Google backs Thinking Machines

Google Cloud has signed a new multibillion-dollar agreement to supply Thinking Machines Lab with the computing power to build and run its artificial intelligence models. (techcrunch.com) The deal was announced April 22 at Google Cloud Next in Las Vegas. Google said Thinking Machines will expand its use of Google Cloud’s AI Hypercomputer, including A4X Max virtual machines built on Nvidia GB300 systems. (googlecloudpresscorner.com) Thinking Machines Lab is the startup founded by former OpenAI chief technology officer Mira Murati. The company says it is building artificial intelligence research and products aimed at giving people and organizations more useful control over AI tools. (thinkingmachines.ai) Cloud computing in this case means renting giant clusters of chips in remote data centers instead of buying and operating them yourself. Training a frontier model can require tens of thousands of graphics processors linked together, so access to cloud capacity has become as important as the model design itself. (cloud.google.com, techcrunch.com) Google said Thinking Machines will be among the first Google Cloud customers to use Nvidia GB300 NVL72 systems through its cloud service. In early testing, Google said the startup saw training and serving speeds double versus prior-generation graphics processors. (googlecloudpresscorner.com, morningstar.com) The package goes beyond chips. Google said the agreement also includes Google Kubernetes Engine, Spanner, Cloud Storage, Cluster Director and Anywhere Cache, the software and storage layers needed to keep large model training jobs moving. (googlecloudpresscorner.com) Google is making a broader push to sell AI infrastructure to outside customers as demand rises. Chief executive Sundar Pichai said this week that more than half of Google’s machine-learning compute investment in 2026 is expected to go to the cloud business, and that direct customer use of Google’s first-party models now exceeds 16 billion tokens per minute. (blog.google) That puts this agreement in the middle of a three-way race between cloud providers, chip suppliers and model builders. Nvidia supplies the hardware, Google rents the systems and networking, and startups like Thinking Machines use both to compete with OpenAI, Anthropic and other large-model developers. (techcrunch.com, blog.google, thinkingmachines.ai) The immediate result is simple: one of the newest AI labs has locked in a large supply of scarce computing capacity. In 2026, that may be as decisive as hiring researchers or publishing new model results. (techcrunch.com, blog.google)