Thinking Machines Valuation: Why the $12B Mark Survived a Failed $50B Round

Thinking Machines Lab is worth $12 billion right now — the exact same number investors agreed to in July 2025, before the company had shipped a single product. In between, the price briefly touched $55-60 billion and then round-tripped all the way back down. I think even $12 billion is generous for a company that just lost two co-founders back to the lab it was built to compete with.

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 writing the largest AI seed check in history — of course this is worth tens of billions. That story survived exactly six months of contact with reality.

What Is Thinking Machines Lab's Valuation in 2026?

Thinking Machines Lab's valuation is $12 billion as of mid-2026, the same post-money price set in its $2 billion seed round in July 2025. A proposed follow-on round targeting $50 billion, and later reported by Bloomberg to be in talks as high as $55-60 billion, fell apart in January 2026, so the price simply reverted rather than resetting.

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 had five months to decide whether a roughly 4x-5x jump was justified by one fine-tuning product and no disclosed revenue, and they said no.

The Bull Case, and Why I'm Not Buying It

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 — fifteen months 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, but it's not a $50 billion business with undisclosed revenue and a single named customer relationship publicly reported.

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, with sources telling Wired he allegedly shared confidential information with rivals — an allegation OpenAI itself has disputed. 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. Headcount has since stabilized around 200 employees as of mid-2026, up from a low point after the departures, but "stabilized" after a founder-level exodus is a recovery story, not a growth story.

Thinking Machines Isn't Alone in the Zero-Revenue 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. Compare that to a company like Perplexity, which at least has $450 million in disclosed ARR supporting its $23 billion mark. Thinking Machines sits closer to the SSI end of that spectrum than the Perplexity end, which is exactly why the $50 billion round didn't clear.

Figures blended from company disclosures, TechCrunch, and Value Add VC coverage of each company. Data 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 eleven months 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 even if the figure isn't public, 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.

My disagreement 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.