 | 10 | 77% | 1 | 8% | 2 | 0 |
| 2015 | 20 | 11 | 55% | 8 | 40% | 1 | 1 |
| 2016 | 29 | 16 | 55% | 11 | 38% | 2 | 2 |
| 2017 | 26 | 13 | 50% | 9 | 35% | 4 | 2 |
| 2018 | 23 | 13 | 57% | 7 | 30% | 3 | 2 |
| 2019 | 36 | 19 | 53% | 14 | 39% | 3 | 3 |
| 2020 | 22 | 12 | 55% | 9 | 41% | 1 | 3 |
| 2021 | 32 | 15 | 47% | 14 | 44% | 3 | 10 |
| 2022 | 32 | 18 | 56% | 12 | 38% | 2 | 8 |
| 2023 | 67 | 26 | 39% | 31 | 46% | 10 | 17 |
| 2024 | 67 | 23 | 34% | 41 | 61% | 3 | 15 |
| 2025 | 104 | 32 | 31% | 59 | 57% | 13 | 29 |
| 2026 | 133 | 33 | 25% | 78 | 59% | 22 | 34 |
In every year from 2010 through 2022, universities supplied at least as many prior-work transitions into the frontier labs as companies (141 university transitions and 87 from companies across those 13 years). In 2023, company transitions passed university transitions (31 to 26) and stayed ahead through 2026. Across 2023–2026, companies supplied 209 of 371 prior-work transitions (56%), ahead of the 114 from universities (31%), with Google, Meta and OpenAI alone supplying 95 (26% of all prior-work transitions in that period). Within each era, Google DeepMind drew 66% of its prior-work transitions from universities before 2021 (88 of 133) and 49% in 2021–2026 (40 of 81). OpenAI drew a company majority in both periods: 60% (31 of 52) before 2021 and 61% (67 of 109) from 2021 on.
Career data comes from the biographies on this site and from OpenAlex author records, as of 22 October 2026. People whose OpenAlex record may be incomplete or mixed are left out of citation figures.