Mira Murati's Thinking Machines Lab Signs Single-Digit Billion Google Cloud Deal — as Meta Poaches 7 Founding Members
Mira Murati’s Thinking Machines Lab struck its first agreement with a major cloud provider on Wednesday, signing a multi-billion-dollar deal with Google Cloud for infrastructure built around Nvidia’s latest GB300 NVL72 chips. Google announced the deal at Cloud Next in Las Vegas. A source familiar with the terms told TechCrunch the value lands in the single-digit billions.
The lab is among the first Google Cloud customers to access GB300 NVL72 systems, which Google says deliver a 2x improvement in training and serving speed over prior-generation Blackwell GPUs. Thinking Machines will run workloads through Google’s AI Hypercomputer stack — A4X Max virtual machines, Jupiter high-bandwidth networking, Kubernetes Engine, Spanner, and Cluster Director for automated remediation.
Why GB300 Matters for Tinker
Thinking Machines’ first product, Tinker, is a platform for building custom frontier AI models using reinforcement learning. RL workloads are among the most network-intensive computations in model development: weight updates traverse thousands of GPUs thousands of times per training run. Google’s Jupiter fabric, which ties together the GB300 NVL72 rack, is specifically designed for that pattern.
The Google deal fills a near-term gap. Thinking Machines signed a separate multi-year partnership with Nvidia in March 2026 for at least one gigawatt of Vera Rubin systems, but that capacity is not expected to come online until early 2027. GB300 access covers the gap while the next-generation silicon is being built out.
Meta Poaching Complicates the Picture
The compute deal landed alongside a more difficult development: Meta has poached seven founding members of Thinking Machines Lab, including the lead engineer behind Tinker. The scale of the talent departure — seven people from what was already a small founding team — raises questions about internal cohesion at a company that has been highly secretive since its February 2025 founding.
Murati has not commented publicly on the departures. The Google Cloud deal may partly serve as a visible signal that the lab’s infrastructure roadmap remains intact despite the team changes.
The Frontier Lab Hyperscaler Race
The deal adds Thinking Machines to a list of frontier AI customers locked into Google’s compute supply chain. Anthropic signed a multi-gigawatt TPU agreement with Google and Broadcom earlier this month. Meta is running its own multibillion-dollar Google Cloud contract. Google Cloud’s revenue backlog more than doubled last year to $240 billion, and the company disclosed at Cloud Next that its models now process 16 billion tokens per minute via direct API use by customers — up from 10 billion last quarter.
The Thinking Machines agreement is not exclusive. The lab retains the option to use multiple cloud providers simultaneously, which matters as it seeks to preserve bargaining leverage in a market where hyperscalers are competing aggressively to lock in frontier AI customers early.
Thinking Machines raised a $2 billion seed round at a $12 billion valuation led by Andreessen Horowitz in early 2025. No subsequent public funding round has been announced.