Mira Murati’s Thinking Machines Lab closed the largest seed round in AI history in July 2025, raising $2 billion at a $12 billion post-money valuation, according to Reuters and TechCrunch. Four months later, Bloomberg reported the company was seeking $50 billion from new investors. By January 2026, those talks had collapsed without a deal. Today Thinking Machines sits at its original $12 billion valuation, with roughly 140 to 169 employees, two live products, and infrastructure commitments from Nvidia and Google that run into the billions.
Mira Murati’s Thinking Machines: $2B Raised, $50B Stalled, Shipping
Thinking Machines Lab raised $2 billion at a $12 billion seed valuation in July 2025, then sought $50 billion from investors four months later before backers passed. Here is the full financial arc: valuation milestones, Nvidia and Google infrastructure deals, and two shipped products through June 2026.
7 min read
Related startups
Two Products in 16 Months
The finetuning API was Thinking Machines Lab’s first commercial offering, launched alongside Tinker, a developer platform for running and customizing open-source models. The approach positioned the lab as infrastructure rather than a consumer-facing assistant provider. In September 2025, TechCrunch reported Thinking Machines had published research arguing that current AI models produce inconsistent outputs under similar inputs, and proposing training techniques to reduce that variance.
The more consequential release came in May 2026. The lab introduced what it calls “interaction models,” a category it defines as AI that can process input and generate a response simultaneously, so that the system can be interrupted mid-response, as a person can be cut off in conversation. The first model, TML-Interaction-Small, responds in 0.40 seconds, which Thinking Machines describes as roughly matching the latency of natural human conversation and as significantly faster than comparable models from OpenAI and Google. Alongside the model, the lab published a technical paper arguing that real-time interactivity should be a native architectural property, not an afterthought layered over a turn-based language model, per our earlier coverage.
The paper-before-product sequencing mirrors how Anthropic introduced Constitutional AI and how DeepMind published AlphaGo research ahead of its commercial applications: establish the conceptual frame before competitors can define the conversation in their own terms. Semafor previewed the interaction models in May 2026, noting the architecture is a single network in which the ability to listen, speak, see, and pause is trained in from the start rather than bolted on.
Nvidia, Google, and the Infrastructure Strategy
Thinking Machines Lab’s compute strategy is built on partnerships rather than owned hardware. In March 2026, Nvidia announced it would invest in the lab and supply chips amounting to at least one gigawatt of Vera Rubin AI accelerator capacity, per Bloomberg. Vera Rubin is Nvidia’s successor to the Blackwell architecture. Allocating a gigawatt of it to a 16-month-old lab reflects Nvidia’s assessment that Thinking Machines is building models that require frontier-grade training at scale. See our earlier coverage of the Nvidia pact for full deal terms.
In April 2026, TechCrunch reported that Google had deepened its partnership with Thinking Machines through a new agreement valued in the single-digit billions. The deal provides access to Google Cloud infrastructure and to Nvidia GB300 chips hosted on Google’s platforms, giving Thinking Machines access to Google’s full AI computing stack without the capital expenditure of self-hosting.
The same month, TechCrunch reported that Meta had attempted to poach talent from Thinking Machines and that the lab had retained most of its key researchers. At a headcount of roughly 140 to 169 employees as of April 2026, retention is a meaningful signal: the team is small enough that losing two or three senior researchers would be materially disruptive.
What It Means
Thinking Machines Lab enters the second half of 2026 with $2 billion in capital, two shipped products, a published technical framework for interaction models, and infrastructure agreements with the two largest AI cloud providers. The $50 billion valuation episode is the most visible entry in the company’s public record, but the more durable data point is that Murati maintained the original seed valuation, kept her team intact through active poaching attempts, and delivered a product roadmap that moved from developer tooling to a novel model category in under 16 months. Whether the interaction-models paradigm establishes itself as a distinct product category or becomes a capability that larger companies absorb into existing offerings is the question the current funding level and headcount will need to answer.
Sources
TechCrunch: Thinking Machines Lab is ex-OpenAI CTO Mira Murati’s new startup (Feb 2025)
Bloomberg: Murati’s Thinking Machines in Funding Talks at $50 Billion Value (Nov 2025)
The Information: Thinking Machines Lab Launches First Finetuning Product (Oct 2025)
Semafor: Mira Murati’s Thinking Machines previews “interaction models” (May 2026)
CNBC: OpenAI CTO Mira Murati announces she’s leaving the company (Sep 2024)
Editorial standards: every claim is sourced. Tips: [email protected]
© 2026 StartupHub.ai. All rights reserved. Do not enter, scrape, copy, reproduce, or republish this article in whole or in part. Use as input to AI training, fine-tuning, retrieval-augmented generation, or any machine-learning system is prohibited without written license. Substantially-similar derivative works will be pursued to the fullest extent of applicable copyright, database, and computer-misuse laws. See our terms.
