The Thinking Machines valuation is heading to roughly $40 billion. Accel is in talks to lead a raise of at least $1 billion, The Information reported on September 3, 2026, with Nvidia weighing participation. That is 3.3x the $12 billion price Andreessen Horowitz paid in July 2025 — but below the $50 billion the company chased last year, and set against revenue reported at only $100 million-plus.
Mira Murati’s startup is 19 months old. It has one open-weights model, one fine-tuning product, and about 100 employees. Investors are being asked to underwrite a price that assumes almost none of that stays true for long.
How much is Thinking Machines raising, and at what valuation?
At least $1 billion, at a valuation of at least $40 billion pre-money, according to The Information. Accel — already on the cap table — is discussing leading the round. Nvidia, an existing investor and the company’s infrastructure partner, may participate. Nothing is signed, and the terms reported are talks, not a close.
The step-up is the headline. The last priced round, in July 2025, valued the company at $12 billion on a $2 billion raise led by Andreessen Horowitz, with Nvidia, AMD, Cisco and Jane Street alongside. That was described at the time as among the largest seed financings ever recorded.
The valuation timeline
| Date | Event | Amount | Valuation |
|---|---|---|---|
| Feb 2025 | Company founded by Mira Murati | — | — |
| Jul 2025 | Seed round led by a16z | $2B | $12B |
| Nov 2025 | Valuation talks reported (unclosed) | — | up to $60B |
| Late 2025 | Valuation sought (not reached) | — | $50B |
| Mar 2026 | Nvidia multiyear chip partnership | undisclosed equity | — |
| Sep 2026 | Accel-led round in talks | $1B+ | ~$40B |
Read that table backwards and the story changes. A company that discussed $60 billion in November 2025 is reportedly pricing at $40 billion ten months later. On the last marked price it is a 3.3x markup. On what the founders were asking, it is a haircut of a third.
What does Thinking Machines actually sell?
Two things: Tinker, an API for fine-tuning language models on the company’s own infrastructure, launched October 1, 2025; and Inkling, an open-weights foundation model released July 15, 2026. The weights are free. The compute to customize them is not — that usage fee is the business.
Inkling is not a small artifact. The company’s launch post describes “975B total parameters, 41B active,” a Mixture-of-Experts design, multimodal input, controllable reasoning effort, and “a context window of up to 1M tokens.”
On Tinker it ships in 64K and 256K context configurations. Thinking Machines launched it at “a 50% discount for a limited time.”
The strategic logic is coherent. Give away the model, charge for the customization loop, and own the layer where enterprise data meets frontier weights. It is the same wedge Tencent used with Hy4 — commoditize the model, monetize the surface around it.
Is the revenue anywhere near the price?
No — and the sources do not even agree on the number. TechCrunch puts the annual revenue run rate at “over $100 million.” The Information, as relayed by PYMNTS, says the company is “generating at least a few hundred million dollars in annualized revenue.” That is a spread of 3x on the single figure that matters most.
Run the multiples on both ends:
- At $100 million ARR, a $40 billion valuation is 400x revenue.
- At $300 million ARR, it is 133x revenue.
- At $500 million ARR — above any reported figure — it is still 80x.
For context, Cognition’s $47 billion round priced at roughly 52x revenue and was already treated as aggressive. Thinking Machines, on the low revenue estimate, is being asked to clear a bar nearly eight times steeper.
The skeptical read is simple: when two credible outlets covering the same round cannot land within 3x of each other on ARR, the number is not audited, not disclosed, and not the thing being priced. What is being priced is Murati, the Nvidia relationship, and option value on a lab that has not yet shipped a closed frontier model.
Why is Nvidia in this round at all?
Because Nvidia is on both sides of the ledger. It is an equity holder from the 2025 seed, and since March 11, 2026 it is the compute supplier under a multiyear partnership that StrictlyVC reported as “worth tens of billions of dollars,” covering roughly one gigawatt of Vera Rubin capacity.
That structure is now standard. Nvidia invests, the startup commits the money to Nvidia silicon, and the revenue books at Nvidia. We have seen the same shape in the $6 billion Poolside deal and in Lambda’s $35 billion Anthropic contract.
