Anthropic’s White House negotiation over its newest AI models, Fable 5 and Mythos 5, did not get smoothed over by CEO Dario Amodei. Instead, one of Anthropic’s quietest cofounders, Tom Brown, stepped in as the dealmaker. Brown is the company’s chief compute officer, and he helped negotiate an agreement that persuaded the U.S. Commerce Department to lift export restrictions on those flagship models.

That is the real twist here: the political leverage came from the person responsible for getting the chips. In a moment when the government tightened the rules around what high-end AI can be exported, Brown helped translate Anthropic’s compute reality into regulatory acceptance. Anthropic’s Head of Public Policy Sarah Heck worked alongside him on the White House talks, but the letter about the export controls that Wired reported was addressed directly to Brown, underlining just how central he became. The episode thrust one of AI’s least visible power centers into view: the engineer who secures computing power also became an important diplomat.

This matters for more than curiosity points. Compute is not a side quest in frontier AI. Anthropic’s approach relies on access to major hardware providers and the ability to route different tasks across them. The company uses Amazon Trainium chips, Nvidia GPUs, and Google TPUs, distributing work depending on what each system does best and what access is available. When that system strains, everything strains: training schedules, model iteration timelines, and product roadmaps. That is why Brown leads the high-stakes compute push, which has included deals across the industry.

And Anthropic has had reasons to push. This spring, the company was dealing with a compute crunch amid rabid demand for its Claude Code product. Brown and his team announced a multi-billion-dollar deal to run Anthropic’s models on SpaceX’s huge data center, Colossus. Beyond that, Brown has spoken on behalf of Anthropic at Amazon Web Services’ re:Invent conference, and he was quoted in a 2023 Google Cloud release hyping an AI infrastructure collaboration. In other words, this is not a “paper” role. It’s a role built on nonstop coordination with big, powerful institutions.

So when the U.S. government becomes involved, the compute leader is often the only person who can speak the language. Export controls are not abstract policy; they can directly change the feasibility of shipping or deploying certain capabilities. In this case, the result was a rollback of those restrictions, which cooled tensions and removed a major constraint on Anthropic’s newest flagship models. The headline implication is clear: political relief showed up because Brown could get the deal done, even when the spotlight usually follows the CEO.

Brown’s path to this kind of influence runs through startups, OpenAI, and a late conversion to AI research. Out of MIT, he worked at the language-teaching startup Lingt, then at the mobile advertising startup MoPub. He cofounded Grouper, a startup that matched pairs of trios for in-person meetups, with Michael Waxman as part of Y Combinator’s winter 2012 batch. Waxman later described Brown as optimistic about AI while also emphasizing Brown’s character traits and integrity, saying, “Knowing Tom’s character and that he’s in the room at the leading edge of some of these really important decisions definitely makes me just feel better and sleep easier at night.” Brown left Grouper after a few years and then pivoted.

When he decided he did not yet have the skills for AI research, Brown reportedly ran a months-long self-study sprint, including a Coursera course, Kaggle projects, and the book “Linear Algebra Done Right.” He even bought a GPU chip with Y Combinator credits. In 2015, Waxman remembers Brown inviting him to a machine learning study group. Brown eventually joined OpenAI through Greg Brockman, and at OpenAI he worked on a project related to the game “StarCraft.” By the time he was publishing, his focus increasingly connected engineering to AI capability and safety. Brown was one of six authors from OpenAI and DeepMind on a 2017 paper that paved the way for Reinforcement Learning from Human Feedback, and his Google Scholar page lists more than 140,000 citations.

Two more landmark papers landed in 2020. Brown helped with “Scaling Laws for Neural Language Models,” which established the scaling laws that became a blueprint for frontier AI, and he was a main author, leading the engineering effort, on “Language Models are Few-Shot Learners,” introducing GPT-3 and the prompt-based control ideas that later powered ChatGPT. During a 2024 discussion with his cofounders, Brown said scaling and safety teams at OpenAI, which reported to Daniela and Dario Amodei, treated the scaling law idea seriously, including the view that there would come a time where “humanity will hand off control to transformative AI.” That group cofounded Anthropic in 2021.

Today, Brown’s role blends technical depth with constant negotiation. He has spoken about compute in ways that reveal his evolving view of speed and impact, noting that he started out not thinking it would be that fast, and that he developed sympathy for people who still think it is too fast. He also pledged, like other cofounders, to donate most of his wealth. Bloomberg estimates his net worth is nearly $8 billion, tied to his Anthropic stake and the company’s $965 billion valuation, and he has pledged to give away 80% of his wealth.

There is a financial timing angle here too. The rollback of White House restrictions comes when Anthropic is eyeing an initial public offering, which would provide employees with a liquidity event and flood the company with cash for expansion. If you sit on a board or run finance, you care about capital events and risk, and if you run compute, you care about supply, time, and access. Brown bridging those two worlds may be the kind of advantage that does not show up in typical investor decks. But in a regulated, hardware-constrained race, it can determine whether the next models ship on schedule. For executives in similar roles, the message is unmissable: when policy intersects with infrastructure, your compute chief may be your best negotiator, not just your technical operator.