Thinking Machines Lab company profile: valuation, funding & revenue

This Thinking Machines Lab company profile covers the company's valuation, revenue, funding rounds, investors and history. Thinking Machines Lab works in artificial intelligence, with a focus on frontier AI research and models. It was founded in 2025 and is headquartered in San Francisco, California, United States. Its latest valuation on record is About US$12 billion (July 2025). Every figure links to a public source, and anything the company hasn't disclosed is marked as such.

Revenue

Funding history

Thinking Machines Lab timeline

Key dates from the sources on this page, oldest first. Metric snapshots are listed after the funding events.

• 2025 — Thinking Machines Lab is founded in San Francisco by Mira Murati (CEO), with John Schulman, Barret Zoph, Lilian Weng and Luke Metz.
• Jul 2025 — Seed: About US$2B, valuing the company at About US$12B. Investors: Andreessen Horowitz (lead); Nvidia, AMD, Cisco, Jane Street and others.
• September 2026 — Reported funding talks: Talks to raise at a valuation of at least US$40 billion; reported annualized revenue run rate above US$100 million.
• March 2026 — Compute partnership: Nvidia investment and a multi-year deal for one gigawatt of Vera Rubin computing capacity.

Investors

Overview

Thinking Machines Lab is an artificial intelligence research and product company founded in San Francisco in February 2025 by Mira Murati, formerly chief technology officer of OpenAI, where she oversaw the development of ChatGPT, DALL-E and GPT-4. The lab's stated aim is to make AI systems more widely understood, customisable and generally capable, with an emphasis on openness: publishing research, releasing open-weight models and building tools that let others adapt models to their needs.

Murati recruited a team of senior researchers, many from OpenAI. John Schulman, an OpenAI co-founder who led its reinforcement learning work, became chief scientist. Barret Zoph, Lilian Weng and Luke Metz were also part of the founding group. The company is organised as a public benefit corporation and moved into new San Francisco offices in the Mission District in 2026.

Products

• Tinker: launched in October 2025, a service that lets researchers and developers fine-tune open-weight language models without managing the underlying computing infrastructure. Users write the training logic; Tinker handles the distributed training.
• Inkling: an open-weight large language model with roughly 975 billion parameters, released in July 2026 under the permissive Apache licence.
• Inkling Small: a smaller 276-billion-parameter version released weeks later, cheaper to run.

Releasing a very large model under an open licence was a notable choice when most frontier labs keep their best models closed. It places Thinking Machines alongside Meta, Mistral AI and Chinese labs in the open-model camp, while selling services like Tinker around them.

Funding

In July 2025, before releasing any product, Thinking Machines raised about US$2 billion in a seed round at roughly a US$12 billion valuation, led by Andreessen Horowitz with Nvidia, AMD, Cisco and Jane Street among the investors. It was one of the largest seed rounds in history. In March 2026, Nvidia made a further investment alongside a multi-year agreement to supply a gigawatt of computing capacity on its Vera Rubin platform.

Fundraising since has been eventful. Late in 2025 the company reportedly sought a valuation around US$50 billion, but those talks did not produce a round. In September 2026, TechCrunch, citing The Information, reported talks to raise new funding at a valuation of at least US$40 billion, with Accel in discussions to lead; reports differed on the amount. Until a round is confirmed, we keep the seed valuation as the last confirmed figure.

Revenue

A source told TechCrunch in September 2026 that the company's annualised revenue run rate exceeded US$100 million, most likely from Tinker and enterprise work. The company hasn't published revenue itself, so we show it as not disclosed. Even taking that figure at face value, a US$40 billion valuation would be several hundred times run-rate revenue, which shows how much of the value rests on expectations.

Talent: gains and losses

Because a frontier lab's main asset is its people, departures matter. In late 2025, co-founder Andrew Tulloch left to join Meta's superintelligence group. In January 2026, Barret Zoph and Luke Metz returned to OpenAI. Murati has continued to recruit, and John Schulman remains chief scientist. Investors in AI labs watch these moves closely, because a handful of researchers can make the difference between leading and lagging.

Competition

Thinking Machines competes with OpenAI, Anthropic, Google DeepMind, Meta, xAI, Mistral AI, Safe Superintelligence and Chinese labs. For talent, it competes with all of them and with well-funded startups. Its distinctive position is its openness, its focus on customisation, and its founder's reputation.

Key facts

Thinking Machines Lab is a San Francisco AI company founded in February 2025 by Mira Murati, with John Schulman as chief scientist. It raised about US$2 billion at roughly US$12 billion in July 2025, led by Andreessen Horowitz, and later received investment and compute commitments from Nvidia. It launched the Tinker fine-tuning service and the open-weight Inkling models. Reports in 2026 described talks at a valuation of at least US$40 billion.

Why open weights matter for business

When a company can download a model's weights, it can run the model on its own servers, keep sensitive data in-house, fine-tune it on proprietary information and avoid paying per token to an outside provider. That appeals to banks, healthcare companies, governments and any business with strict data rules. The trade-off is that running a large model requires significant computing resources and expertise. Services like Tinker aim to remove part of that burden, making customisation practical for more organisations. If enterprises increasingly prefer models they control, open-weight providers could capture a meaningful share of the market.

Seed rounds that look like growth rounds

A US$2 billion "seed" sounds contradictory, since seed rounds are normally the smallest. In frontier AI, the term mostly signals that it was the company's first institutional round. The size reflects the cost of training models: computing power for a single large training run can cost hundreds of millions of dollars. Investors backing such rounds are betting on the team and the possibility of building one of a handful of leading AI labs. Our guide to pre-revenue valuation methods explains how investors put a price on companies with little revenue.

Questions investors ask about AI labs

• How much computing capacity has the lab secured, and on what terms?
• Can it retain its senior researchers against offers from larger rivals?
• Is there a clear route from research to revenue, through products, APIs or licensing?
• How does its model quality compare on independent benchmarks?

What to look for next

The milestones that would change the picture are a confirmed new funding round, adoption of Inkling by enterprises and developers, Tinker's growth, further hires or departures, and any disclosure of revenue.

Bottom line for readers

Developers and enterprises can evaluate Inkling and Tinker as alternatives to closed APIs, especially where data control matters. Founders can see how a strong team can raise enormous sums before a product exists. Investors should treat the valuation as a bet on a small group of researchers and their ability to compete with far larger labs.

Latest developments (October 2026)

Thinking Machines Lab signed a multi-year partnership with Nvidia in March 2026 for at least a gigawatt of computing capacity, and in July 2026 released Inkling, a multimodal model. In September 2026 it was reported to be discussing a raise of US$5 billion to US$6 billion at a valuation of at least US$40 billion, with Nvidia possibly investing, according to Let’s Data Science. Earlier talks at higher valuations collapsed, so our profile keeps the US$12 billion seed valuation until a round closes. See our new AI labs guide.

Sources

Every figure on this page comes from the public sources below. Where sources disagreed, we used the company's own announcement or a primary news report. Private-company data changes; check the date beside each figure.

1. Wikipedia
2. TechCrunch