Tom Brown

Tom B. Brown is an American software engineer and AI researcher, co-founder and Chief Compute Officer of Anthropic, the public-benefit corporation that develops the Claude family of large-language models. He was previously a research-engineering lead at OpenAI, where he was the lead author of the GPT-3 paper. As of May 2026, he runs the technical organizations responsible for converting Anthropic's compute into trained models and inference capacity, including the AWS Trainium2 partnership announced as Project Rainier at AWS re:Invent in December 2024.

At a glance

• Education: Master of Engineering in computer science, Massachusetts Institute of Technology (2010), with prior coursework in computational cognitive science.
• Current role: Co-founder and Chief Compute Officer of Anthropic, since 2021.
• Key contributions: Lead author of "Language Models are Few-Shot Learners" (the GPT-3 paper, 2020); co-author of "Scaling Laws for Neural Language Models" (2020), "Deep Reinforcement Learning from Human Preferences" (2017), and "Constitutional AI: Harmlessness from AI Feedback" (2022); engineering and infrastructure leadership for the Claude model line and the Project Rainier Trainium2 cluster.
• Public profile: Y Combinator co-founder background (Grouper, W12); featured guest on the Y Combinator Lightcone Podcast (August 2025); customer-keynote speaker at AWS re:Invent 2024.
• X / Twitter: @nottombrown
• GitHub: @nottombrown
• LinkedIn: Tom Brown

Origins

Public biographical material on Brown is comparatively thin compared with the Anthropic co-founders who have Wikipedia entries. The most-cited primary accounts of his pre-OpenAI career are the August 2025 Y Combinator Lightcone podcast episode and the August 2017 Medium post "How I made the switch to AI research."

Brown attended MIT and completed a Master of Engineering in computer science around 2010, with coursework in computational cognitive science. He has spoken publicly on the Lightcone podcast about an early-stage academic record that included a B-minus in linear algebra, citing the experience as part of the case that the path into AI research is open to engineers without a research-PhD background.

Career

After completing the MIT degree around 2010, Brown spent roughly eight years building consumer and developer-facing software startups before transitioning to AI research. The two named startups in the period are MoPub, the mobile advertising platform where Brown worked as a founding engineer scaling the ad-serving API, and Grouper, the social-club company he co-founded with Michael Waxman through Y Combinator's W12 batch in 2012. He served as Grouper's Chief Technology Officer from 2012 through approximately 2014. MoPub was acquired by Twitter in 2013, and Grouper wound down around the time Brown began the AI transition.

In late 2014 and early 2015 Brown spent approximately six months in self-directed machine-learning study, taking online courses and joining South Park Commons. After the self-study window he reached out to Greg Brockman at the time of OpenAI's launch announcement and joined the lab as a research engineer.

Brown's first OpenAI tenure ran approximately from May 2016 to May 2017, working on infrastructure projects including the StarCraft research environment. He then spent a year at Google Brain before returning to OpenAI in late 2018 as Research Engineering Lead for what would become the GPT-3 effort. He remained in that role through the December 2020 departure with Dario Amodei, Daniela Amodei, and the cohort that founded Anthropic.

In early 2021 the group formally incorporated Anthropic as a Delaware Public Benefit Corporation. The seven public co-founders are Dario Amodei, Daniela Amodei, Brown, Sam McCandlish, Jared Kaplan, Jack Clark, and Chris Olah. Brown took responsibility for the training and infrastructure organization, with his Chief Compute Officer title documented on his Crunchbase profile and confirmed by the AWS re:Invent 2024 keynote. The first public Claude model launched in March 2023, and the company has shipped successive Claude generations through Claude Opus 4.7 in April 2026.

At AWS re:Invent 2024, Brown appeared as the sole customer speaker during Peter DeSantis's Monday Night Live keynote to announce Project Rainier, a Trainium2 compute cluster scaled to deliver more than five times the exaflops of Anthropic's prior training infrastructure. AWS subsequently disclosed in 2025 that Project Rainier was activated with approximately 500,000 Trainium2 chips dedicated to Anthropic workloads.

Affiliations

• MoPub: Founding engineer, late 2000s to early 2010s.
• Grouper (YC W12): Co-founder and CTO, 2012 to approximately 2014.
• OpenAI: Member of Technical Staff, 2016-05 to 2017-05.
• Google Brain: Research staff, approximately 2017 to 2018.
• OpenAI: Research Engineering Lead (GPT-3), 2018-12 to 2020-12.
• Anthropic: Co-founder and Chief Compute Officer, 2021 to present.