It is good business for Nvidia and it is circular. A $1 billion equity check that returns as a multi-billion-dollar compute commitment is not an independent validation of the $40 billion price. It is the supplier funding its own order book.
Who has left, and does it matter?
Several of the people the $12 billion valuation was underwritten on are gone. Andrew Tulloch was recruited by Meta Superintelligence Labs in October 2025. In January 2026, more co-founders followed: TechCrunch reports Lilian Weng and Luke Metz returned to OpenAI, and Barret Zoph left for Google.
Thinking Machines was founded in February 2025 around roughly 30 researchers pulled from OpenAI, Meta AI and Mistral. Headcount reached about 100 in 2026. The seed thesis was the roster.
Investors are now paying 3.3x more for a thinner one. That is defensible if the product is the asset rather than the people — but the company’s pitch in 2025 was explicitly the people.
Why this matters for the wider AI market
This round is a live test of whether pre-revenue-multiple pricing survived into late 2026. The market has stayed open — a16z closed a $1.1 billion Machine Age fund in August, and Etched doubled to $21 billion in under a month. Capital is not the constraint.
Price discipline is. Three signals in this deal argue the mood has shifted:
- The number came down. $60 billion discussed, $50 billion sought, ~$40 billion reportedly on the table. Down rounds against ambition are still down rounds against ambition.
- The lead is an insider. Accel already owns a position. Insider-led rounds price differently than competitive outside bids.
- The raise is smaller. $1 billion, versus $2 billion at the seed — for a company that now has a gigawatt-scale compute bill.
For public-market investors, the read-through is narrower than the headline suggests. The clearest beneficiary of a closed round is Nvidia, which converts private AI equity into shipped silicon regardless of whether Thinking Machines ever justifies $40 billion.
This post is reporting and analysis, not financial advice.
Frequently asked questions
Is the Thinking Machines valuation confirmed?
No. The Information reported talks on September 3, 2026, and TechCrunch and PYMNTS relayed them. No round has been announced by the company. Terms in talks change.
How much has Thinking Machines raised in total?
$2 billion in the July 2025 seed at a $12 billion valuation, plus an undisclosed Nvidia investment in March 2026. The current $1 billion would bring disclosed equity to roughly $3 billion.
What is Tinker?
An API launched October 1, 2025 that lets customers fine-tune language models on Thinking Machines’ infrastructure. Users submit jobs; the company runs them and charges usage-based compute fees.
What is Inkling?
Thinking Machines’ first open-weights model, released July 15, 2026 — 975B total parameters with 41B active, Mixture-of-Experts, multimodal, up to a 1M-token context window.
Why would a company take a lower valuation than it asked for?
Compute bills do not wait. A gigawatt-scale Nvidia commitment requires cash now, and a priced round at $40 billion beats an unpriced one at $60 billion.
Who is leading the round?
Accel is in talks to lead, per The Information. Nvidia, an existing investor, is considering participating. Neither is confirmed.
How does this compare to OpenAI?
OpenAI closed $122 billion at an $852 billion valuation in March 2026, per CNBC — the largest funding round on record. Thinking Machines is raising roughly 1/122nd of that at about 1/21st the valuation.
The bottom line
The Thinking Machines valuation at $40 billion is a markup and a markdown at the same time. Up 3.3x from the last priced round; down a third from what the company was asking ten months ago. That is what a cooling-but-open market looks like.
Watch three things next. Whether Accel actually signs. Whether the company discloses a revenue figure that closes the 3x gap between the two published estimates. And whether anyone outside the existing cap table bids.
If the round closes at $40 billion with Accel and Nvidia and no new outside lead, the price is a negotiation between people who already own the asset — not a market clearing one.
Sources
- TechCrunch — Accel reportedly in talks to lead $1B round for Thinking Machines at $40B valuation
- The Information — Thinking Machines Lab In Talks to Raise Billions at Roughly $40 Billion Valuation
- PYMNTS — Thinking Machines Lab Seeks $1 Billion at $40 Billion Valuation
- Thinking Machines Lab — Inkling: Our Open-Weights Model
- StrictlyVC — Thinking Machines Lands a Huge Deal with Nvidia
- CNBC — OpenAI closes record-breaking $122 billion funding round
Leave a Reply