Google signed a multibillion-dollar deal with Mira Murati's Thinking Machines Lab on 22 April, its third major frontier AI compute partnership this month alongside Anthropic and Meta.
Fourteen months after Mira Murati left OpenAI with no product and no office, her startup Thinking Machines Lab has locked in a multibillion-dollar compute contract with Google Cloud — a deal in the single-digit billions, according to TechCrunch's 22 April exclusive — making it the third frontier AI developer to secure access to Google's Nvidia GB300 capacity within a single month.
mira murati · thinking machines lab · google cloud
The speed of the agreement illustrates how dramatically the AI infrastructure market has shifted since 2023. Thinking Machines Lab, incorporated in February 2025, had not yet shipped a product when it raised a $2 billion seed round at a $12 billion valuation. Its first offering, Tinker — a platform that automates the creation of custom frontier AI models using reinforcement learning — launched in October 2025. The Google deal gives Thinking Machines access to compute capacity for training and deploying those models at scale, alongside infrastructure services tied to Google's latest AI systems. Google noted that its platform can handle the reinforcement learning workloads that Tinker's architecture depends on — a capability that neither AWS nor Microsoft Azure has been advertising with the same specificity.
The numbers that frame the deal reveal how concentrated frontier AI infrastructure has become. TechCrunch placed the contract in the "single-digit billions," a range spanning $1 billion to $9 billion depending on delivery timeline and utilisation commitments. Nvidia's GB300 chips, which underpin Google's latest compute cluster, currently command a 60-day lead time for independent purchasers at list prices of approximately $40,000 per unit (Nvidia quarterly earnings, February 2026). Thinking Machines has also separately committed to one gigawatt of Nvidia compute directly, per Axios reporting from March 2026, giving it one of the largest independent compute reservations of any pre-revenue startup in history. Murati's company raised $2 billion at a $12 billion valuation — a ratio that would have been almost inexplicable in 2022 but has become a marker of how compressed the timeline from founding to frontier is in 2026, with OpenAI now at $25 billion and Anthropic approaching $19 billion in annualised revenue (The Information, April 2026).
Continue reading to see the full article
“TechCrunch placed the contract in the "single-digit billions," a range spanning $1 billion to $9 billion depending on delivery timeline and utilisation commitments.”
The deal makes Thinking Machines the third frontier AI developer in April to agree terms with Google Cloud, behind Anthropic — which signed for up to 5 gigawatts of capacity for Claude training and deployment — and Meta, which has been building GPU reserves across multiple cloud providers. That clustering matters. Google is running a $75 billion capital expenditure programme in 2026 (Alphabet earnings, Q1 2026) and needs enterprise customers to justify and amortise the build-out. Signing compute agreements with Anthropic, Meta, and Thinking Machines in the same month establishes Google Cloud as a neutral provider to all frontier labs — a commercially and strategically significant position as Microsoft remains deeply aligned with OpenAI and Amazon has prioritised its own Bedrock infrastructure for AI workloads.
Key Takeaways
→mira murati: Thinking Machines Lab is an AI startup founded in February 2025 by Mira Murati, former CTO of OpenAI.
→thinking machines lab: Thinking Machines Lab is an AI startup founded in February 2025 by Mira Murati, former CTO of OpenAI.
→google cloud: Thinking Machines Lab is an AI startup founded in February 2025 by Mira Murati, former CTO of OpenAI.
→ai infrastructure: Thinking Machines Lab is an AI startup founded in February 2025 by Mira Murati, former CTO of OpenAI.
Google benefits most directly by locking in Thinking Machines as a long-term customer and demonstrating that its custom AI infrastructure can attract customers away from AWS and Microsoft Azure on performance and availability grounds alone. Murati gets compute without equity dilution, and the deal likely delays any near-term pressure to generate public revenue. For Nvidia, every enterprise deal that routes through GB300 hardware is a direct revenue event regardless of which cloud provider wraps the silicon. The party with the most ambiguous position is OpenAI itself: Thinking Machines Lab was founded largely by people who left OpenAI, and a well-capitalised competitor with Google's infrastructure behind it is a different kind of rival than a scrappy new entrant.
mira murati · thinking machines lab · google cloud
The risk the deal does not advertise is governance. Murati departed OpenAI in October 2024 amid board-level disagreements over the organisation's for-profit transition. Thinking Machines was incorporated as a for-profit Delaware C-corporation with no published AI safety commitments or external oversight structure — a model that several AI safety researchers have described as governance-light at a moment of increasing regulatory scrutiny. The EU AI Act, in full enforcement since August 2025, places Thinking Machines' Tinker platform in the high-risk category due to its custom model generation capabilities. Compliance certification under the Act is required by October 2026. If the company misses that deadline, it faces fines of up to 3 percent of global annual turnover — a meaningful sum at any valuation above $10 billion.
Advertisement
The next signal to watch is Tinker's first disclosed enterprise client, expected by mid-May according to sources familiar with the company's plans cited by TechCrunch. That disclosure will test whether Murati's approach — build fast, secure infrastructure, then monetise — can generate the revenue multiples that a $12 billion seed valuation implies. If it can, the Google deal will look prescient. If it cannot, the implied valuation faces its first serious public stress test, and the question of whether the frontier AI market can support a third non-incumbent player with serious compute access will move from theoretical to urgent.
What is Thinking Machines Lab and what does it build?
Thinking Machines Lab is an AI startup founded in February 2025 by Mira Murati, former CTO of OpenAI. Its product, Tinker — launched October 2025 — automates the creation of custom frontier AI models using reinforcement learning. The company raised $2 billion at a $12 billion valuation before generating public revenue.
How large is the Google-Thinking Machines deal?
TechCrunch reported the contract as in the "single-digit billions," placing it between $1 billion and $9 billion. The deal gives Thinking Machines access to Google Cloud's Nvidia GB300-based AI compute infrastructure for model training and deployment, alongside Google's proprietary AI systems.
Why is Google signing deals with multiple AI startups simultaneously?
Google's $75 billion capex programme in 2026 requires enterprise customers to justify the infrastructure investment. Signing compute deals with Anthropic, Meta, and Thinking Machines in the same month positions Google Cloud as a neutral frontier AI provider — valuable as Microsoft remains aligned with OpenAI and Amazon's GPU availability has been constrained by Bedrock priorities.
What are the regulatory risks for Thinking Machines Lab?
The EU AI Act, in full enforcement since August 2025, classifies Tinker in the high-risk category due to its custom model generation capabilities. Compliance certification is required by October 2026. The company also has no published AI safety commitments or external oversight structure, a governance gap that regulators in both the EU and US have flagged.