Notable contributions

Brown's body of work is concentrated on training infrastructure, distributed-systems engineering for large-scale model runs, and lead authorship on one of the most-cited papers in modern AI. Many of the papers below are co-authored at scale, with Brown listed as first or senior author on a subset.

• "Deep Reinforcement Learning from Human Preferences" (June 2017). NeurIPS paper co-authored with Paul Christiano, Jan Leike, Miljan Martic, Shane Legg, and Dario Amodei, jointly between OpenAI and DeepMind. The paper introduced the reinforcement-learning-from-human-feedback methodology that became a foundational training step for the modern instruction-following assistant. Anthropic press materials describe Brown as a co-inventor of RLHF.
• "Scaling Laws for Neural Language Models" (January 2020). The OpenAI scaling-laws paper led by Jared Kaplan and Sam McCandlish on the empirical relationship between compute, model size, dataset size, and language-model loss. Brown is among the named co-authors.
• "Language Models are Few-Shot Learners" (May 2020, arXiv 2005.14165). The 175-billion-parameter GPT-3 paper, with Brown as lead author. The paper established large-scale autoregressive pre-training as the dominant paradigm in language modeling, demonstrated few-shot prompting from a single trained model, and is widely cited as one of the most influential AI publications of the past decade. Senior co-authors include Dario Amodei, Sam McCandlish, Jared Kaplan, Jack Clark, Ilya Sutskever, and Alec Radford.
• "Extracting Training Data from Large Language Models" (December 2020). USENIX Security paper on extraction attacks against language models, co-authored with Nicholas Carlini, Florian Tramer, Eric Wallace, and others.
• "Constitutional AI: Harmlessness from AI Feedback" (December 2022). Anthropic methodology paper introducing the alignment technique in which a model trains to follow a written constitution of principles via self-critique and revision. Brown is among the named co-authors.
• "Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback" (April 2022). Anthropic paper documenting the RLHF pipeline used in early Claude precursors.
• Project Rainier and the Trainium2 partnership (December 2024). Anthropic's compute partnership with AWS, announced at AWS re:Invent during Brown's customer keynote on Monday Night Live and scaled in 2025 to approximately 500,000 Trainium2 chips.

Investments and boards

• Anthropic (AI): Co-founder and Chief Compute Officer, 2021 to present. Public Benefit Corporation incorporated in Delaware. Approximately $73 billion cumulative funding through the February 2026 Series G at a $380 billion post-money valuation.

No public personal angel-investor activity on record outside the Anthropic role in AI, semiconductors, datacenters, software, or energy as of May 2026. Brown's footprint in this section is concentrated in his founding and operating role at Anthropic rather than a parallel investing program.

Network

Brown's longest-running professional relationships are with his fellow Anthropic co-founders, all of whom he worked with at OpenAI before the 2021 founding: Dario Amodei, the chief executive who led the GPT-3 effort with Brown as lead author of the resulting paper; Daniela Amodei, the president; Sam McCandlish, with whom he co-authored the scaling-laws paper; Jared Kaplan, the chief science officer and lead author of the same paper; Jack Clark, the policy lead; and Chris Olah, the interpretability lead. The seven-person founding cohort has been comparatively stable since 2021.

His broader OpenAI co-author network includes Greg Brockman, the OpenAI president whom Brown contacted to enter the lab and a co-author on the GPT-3 paper; Ilya Sutskever, the former OpenAI Chief Scientist; Andrej Karpathy, an early OpenAI research scientist; Mira Murati, the former OpenAI Chief Technology Officer; and John Schulman, the OpenAI co-founder and reinforcement-learning lead. The 2017 RLHF paper, co-authored with Christiano, Leike, Shane Legg, and Dario Amodei across OpenAI and DeepMind, is the documented start of the cross-lab collaboration that produced modern instruction-following assistants. His pre-OpenAI relationships include Grouper co-founder Michael Waxman and the Y Combinator W12 batch cohort.

