Thinking Machines Lab funding: Every round, investor, and valuation (2026)

Thinking Machines Lab is the AI company Mira Murati founded in February 2025, five months after she left her job as chief technology officer of OpenAI. It builds open-weight AI models and Tinker, a service that lets researchers and companies fine-tune those models on their own data without running the training hardware themselves.

The company has raised about $2 billion across two rounds, all of it equity. The first was a seed round in July 2025 that valued the company at $12 billion. The second was a strategic investment from Nvidia in March 2026, whose size was never disclosed.

A third round is in talks. Accel is reported to be leading it at a valuation of at least $40 billion, though that is below the $50 billion the company sought in late 2025. Let's dive into Thinking Machines Lab's funding history.

Thinking Machines Lab raised its seed round on the strength of its team. When investors committed, the company had no product, no revenue, and a public description of its plans that ran to a few paragraphs. Every round since has been shaped by what that team built, and by who left it.

Murati announced Thinking Machines Lab in February 2025 with about 30 researchers and engineers, most of them from OpenAI, Meta, and Mistral. The group included John Schulman, an OpenAI co-founder, and Barret Zoph, who had run post-training at OpenAI. That roster is what investors were paying for.

Reuters reported in April 2025 that Andreessen Horowitz was in talks to lead a round at a $10 billion valuation. By the time the deal closed in July, the price had moved up. The company confirmed about $2 billion at a $12 billion valuation, and Bloomberg had earlier put the pre-money figure at $10 billion.

Andreessen Horowitz led the round. Nvidia, Accel, ServiceNow, Cisco, AMD, and Jane Street joined it, according to the company. Albania's government also said in May 2025 that it would invest $10 million, which needed a change to the country's budget. Murati was born and raised in Albania.

The seed came with unusual terms. The Information reported that Murati's vote on the board equals the votes of all other directors combined, plus one, and that founding shareholders hold shares with 100 times the voting power of ordinary stock. The company did not comment on the report.

Tinker launched in October 2025 as the company's first product. The same month, co-founder Andrew Tulloch left to join Meta, which had spent the summer offering large pay packages to researchers across the industry.

In November 2025, Bloomberg reported that Thinking Machines Lab was in early talks to raise at about $50 billion, with some people involved putting the figure at $55 billion to $60 billion. That would have been more than four times the seed price, four months after it closed. The round never came together.

In January 2026, the company fired CTO Barret Zoph. Zoph and co-founder Luke Metz rejoined OpenAI, along with researcher Sam Schoenholz. Soumith Chintala, the co-creator of PyTorch who had joined from Meta, became CTO.

The next money came from a supplier. In March 2026, Nvidia announced a multiyear partnership under which Thinking Machines Lab will deploy at least one gigawatt of Nvidia's next-generation Vera Rubin systems, starting in early 2027. Nvidia said it had also made a "significant investment" in the company. Neither side disclosed the amount or the valuation.

In September 2026, The Information reported that Accel, an existing investor, was in talks to lead a new round at a pre-money valuation of at least $40 billion. The first report put the round at $1 billion or more. A later report put it at $5 billion to $6 billion, with Nvidia contributing about $2.5 billion. As of early October 2026, no close had been announced.

Thinking Machines Lab has only one confirmed valuation. Everything after the seed is either undisclosed or still in negotiation, so the history is mostly a record of what investors have been willing to discuss.

The April talks and the July close both put the company at about $10 billion before the new money, so the $12 billion figure is that price plus the cash raised. The failed November talks would have more than quadrupled the seed valuation without a revenue figure to point to.

The September talks set the bar lower. A $40 billion pre-money price is about 3.3 times the seed valuation, and it comes after the company had shipped two products and after reports of its first meaningful revenue. It also comes after the departures of several co-founders, which investors had to price in.

The comparison with Murati's former employer is stark. OpenAI's $122 billion round valued it at $852 billion in March 2026, and Anthropic's funding rounds took it close to $1 trillion two months later. Thinking Machines Lab is a much smaller bet, but it is priced on the same assumption: that a small team of top researchers can build a frontier lab.

Exact stakes are private, but Murati's voting rights mean she controls the company regardless of how much equity the investors hold.

Thinking Machines Lab's investors fall into three groups: a lead venture firm, chipmakers and cloud vendors that sell it hardware or software, and a trading firm and a government that took small, strategic positions.

Andreessen Horowitz led the seed and is the largest disclosed financial backer. The firm was in talks to lead as early as April 2025, before the company had published any research. TNW reported in September 2026 that Andreessen Horowitz was briefing its own limited partners on the new round, which suggests it plans to keep its position.

Nvidia invested twice: in the seed and again in March 2026. Its second check came bundled with a deal to supply at least one gigawatt of computing power, which makes Nvidia both an owner and a supplier. Analysts quoted by Reuters described this kind of arrangement as a circular flow of capital, since some of the money comes back as chip orders that feed Nvidia's data center business.

Nvidia is reported to be discussing about $2.5 billion more in the round now in talks. If it closes, Nvidia would likely become the largest outside investor. AMD, Nvidia's main rival in AI chips, also joined the seed, which gives Thinking Machines Lab a relationship with both of the companies that make the GPUs it trains on.

