Thinking Machines Valuation 2026: $12B Priced, $40B+ Talks as of October 21

Thinking Machines Lab's last priced valuation is still $12 billion — the exact same number investors agreed to in July 2025, before the company had shipped a single product. Talks in late 2025 reportedly floated $50-60 billion and did not close, and the company is now the subject of new talks for a $40 billion-plus round that, as of October 21, 2026, has not closed either. Even $12 billion looks generous for a company that has since lost several co-founders, most of them to the lab it was built to compete with: Barret Zoph and Luke Metz went back to OpenAI in January (Zoph has since moved to Google), and Lilian Weng returned to OpenAI in July. The new number deserves the same skepticism.

The consensus read on Mira Murati's Thinking Machines Lab has always been simple: pedigree is destiny. Former OpenAI CTO, a team stacked with ex-OpenAI researchers, and a16z leading one of the largest seed rounds in Silicon Valley history — of course this is worth tens of billions. That story survived exactly six months of contact with reality, and the company's full funding record is on its Pulse hub.

Thinking Machines Lab: Key Numbers (as of October 21, 2026)

Only the July 2025 seed is a closed, priced round. Every later valuation figure is reported talks, and revenue figures come from unnamed sources, not company disclosures.

What Is Thinking Machines Lab's Valuation in 2026?

Thinking Machines Lab's last priced valuation is $12 billion, the post-money price set in its $2 billion seed round in July 2025. In November 2025, Bloomberg reported early talks for a new round at a roughly $50 billion valuation, possibly rising to $55-60 billion (per Reuters); those talks did not close, so the price simply reverted rather than resetting. On September 3, 2026, TechCrunch reported, citing The Information, that the company is in talks to raise $1 billion at a valuation of at least $40 billion; The Information has since put the raise at $5-6 billion on a pre-money valuation of at least $40 billion, per The Next Web, with existing investor Accel in talks to lead — but that round has not closed.

That reversion is the part nobody is talking about with enough skepticism. Markups that don't happen aren't neutral events — they're a rejection. Prospective backers were asked whether a roughly 4x-5x jump, a few months after the seed, was justified by one fine-tuning product and no disclosed revenue, and they walked away. The new talks come with reported revenue figures, not disclosed ones — TechCrunch reported the company's "annual revenue run rate stands at over $100 million, according to a source with knowledge of the company's financials," and The Information has put it in the hundreds of millions — but numbers from unnamed sources are not audited figures, and the round still isn't signed.

Thinking Machines Lab Valuation History

The September 2026 round is reported as in talks, not closed, and the November 2025 talks did not close; $12B (July 2025) remains the last disclosed priced round. Nvidia also invested in March 2026 alongside its chip partnership, but neither company said how much (per The Next Web). Checked October 21, 2026.

The Bull Case, and Why It Doesn't Hold Up

The bull case isn't crazy on paper. Thinking Machines raised its $2 billion seed from a syndicate that included Nvidia, AMD, ServiceNow, Cisco, Accel, and Jane Street alongside a16z — strategic money, not just financial money, betting that the team that built ChatGPT's underlying infrastructure could do it again independently. Nvidia backed that conviction further with a multiyear chip supply deal signed in March 2026, and the company shipped Nvidia's preferred customer profile: compute-hungry, well-capitalized, technically credible.

But strategic investment from chip vendors is a demand signal for GPUs, not proof of product-market fit. Tinker, the company's first shipped product, only launched in October 2025 — about a year after Murati left OpenAI and three months after the seed closed. A fine-tuning API priced per million tokens plus storage is a real business, and a reported run rate over $100 million suggests it's selling. But that number comes from an anonymous source, not the company, and against a $40 billion pre-money price The Next Web works out the multiple at roughly 80 to 400 times revenue, depending on where in The Information's "hundreds of millions" the real figure sits. That's a valuation built on what Tinker might become, not on what it is.

The January 2026 Exodus Nobody Priced In

Here's the part that should worry anyone still underwriting a mega-round: in January 2026, co-founders Barret Zoph and Luke Metz returned to OpenAI, along with researcher Sam Schoenholz. Zoph was reportedly fired by Murati after telling her he was considering leaving. A source close to Thinking Machines alleged to Wired that he had shared confidential company information with competitors, which Wired could not verify with Zoph; OpenAI's Fidji Simo wrote in a staff memo that OpenAI doesn't share Murati's concerns about him. Soumith Chintala stepped in as the new CTO.

A single senior departure at a research lab is normal turnover. Two co-founders returning to the exact company the startup was founded to compete with, during the same window a $50 billion round was on the table, is not noise — it's a data point about internal conviction that outside investors don't get to see until it's already happened. TechCrunch put headcount at around 140 in April 2026, with the lab hiring researchers away from Meta even as Meta picked off its founding members — but rebuilding the bench after a founder-level exodus is a recovery story, not a growth story.

Thinking Machines Isn't Alone in the Pedigree-Priced Club

What makes Thinking Machines interesting isn't that it's uniquely overpriced — it's that it's the clearest test case for a pattern across the entire frontier-lab category. Safe Superintelligence is valued at $32 billion with zero disclosed revenue, on Ilya Sutskever's name alone, and Flapping Airplanes went from a $180 million seed to a $6 billion valuation in nine months with no shipped product at all. Compare that to a company like Perplexity, which at least has $450 million in disclosed ARR supporting its $20 billion mark (Sep 2025, reported by The Information via TechCrunch). When the $50 billion round was being shopped, Thinking Machines had disclosed no revenue, which put it at the SSI end of that spectrum — and that round didn't clear. A reported $100 million-plus run rate moves it toward the middle, though The Information's reported "hundreds of millions" could put it much closer to Perplexity's $450 million; neither figure is a company disclosure.

Figures blended from company disclosures, TechCrunch, and Value Add VC coverage of each company. Thinking Machines row updated October 1, 2026; other rows as of August 2026.

What the headline misses

It would be lazy to write this off as a pure failure. A $12 billion valuation that survived a leadership exodus, a failed markup attempt, and more than a year of intense scrutiny is still a real number — most startups don't have $2 billion in the bank to weather that kind of storm. Tinker has real paying usage — TechCrunch's source puts the run rate over $100 million, even though the company hasn't published a figure — Inkling gives the company an actual open-weight release to point to as of July 2026, and Google Cloud's compute partnership is a legitimate signal that infrastructure players see something worth supporting. The failed round also isn't unique to Thinking Machines — plenty of 2021-vintage growth rounds died the same way without killing the company underneath them.

The disagreement here isn't with $12 billion existing. It's with treating $12 billion as a floor rather than a question mark. A seed valuation set before any product shipped, that survived only because a bigger round failed to materialize, is not the same thing as a market-tested price. It's an un-marked-down mark, and those are exactly the numbers that get quietly cut in the next financing cycle if Tinker and Inkling don't convert reputation into recurring revenue within the next two or three quarters.