Position in the field

As of May 2026, Brown is one of the most consequential lead authors of the modern large-language-model era. The GPT-3 paper at arXiv 2005.14165 has accumulated more than 80,000 citations, sitting alongside the original transformer paper as one of the most-cited AI publications of the past decade. The lead-authorship credit on a paper of this footprint, combined with the co-founding of Anthropic and the Chief Compute Officer remit, places Brown in a small group of senior frontier-lab leaders whose research record and operating role are both widely cited.

His career path is structurally distinctive among Anthropic's founding cohort. Where Dario Amodei and Jared Kaplan came in through physics doctorates and Chris Olah through self-directed research without a degree, Brown's path ran through eight years of consumer and developer-platform startup work, six months of self-directed machine-learning study at South Park Commons, and a research-engineering trajectory through OpenAI, Google Brain, and back to OpenAI. The credential profile distinguishes him from most peers in lead-author positions on frontier-tier model papers.

The Chief Compute Officer role at Anthropic has no exact analog at peer frontier labs; the closest comparable position is the chief-infrastructure or VP-of-compute role that other labs split across multiple senior leaders. Brown's public profile is concentrated in technical and infrastructure conversations rather than policy and AGI framing, and he has not pursued the high-frequency podcast or congressional-testimony cadence of the Amodei siblings.

Outlook

Open questions over the next 6 to 18 months:

• Project Rainier scale-up. The performance and cost economics of the AWS Trainium2 cluster relative to Nvidia-based training, and whether the partnership extends to a successor generation of AWS silicon for the next Claude training run.
• Compute supply chain. Anthropic's compute mix between AWS Trainium, Google TPU, and Nvidia GPU as the next training cycle begins, given the company's multi-cloud posture and the AWS strategic-investment relationship.
• Successor model training cadence. The pre-training timeline for the Claude generation beyond the 4.x line, with Brown's organization as the operational counterpart to the engineering build-out.
• Public commentary cadence. Whether Brown's public-talk cadence increases as Anthropic's infrastructure footprint becomes a more prominent part of the company's external profile.
• Engineering-leadership recruiting. Whether the senior-engineering organization at Anthropic continues to recruit from the OpenAI and Google Brain alumni networks at the rate observed through the 2024 to 2025 period.

Sources

• Tom Brown - LinkedIn. Brown's official LinkedIn profile listing the MIT degree, the MoPub and Grouper roles, and the OpenAI to Anthropic trajectory.
• Tom Brown - Co-Founder and Chief Compute Officer @ Anthropic. Crunchbase profile documenting the Chief Compute Officer title and career history.
• Tom B Brown - Google Scholar. Google Scholar profile listing publications and citation counts, with the GPT-3 paper above 80,000 citations.
• Language Models are Few-Shot Learners. The May 2020 GPT-3 paper led by Brown.
• Scaling Laws for Neural Language Models. The January 2020 OpenAI scaling-laws paper co-authored by Brown, led by Jared Kaplan and Sam McCandlish.
• Deep Reinforcement Learning from Human Preferences. The June 2017 NeurIPS paper introducing RLHF, jointly between OpenAI and DeepMind, with Brown among the listed co-authors.
• Constitutional AI: Harmlessness from AI Feedback. The December 2022 Anthropic Constitutional AI paper with Brown among the named co-authors.
• How I made the switch to AI Research. Brown's August 2017 Medium post documenting the transition from startup engineering to AI research at OpenAI.
• Anthropic Co-founder: Building Claude Code, Lessons From GPT-3 & LLM System Design. Y Combinator Lightcone Podcast episode, August 21, 2025, with the most detailed publicly available account of Brown's career path and Anthropic compute strategy.
• In Conversation With Anthropic Co-Founder Tom Brown. Salesforce Ventures fireside-chat coverage from December 2023.
• AWS re:Invent 2024 - Customer Keynote Anthropic. Brown's December 2024 customer-keynote at Peter DeSantis's Monday Night Live keynote, announcing Project Rainier and the Trainium2 partnership.
• Anthropic to Train Next Generation Project Rainier LLM Using AWS Processors. Industry coverage of the Project Rainier announcement and Trainium2 deployment scale.
• Grouper (YC W12). Y Combinator company page for Brown's pre-OpenAI startup.
• Feature image: text-mode card generated via scripts/make_lab_card.py, used as a fallback because no Anthropic press-kit headshot, Wikipedia portrait, or other credit-cleared photograph of Brown was located in May 2026.