Accel joined the seed and is now in talks to lead the next round. A lead role usually means setting the price and the terms, so Accel's move from participant to lead is a sign of its conviction after the earlier $50 billion talks fell apart.

Cisco and ServiceNow are enterprise software and networking companies that sell to the same large customers who might one day buy customized AI models. Both invested in the seed. Their stakes give them early access to a lab whose pitch centers on models that companies can adapt to their own work.

Jane Street, the quantitative trading firm, is the odd name on the list. Trading firms run large computing operations of their own and are natural users of custom models. Bridgewater Associates, another finance firm, became one of Tinker's first named customers.

Albania's $10 million is the most unusual check on the list. Albanian state broadcaster RTSH reported the government's decision in May 2025, and it is one of the few cases of a national government buying into a private AI lab at the seed stage.

TechCrunch's September 2026 coverage also listed GV, Lightspeed, and Conviction Partners as backers. The company's own seed announcement did not name them, so their role is unconfirmed.

Thinking Machines Lab spends most of its capital on two things: computing power to train and run models, and the researchers who build them. Its compute commitments now far exceed the money it has raised.

Compute is the largest line by a wide margin. The Nvidia agreement covers at least one gigawatt of Vera Rubin systems, and industry executives quoted by Reuters put the cost of one gigawatt of AI computing at around $50 billion. That is an industry estimate, not the contract value, but it shows why the company needs to keep raising.

In April 2026, the company signed its first cloud deal, with Google Cloud. TechCrunch's source valued it in the single-digit billions. It gives Thinking Machines Lab access to Nvidia GB300 chips on Google's cloud platform, and it is not exclusive.

In September 2026, the company added a $65 million-a-year inference deal with Crusoe to serve its models to customers. Crusoe runs the workloads on dedicated Nvidia systems, and the contract adds Thinking Machines Lab to a customer list that has helped Crusoe's own funding rounds.

Talent is the second cost and the biggest risk. Posted salaries for software engineers run from $350,000 to $475,000, and headcount passed 150 by May 2026, according to Business Insider.

Holding on to that team has been hard. Business Insider counted 13 departures from the 42-person founding team by May 2026, seven of them to Meta, five to OpenAI, and one to xAI. Co-founder Lilian Weng stepped down in July 2026 and later rejoined OpenAI. Meta's ownership structure, with Mark Zuckerberg in control, let it make the large offers that pulled several researchers away.

The rest goes into products. Tinker trains small add-on layers for open-weight models rather than retraining the whole model, which keeps the cost down for customers. It was free at launch, moved to usage-based pricing in November 2025, and became generally available in December 2025.

In July 2026, the company released Inkling, its first general-purpose model. Inkling has 975 billion parameters, handles text, images, and audio, and is free to download under an open license. Reuters described it as competitive, especially on agent tasks, though other models kept the overall lead.

The business model links the two products. Inkling is free, but customers pay Tinker usage fees to adapt and run it. That puts Thinking Machines Lab in competition with fine-tuning platforms such as Together AI, whose Together AI funding history shows a similar bet on open models. TechCrunch reported in September 2026 that the company's annual revenue run rate had passed $100 million.

Thinking Machines Lab's growth focus is enterprises and research groups that want to customize open models rather than rent closed ones. TNW described it as one of the few US labs offering an open-weight alternative to Chinese models such as DeepSeek and Kimi. In Europe, it competes with a lab that took a similar path, as Mistral's ownership and funding shows.

The product roadmap points toward real-time AI. In May 2026, the company previewed "interaction models," which handle several kinds of input in chunks of about 200 milliseconds so a person and the model can work together live, with a wider launch promised later in 2026. The Nvidia systems arriving in 2027 are meant for training and serving its next models.

The company has no reported IPO plans, and its next step is closing the round now in talks. The main risks are talent, since departures to rivals such as xAI's well-funded lab continue, and price, since a $40 billion valuation could sit at as much as 400 times the reported run rate. The new capital will mostly go to compute that has already been ordered, so the round's size will decide how much of that capacity Thinking Machines Lab can pay for itself.

Thinking Machines Lab has raised about $2 billion in disclosed funding, all of it equity. Nearly all of it came from the seed round in July 2025. Nvidia made a further investment in March 2026, but its size was not disclosed. A new round of $1 billion to $6 billion is in talks.

Thinking Machines Lab's lead investor is Andreessen Horowitz. Other backers include Nvidia, Accel, AMD, Cisco, ServiceNow, Jane Street, and the government of Albania. Accel is in talks to lead the next round, with Nvidia expected to take part.

Thinking Machines Lab was valued at $12 billion after its July 2025 seed round. It is in talks to raise at a pre-money valuation of at least $40 billion. Earlier talks at about $50 billion in late 2025 did not close.

No. Thinking Machines Lab is private and has not filed for an IPO. Its shares are not traded on any stock exchange.

Thinking Machines Lab is owned by its founders, employees, and investors. Founder and CEO Mira Murati controls the company through supervoting founder shares and a board vote that outweighs all other directors combined. Andreessen Horowitz and Nvidia are likely the largest outside shareholders, though exact stakes are not public.