Mira Murati's Thinking Machines Lab Signs Multi-Billion-Dollar Compute Deal With Google Cloud

Mira Murati’s startup Thinking Machines Lab has signed a new multi-billion-dollar compute agreement with Google Cloud, announced April 22 at Google Cloud Next 2026 in Las Vegas. The deal, valued in the single-digit billions, is one of the largest infrastructure commitments Google Cloud has made to an AI startup outside its own research divisions.

The agreement gives Thinking Machines access to Google’s new A4X Max virtual machines, powered by Nvidia’s GB300 chips. Google claims these VMs deliver a 2x throughput improvement over the previous generation — a meaningful edge for a company whose core product, Tinker, automates the creation of custom frontier AI models. The partnership is non-exclusive, so Thinking Machines can continue training on other cloud providers’ hardware.

This is an expansion of a relationship that started in 2025, shortly after Murati founded the company following her departure from OpenAI. Thinking Machines raised a $2 billion seed round at a $12 billion valuation earlier this year — an extraordinary early-stage number that reflects both Murati’s reputation and the market’s appetite for credible alternatives to the dominant AI labs.

The timing matters. Announced on the same stage as Google’s broader Cloud Next keynote, the deal signals Google’s intent to use cloud infrastructure partnerships as a competitive weapon against AWS and Azure. Amazon and Microsoft have both signed similar anchor deals with AI startups — Amazon with Anthropic, Microsoft with OpenAI — and Google is making the same play with a shorter list of high-conviction bets. Thinking Machines Lab joins Mistral and Cohere on Google Cloud’s roster of strategic AI partners.

What Thinking Machines actually ships at scale remains the open question. Tinker — its flagship product for building bespoke frontier models — has not yet had a broad public release. But the Google deal buys time and compute. With access to GB300 clusters and a multi-billion-dollar runway, Murati’s team can train larger models than most research labs can afford.

For Google Cloud, the optics matter as much as the revenue. Landing a company led by the former CTO of OpenAI — someone with direct knowledge of how GPT-4 and o1 were built — is a credibility signal. It tells other AI startups that Google’s infrastructure is capable enough for the most demanding workloads, and that Google is willing to write checks to prove it.

The deal does not include an equity stake by either party, and both companies declined to disclose the exact financial terms. A Thinking Machines spokesperson confirmed the agreement is “long-term” but would not specify a duration.

For developers, the practical takeaway is that Tinker’s model-building platform will have better compute access going forward. If Thinking Machines delivers on its premise of democratizing frontier model creation, the Google Cloud partnership is what makes that possible at scale.