Three weeks ago, this publication reported that Meta had gutted Mira Murati's Thinking Machines Lab, hiring seven founding members after she rejected a roughly $1 billion acquisition offer. Andrew Tulloch's package alone was reportedly worth $1.5 billion over six years. It looked like the beginning of the end for the startup that was supposed to challenge OpenAI.

It was not. On Tuesday, Thinking Machines Lab signed a multi-billion-dollar agreement with Google Cloud to access AI infrastructure powered by Nvidia's latest GB300 chips. The deal, first reported by TechCrunch, is valued in the single-digit billions and marks the startup's first formal partnership with a cloud services provider.

What the Deal Includes

Thinking Machines gets access to Google's latest AI systems built on Nvidia's GB300 NVL72 systems, which offer a claimed 2X improvement in training and serving speed compared to prior-generation GPUs. The deal also includes infrastructure services to support model training and deployment, plus access to Google Cloud's storage, Kubernetes engine, and Spanner database product.

"Google Cloud got us running at record speed with the reliability we demand," said Myle Ott, a founding researcher at Thinking Machines, in a statement. The deal is not exclusive, leaving Murati free to work with multiple cloud providers. But it signals that Google is aggressively locking in frontier labs early in their compute lifecycle.

The Strategic Context

Google Cloud has been on a signing spree. Earlier this month, Anthropic expanded its compute partnership with Google and Broadcom for multiple gigawatts of TPU capacity. Anthropic also signed a separate agreement with Amazon for up to 5 gigawatts. The hyperscaler war for AI lab contracts is now the most expensive land grab in cloud computing history.

For Thinking Machines, the timing is everything. Murati left OpenAI in September 2024 and founded the company in February 2025. By July, it had raised a $2 billion seed round at a $12 billion valuation led by Andreessen Horowitz. The company launched its first product, Tinker, in October. It automates the creation of custom frontier AI models using reinforcement learning, a training approach that has underpinned recent breakthroughs at DeepMind and OpenAI.

The scale of the Google Cloud deal reflects how computationally expensive that reinforcement learning work gets. Google's press release specifically noted support for Thinking Machines' reinforcement learning workloads.

The Meta Shadow

The deal arrives against a backdrop of aggressive talent raiding. Meta hired seven founding members, including Joshua Gross, who built Tinker. Other departures scattered across the industry: Barret Zoph and Luke Metz went back to OpenAI, Karan Mohan Chaplot went to xAI. Business Insider reported this week that Meta quietly hired two more founding members, making it the latest in a string of departures from the high-profile startup.

Murati had previously partnered with Nvidia, which invested during the seed round. But the Google deal is qualitatively different. It is not an equity investment. It is a commercial partnership for production-grade infrastructure. It says: this company is building something real, at a scale that requires billions in compute, and Google is betting that hosting it will be worth the deal.

What It Means

The AI lab landscape has a clear pattern now. Labs raise billions, sign multi-billion compute deals, lose talent to bigger companies, and keep building anyway. Thinking Machines is following the same script as Anthropic, which lost half its founding team to various companies over the past two years and now generates $30 billion in annualized revenue.

The question is not whether Murati can survive losing seven founding members. The question is whether what she is building with the remaining team, plus Google's GB300 infrastructure, plus Nvidia's investment, is enough to justify a $12 billion valuation for a company that still has not publicly revealed what its frontier models can do.

First reported by TechCrunch (Rebecca Bellan). Additional reporting from Business Insider, Latestly, NewsBytesApp.