What does $40 billion buy when the product is a training API and several of the people who made the company famous have already left? That’s the uncomfortable question under the Thinking Machines Lab $40 billion valuation. It surfaced on November 6 in a Business Insider report. The company still hasn’t confirmed it.
The reported deal is roughly $5 billion in new money, with Accel leading, at a $40 billion pre-money price. That would more than triple what investors paid in July 2025. Business Insider said it could close as soon as that week.
The size isn’t the interesting part. Giant AI rounds stopped surprising anyone a while ago. The interesting part is what investors think they’re paying for, and how much of the answer depends on people who may not stay.
What’s been reported, and what hasn’t been confirmed
Start with the sourcing, because it’s thinner than the headlines suggest.
Everything about the November round traces back to one outlet. Business Insider cited unnamed sources. A Thinking Machines representative declined to comment and Accel didn’t respond, according to Let’s Data Science, which relayed the story. We couldn’t open the Business Insider piece directly, so every detail here comes from outlets summarizing it.
Some write-ups already describe the money as “raised.” Nobody has shown that. Treat the Thinking Machines Lab $40 billion valuation as a round in progress, not a done deal.
Here’s what the report says:
- Size: about $5 billion
- Lead investor: Accel
- Price: $40 billion pre-money
- Nvidia: reportedly in separate talks to put in about $2.5 billion
That last line matters more than it looks. AI News Weekly’s summary doesn’t say whether Nvidia’s money sits inside the $5 billion or on top of it. If it’s on top, total new money would approach $7.5 billion. No one has cleared that up.
The numbers: how the Thinking Machines Lab $40 billion valuation came together
The figure didn’t appear out of nowhere. It has drifted through the press since September, and the path tells you something.
- July 2025: a $2 billion seed round at a $12 billion post-money valuation, with Andreessen Horowitz leading. Accel and Nvidia joined that round too.
- Late 2025: the company reportedly sought $50 billion or more. Those talks never closed.
- September 3, 2026: The Information reported talks for “$1 billion-plus” at roughly $40 billion.
- September 6: The Next Web, relaying The Information, put the talks at $5–6 billion. Accel was in line to lead, and Nvidia would supply roughly half.
- November 6: Business Insider reported about $5 billion at $40 billion pre-money, with Accel leading.
Two things jump out. The round grew while the price held steady. And $40 billion sits at least 20% below the $50 billion the company wanted less than a year ago.
So it’s an odd kind of win. A tripled valuation that’s also a climb-down.
The math is simple enough. $40 billion over a $12 billion seed mark works out to about 3.3x. Add $5 billion to a $40 billion pre-money price and you get roughly $45 billion post-money. The Next Web’s version, at the top of the $6 billion range, landed nearer $46 billion.
What Thinking Machines Lab sells today: Tinker
For a company priced like a frontier lab, Thinking Machines has a modest storefront. Its product is Tinker, which the company’s own page calls “frontier-grade infrastructure for training your own models.”
In plain terms, Tinker is a fine-tuning API. You bring a model and your data. Tinker computes the gradients, updates the weights and samples outputs, so you don’t have to run a training cluster yourself. If “weights” and “gradients” sound abstract, our explainer on how large language models work walks through what training actually changes.
A few details from the product page:
- It supports many open-weight model families, including Qwen, Kimi, GLM, DeepSeek, GPT-OSS and Nemotron. It also runs the company’s own Inkling models.
- Customers use it for forecasting, search agents and classification.
- Pricing is usage-based: per million tokens for training, sampling and prefill, plus $0.10 per GB per month for stored checkpoints.
- Serverless inference is still in beta.
That’s a real business. It’s also a developer tool, and nothing on the page points to a consumer product.
How much does it earn? Sources disagree. The Information’s Stephanie Palazzolo reported annualized revenue in the “hundreds of millions,” self-reported and unaudited. TechCrunch, citing a different source, put it above $100 million. Neither source breaks the figure out by product, so be wary of anyone quoting a precise Tinker run rate.
The Next Web’s math puts the price at roughly 80 to 400 times revenue. A range that wide barely counts as a number.
So what are investors actually paying for?
If not revenue, what holds up the Thinking Machines Lab $40 billion valuation? We’d split it into four bets.
The people who stayed
John Schulman, a co-founder and the chief scientist, is still there. Soumith Chintala joined to run technical work. In frontier AI, a handful of senior researchers can be most of the asset. Investors are betting this group can still do frontier work.
A product that already earns
Tinker shows the team can ship something customers pay for. The revenue figures conflict, but nobody disputes that money comes in.
An option on a frontier model
This is the big one. We can’t see anyone paying 80 times revenue or more for a fine-tuning API alone. The price only makes sense if investors expect the Inkling models, or whatever follows them, to compete at the frontier. That’s a bet on a future product.
Nvidia, on both sides of the table
This part is our own inference, not reporting. In the September version of the deal, Nvidia would supply roughly half the money. Nvidia also sells the chips that AI labs spend their raises on. When a supplier funds a customer, some of that cash can come back as orders.
The round can be perfectly real and still tell you less about outside demand than the headline suggests. For why frontier labs spend so heavily on compute, see our breakdown of AGI hardware requirements.
The talent question: how many founders have left?
Here’s where the price gets hard to square. According to The Next Web:
- Andrew Tulloch left for Meta in late 2025.
- Barret Zoph, a co-founder and the CTO, left in January. OpenAI announced his return, and TNW says he has since moved to Google.
- Co-founder Luke Metz left.
- Lilian Weng stepped down in July, citing health. OpenAI re-signed her two days later.
Some coverage puts the share of departed founders at about a third. One headline counts ten founding exits. We couldn’t verify either tally on a page we could open, so treat both as rough.
Even the confirmed list stands out. The $12 billion seed round was, more than anything, a bet on a specific group of people. Several of them now work at the very labs Thinking Machines wants to compete with. Investors would now pay more than three times as much for a team that’s partly different.
The counterpoint is fair. Schulman stayed, and Chintala arrived, so the lab still has a real core. It’s a different core from the one investors backed in July 2025.
What it says about the frontier-lab funding race
Step back and the Thinking Machines Lab $40 billion valuation looks less like an outlier and more like a symptom. Our read: money keeps chasing the few teams that could, in theory, build frontier models. Being on that short list carries a premium before the product exists. We’ve tracked who’s actually leading that contest in AGI Development Race: Who Is Ahead in 2026?
There’s a human side to the math, too. When a valuation rests on people, those people become the most contested resource in the industry. Every exit turns into a valuation event. The departures above show the pressure runs both ways: labs poach researchers, and they lose them just as fast.
What to watch next
- Does the round close? A company statement would settle the “raised” versus “raising” muddle.
- Where does Nvidia’s money sit? Inside the $5 billion, or on top of it?
- Do official revenue figures surface? Self-reported, unaudited numbers can’t carry this price for long.
- Can Inkling compete? That’s the frontier bet in a single question.
- Does the team hold?
Our guess, and it’s only a guess: the round closes near the reported terms. After the abandoned $50 billion ask, $40 billion already looks like the compromise price. Investors will probably pay. The harder question is whether a valuation like this means much when the lab’s most valuable assets can resign by email.
Would you pay a premium for a team over a product? Most of this market seems to have decided it would.