Big cloud deal for compute

Thinking Machines Lab, Mira Murati’s artificial intelligence startup, has signed a multibillion-dollar deal to expand its use of Google Cloud for model training. (techcrunch.com) Google announced the agreement on April 22 at Cloud Next in Las Vegas and said it will give Thinking Machines more AI infrastructure capacity on its AI Hypercomputer platform. Google said the setup includes A4X Max virtual machines built on Nvidia GB300 graphics processors. (googlecloudpresscorner.com) TechCrunch reported the contract is worth in the single-digit billions, citing people familiar with the deal. Google’s public announcement did not disclose the price, but it said the new agreement expands Thinking Machines’ footprint on Google Cloud. (techcrunch.com) (googlecloudpresscorner.com) Cloud deals like this are about access to scarce computing power: the chips, networking, and storage needed to train large AI systems. Google said A4X Max uses Nvidia GB300 NVL72 systems and its Jupiter network fabric to link clusters that can scale to tens of thousands of graphics processors. (cloud.google.com) That puts Google deeper into the fight to become the default infrastructure provider for frontier AI labs, a market Microsoft has dominated with OpenAI and Amazon has pursued with Anthropic. Thinking Machines gives Google a fast-growing customer founded by OpenAI’s former chief technology officer. (techcrunch.com) (reuters.com) Murati founded Thinking Machines Lab in February 2025 after leaving OpenAI. In July 2025, the company said it raised about $2 billion at a $12 billion valuation in a round led by Andreessen Horowitz. (thinkingmachines.ai) (reuters.com) Google said Thinking Machines will use more than raw compute in the arrangement, including Google Kubernetes Engine, Spanner, Cluster Director, Cloud Storage, and Anywhere Cache. That means the startup is buying a full operating stack for training models and running services around them, not just renting chips by the hour. (googlecloudpresscorner.com) Google has also used the deal to showcase its newest Nvidia-based systems. Its cloud division said A4X Max is now in production, with 72 Blackwell Ultra graphics processors and 36 Grace central processors in each GB300 NVL72 system, plus double the network bandwidth of the prior A4X generation. (cloud.google.com) The concentration risk is getting clearer at the same time the contracts are getting larger. A small number of labs can afford multibillion-dollar commitments, and a small number of cloud providers can supply the power, cooling, networking, and chip inventory those labs need. (techcrunch.com) (cloud.google.com) For Thinking Machines, the immediate result is simpler: guaranteed access to the kind of compute that is now a gatekeeper for frontier AI work. For Google, it is another bid to lock in the labs that may define the next generation of models. (techcrunch.com) (googlecloudpresscorner.com)