Foundation Labs

Runs the lab behind Claude. Shapes the safety-first frontier.

Recent: Anthropic is reportedly preparing what could be a record ~$2T IPO; Amodei is publicly worried the wealth event will dilute mission alignment among staff (Fortune, Aug 2026).

Watch for: IPO timing and terms — the first frontier lab to test public markets at full scale.

The operator behind Anthropic — runs business operations, policy, hiring, and the safety org.

Recent: Named to the Time 100 Most Influential People (April 2026) alongside Dario; #2 on Forbes' 2026 America's Richest Self-Made Women.

Watch for: How she scales the commercial and operations org through IPO preparation.

The physicist behind scaling laws — the framework the whole industry uses to allocate compute.

Recent: Entered the Forbes Billionaires list in 2026 (#567) as Anthropic's valuation surged; continues to set Anthropic's research direction.

Watch for: Whether scaling-law orthodoxy holds as labs shift spend from pre-training to agentic RL.

Founded mechanistic interpretability as a field. The person actually looking inside the models.

Recent: Addressed the Vatican's AI gathering in June 2026, telling the industry it cannot govern itself.

Watch for: Whether interpretability results become deployment gates for frontier models rather than research artifacts.

Steers GPT and the most aggressive AI deployment strategy on earth.

Recent: Shipped the GPT-5.6 family (Sol/Terra/Luna) in July 2026 after conceding GPT-5 was a misfire; says better models are shelved for lack of GPU capacity.

Watch for: The GPT-6 Astra rollout and the trillions-of-dollars data center commitments behind it.

OpenAI's builder-in-chief; owns the compute strategy behind the models.

Recent: Called GPT-6 Astra 'the start of the AGI era' at its September 2026 launch, leaning hard on its computer-use abilities.

Watch for: Whether the 'compute is the new oil' thesis translates into OpenAI-owned infrastructure at scale.

Sets OpenAI's research agenda; oversaw GPT-4 and succeeded Ilya Sutskever.

Recent: Declared a fully automated AI researcher OpenAI's 'North Star,' targeting an autonomous research intern by September 2026 (MIT Tech Review, March 2026).

Watch for: Whether the automated-researcher milestone lands on schedule — the clearest public test of recursive AI R&D claims.

Built DeepMind, won a Nobel, and still frames the AGI-by-2030 case better than anyone.

Recent: Stepped down as DeepMind CEO in August 2026 amid Gemini delays and staff attrition, moving to chairman and Alphabet chief scientist while continuing to run Isomorphic Labs.

Watch for: Whether the reduced operating role frees him to fix frontier research — or marks the start of a longer exit.

Now runs Google's frontier AI day to day — Gemini models, research, app, and developer teams.

Recent: Took operational control of DeepMind in August 2026 after Hassabis stepped aside, tasked with closing the gap after Gemini 3.5 Pro slipped months behind schedule.

Watch for: Whether Gemini 3.5 Pro ships and re-establishes Google at the frontier under his leadership.

Europe's most credible frontier lab founder. Open-weight champion.

Recent: In talks to raise ~€3B at a ~€20B valuation (June 2026), with ARR around $400M and a stated path to $1B by year-end.

Watch for: European sovereignty contracts and whether Mistral becomes the continent's IPO candidate.

Ex-FAIR, worked on LLaMA. Decides what Mistral trains and why.

Recent: Leads model research at Mistral as of mid-2026, keeping open-weight releases competitive on a fraction of US lab compute.

Watch for: Whether Mistral's next open-weight release stays within reach of the closed frontier.

Built the world's largest single-site training cluster and folded it into a $1.25T company.

Recent: Merged xAI into SpaceX in February 2026 (rebranded SpaceXAI, Nasdaq listing in June); Colossus 2 in Memphis became the first gigawatt-class training cluster.

Watch for: Space-based data centers and whether Grok's cadence survives the merger and co-founder departures.

The $14B Scale AI deal made him Meta's frontier bet. Leads Meta Superintelligence Labs.

Recent: Shipped Muse Spark in April 2026, Meta's first major frontier model since the Superintelligence Labs reorg.

Watch for: Whether Meta's parallel AI orgs (MSL vs. Applied AI under Bosworth) converge or collide.

Turing Award winner who left Meta to bet against LLMs entirely — world models instead.

Recent: AMI Labs raised a $1.03B seed at a $3.5B pre-money valuation in March 2026, backed by Nvidia, Bezos Expeditions, Samsung, and Toyota Ventures.

Watch for: The first world-model demo that does something LLMs demonstrably cannot.

Ex-OpenAI CTO with the deepest bench of any startup lab — Schulman, Zoph, Weng, Tulloch.

Recent: Released Inkling, Thinking Machines' first broadly available model, in July 2026, and committed to a gigawatt of Nvidia compute from 2027.

Watch for: Whether the fine-tuning-first thesis (Tinker + Inkling) generates revenue before the $12B valuation needs re-marking.

OpenAI's former chief scientist running the purest bet in AI: one goal, no interim products.

Recent: Landed a long-term Nvidia partnership and investment in July 2026, with access to Vera Rubin compute at order-of-magnitude scale; a first model is reportedly imminent.

Watch for: Whether SSI ships anything public at all — at a $32B valuation on $6B raised, silence is the strategy until it isn't.

China's defining AI founder. Efficiency-first frontier models funded by his own hedge fund.

Recent: Raised $7.4B at a $52B valuation in June 2026 and became the world's wealthiest AI-model founder; a leaked briefing outlined an AGI-first plan and a chip program.

Watch for: The in-house chip effort — DeepSeek building silicon would change the export-control calculus.

Transformer-XL/XLNet author turned founder of China's #2 private AI lab (Kimi).

Recent: Unveiled Kimi K3 at Nvidia GTC 2026 — at 2.8T parameters the largest open-weights model ever, with native vision and 1M-token context; Moonshot hit a $35B valuation in July.

Watch for: Whether giant open-weights releases keep closing the gap with US closed frontier models.

Took China's first LLM company public. GLM models now benchmark against the US frontier.

Recent: Listed Zhipu on the Hong Kong exchange in January 2026 ($558M IPO); GLM-5.3 (August) claims benchmark wins over Claude Opus on Terminal-Bench.

Watch for: The stated break-even target of 2026-27 — the first public test of whether a Chinese frontier lab can fund itself.

'Attention Is All You Need' co-author building the enterprise and sovereign-AI counterweight to the AGI labs.

Recent: Pushed the sovereign AI case to G7 leaders in June 2026, framing the choice as 'sovereign AI or digital serfdom'; Cohere is at roughly $240M revenue.

Watch for: Government and sovereign-cloud deployments — Cohere's wedge against labs chasing superintelligence.