Mira Murati

Who they are

Mira Murati is co-founder and CEO of Thinking Machines Lab — former OpenAI CTO who launched Tinker, a fine-tuning API for open-weight AI models, after raising a $2 billion seed round at a $12 billion valuation.

Person

Murati studied Mechanical Engineering at Dartmouth, graduating in 2012 — an engineering foundation that shaped how she thinks about building systems, not just shipping software. Before AI, she moved through Goldman Sachs as an intern, then Zodiac Aerospace, then Tesla as Senior Product Manager on the Model X, and then Leap Motion in a product/engineering role. She joined OpenAI in its early years, climbed to CTO, and briefly served as Interim CEO during the November 2023 board crisis that briefly ousted Sam Altman. She left OpenAI in September 2024 and launched Thinking Machines Lab in February 2025 — structured as a public benefit corporation, focused on customizable, human-in-the-loop AI. She writes occasionally: a piece in Daedalus on language and creativity in large models is her most cited long-form work, and she's spoken at the Johns Hopkins Bloomberg Center, Dartmouth's Thayer School, and the WIRED Big Interview series. The through-line is operator-turned-builder: each move added technical credibility, product depth, and institutional weight before she finally stepped out to found her own lab.

Company

The most recent development is a multibillion-dollar deal Thinking Machines Lab signed with Google Cloud in April 2026, expanding use of Google's AI Hypercomputer infrastructure powered by Nvidia GB300 systems. That followed a March 2026 strategic partnership with NVIDIA — an undisclosed investment plus a multi-year agreement to deploy at least one gigawatt of Vera Rubin computing capacity starting early 2027. Both deals came after turbulence: a $50 billion funding round stalled at end of 2025 amid unclear revenue paths beyond Tinker, and by January 2026 those talks had collapsed. The company also weathered a talent exodus — by April 2026, Meta had hired seven founding members including Mark Jen, Yinghai Lu, and Andrew Tulloch, and co-founders Barret Zoph and Luke Metz returned to OpenAI in January 2026. Soumith Chintala, creator of PyTorch, joined as CTO in January 2026, and in May 2026 the company previewed TML-Interaction-Small, its first interaction model, processing audio, video, and text simultaneously with a turn-taking latency of 0.40 seconds.

Market

Thinking Machines Lab competes directly with OpenAI, Anthropic, Google DeepMind, and Meta AI — a field defined by extreme valuations, massive compute expenditure, and intense competition for a small pool of frontier researchers. The AI sector in 2025–2026 is being shaped by chip supply constraints, US export controls, and the comp wars that are actively raiding early-stage labs like Thinking Machines. Its differentiated bet is customizable, fine-tunable AI — Tinker targets developers and research institutions who want to adapt open-weight models — carving a niche between closed API providers and raw open-source, though the $50 billion round stalling suggests the market is still assessing whether that niche converts to durable revenue.

Network

Murati's inner circle at Thinking Machines Lab is heavily drawn from OpenAI alumni: John Schulman (Chief Scientist), Lilian Weng (co-founder), and Soumith Chintala (CTO, recruited from Meta) form the current leadership core after Barret Zoph's January 2026 departure back to OpenAI. She also invested in Worktrace AI, an automation startup founded by former OpenAI PM Angela Jiang, signaling she's building a broader network of bets around ex-OpenAI operators.

John Schulman· Chief Scientist, Thinking Machines Lab
Lilian Weng· Co-founder, Thinking Machines Lab
Soumith Chintala· CTO, Thinking Machines Lab (joined January 2026)
Angela Jiang· Founder, Worktrace AI (Murati investor)
Barret Zoph· Former CTO, Thinking Machines Lab (departed January 2026 to OpenAI)

How they likely show up

Career moved from mechanical engineering → product management → CTO → founder, each step adding a new layer of capability → likely thinks in systems and compounding bets rather than single-product wins.
Served as Interim CEO during OpenAI's November 2023 crisis without seeking the permanent role → suggests comfort operating under pressure, low ego about title.
Structured Thinking Machines Lab as a public benefit corporation with stated focus on AI alignment and human-in-the-loop design → values and governance architecture are deliberate choices, not marketing language.
Public writing signal is 'occasional' — a Daedalus essay, targeted talks at Hopkins and Dartmouth — rather than high-volume LinkedIn output → communicates when she has something precise to say, not to maintain a persona.
Raised a $2 billion seed before shipping a product, then launched Tinker as a developer fine-tuning API → willing to operate at long time horizons with large resource commitments before revenue clarity.
Talent losses to Meta, OpenAI, and xAI have been publicly reported as an ongoing challenge → she's managing a retention and culture problem in real time, which will be front of mind.

Conversation tips

→ Reference the Daedalus essay on language and creativity — it's her most substantive long-form statement of view, and citing it signals you read past the headlines.
→ Ask about the public benefit corporation structure — it's a deliberate governance choice and she'll have a specific answer about what it constrains and enables.
→ Don't treat the talent departures as a gotcha; she's already navigating them publicly. A more useful angle is asking what the reconstituted team around Chintala and Schulman is optimizing for.
→ She came up through product (Tesla Model X, Leap Motion) before AI — grounding a conversation in product decisions rather than pure research will land better than treating her as only a researcher.
→ Avoid generic questions about 'the future of AI' — her public talks are specific and technical; match that register.

Toolbox

Openers

Open on the NVIDIA Vera Rubin deal — a March 2026 commitment to deploy at least one gigawatt of next-generation compute starting early 2027. That's a very specific infrastructure bet worth unpacking: what does a lab of 208 people plan to do with gigawatt-scale compute?
Mention the Daedalus essay on language and creativity in large models — it's her most substantive published view on what these systems actually are, and it's a clean signal you engaged with her thinking directly rather than just her press coverage.
Reference the TML-Interaction-Small preview from May 2026 — a multimodal model with a 0.40-second turn-taking latency. It's the first concrete model she's shipped publicly, and it marks a shift from infrastructure-building to model-shipping.

Discovery questions

Tinker launched as a fine-tuning API for open-weight models — how has the actual use pattern from research institutions and enterprise users shaped where you're taking the product next?
You structured Thinking Machines Lab as a public benefit corporation from day one — what does that structure practically constrain or enable when you're in a funding negotiation or a talent war?
Soumith Chintala joining as CTO in January 2026 brought the PyTorch lineage into the lab — how has that changed how the team thinks about the tooling layer underneath your models?

Avoid

Don't lead with questions about the November 2023 OpenAI board crisis or Sam Altman — she addressed it publicly on the Pivot Podcast and at Johns Hopkins and has clearly moved on; reopening it reads as tabloid interest, not substantive preparation.

Make it yours