The archive · AI & Models · Strategic decision · 2025–2026
Thinking Machines' open-weights bet: Murati's lab talks $40B after a $2B seed
Mira Murati's ex-OpenAI lab bets open-weight models plus paid compute beat closed labs; record $2B seed, then $40B round talks on $100M+ revenue.
Thinking Machines
What the business is
Thinking Machines is an AI lab founded in early 2025 by former OpenAI CTO Mira Murati with roughly two dozen OpenAI veterans; it ships open-weight models such as Inkling and Tinker, a platform where customers pay usage-based compute fees to adapt those models on their own data.
Starting capital：$2B seed at a $12B valuation (July 2025), led by Andreessen Horowitz with Nvidia, GV, Lightspeed and Conviction Partners
How it started
Mira Murati, OpenAI's former CTO, founded Thinking Machines in early 2025 and roughly two dozen OpenAI researchers followed her. Investors backed the lab largely on that pedigree, closing a $2 billion seed round in July 2025 — one of the largest seed financings in history — at a $12 billion valuation, led by Andreessen Horowitz with Nvidia, GV, Lightspeed and Conviction Partners.
What happened
By November 2025 the lab was in talks at a $50–60 billion valuation on the strength of one shipped product, Tinker. Those talks collapsed in January 2026 after co-founder and CTO Barret Zoph left mid-all-hands and returned to OpenAI within the hour, joined by co-founder Luke Metz and researcher Sam Schoenholz; the private mark reset to the $12 billion seed price, and founding researcher Lilian Weng left for OpenAI in July. In that same month the company shipped Inkling, an open-weight model that earns revenue through usage-based compute fees on Tinker, and by September its annualized revenue run rate exceeded $100 million.
How it ended up
As of 2026-09-06 Thinking Machines is live and in late-stage funding talks: Accel is reported to be leading a $1B raise at a valuation of at least $40B, with Nvidia also discussed, after the company's first $50B valuation attempt fell apart under a leadership exodus.
Background
Thinking Machines is an AI lab founded in early 2025 by Mira Murati, OpenAI's former chief technology officer, together with roughly two dozen OpenAI veterans. Its founding bet was that a lean lab could compete with closed frontier labs not by selling API tokens, but by releasing open-weight models and charging usage-based compute fees to adapt them on a customer's own data.
Investors initially paid for pedigree: in July 2025 the lab closed a $2 billion seed round at a $12 billion valuation, one of the largest seed financings in history, led by Andreessen Horowitz with Nvidia, GV, Lightspeed and Conviction Partners. By November 2025 it was in talks at a $50–60 billion valuation on the strength of one shipped product, Tinker.
Those talks collapsed in January 2026 after co-founder and CTO Barret Zoph left mid-all-hands and returned to OpenAI, with co-founder Luke Metz and researcher Sam Schoenholz following; the private mark reset to the $12 billion seed price, and founding researcher Lilian Weng left for OpenAI in July. In July 2026 the company shipped Inkling, an open-weight model monetized through usage-based compute fees on Tinker.
By September 2026 the company's annualized revenue run rate exceeded $100 million, and TechCrunch and The Information reported on 2026-09-03 that Accel was in talks to lead a $1 billion round at a valuation of at least $40 billion — roughly 400 times run-rate revenue.
What has to be true
- Murati's OpenAI pedigree let a pre-product lab close a $2 billion seed at $12 billion, among the largest in AI history, with a16z, Nvidia, GV, Lightspeed and Conviction (TechCrunch).
- The open-weights-plus-paid-compute model inverts closed labs: give away weights and charge for the infrastructure used to adapt them on proprietary data (TFN).
- The first valuation attempt failed on people, not product: Zoph, Metz and Schoenholz returned to OpenAI in January 2026 and the $50–60B talks collapsed, resetting the mark to $12B (TFN).
- A shipped product changed the story: Inkling and Tinker pushed annualized revenue above $100 million by September 2026 (TechCrunch, TFN).
- The second attempt prices the company at roughly 400x run-rate revenue — a bet that the compute relationship, not the research pedigree, now carries the valuation (TFN).
What can be applied
Pedigree buys the first round, products buy the second: a founder brand raised a record seed, but the valuation only held once a shipped, revenue-generating model existed.
Aftermath
As of 2026-09-06 Thinking Machines remains in live operation with revenue growing: the $40 billion round was still in discussion, with Accel reported to lead and Nvidia also interested, and no final agreement announced. The company's open question is whether investors will pay twice after watching the first valuation attempt collapse under a leadership exodus — the departures of Zoph, Metz, Schoenholz and Weng back to OpenAI left a smaller founding team carrying the product and the pitch.
Sources
- Accel reportedly in talks to lead $1B round for Thinking Machines at $40B valuation
- Thinking Machines in talks for $1B at $40B valuation as Mira Murati's AI lab races ahead
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card