Thinking Machines–Google deal
- A major AI infrastructure agreement was signed between Thinking Machines Lab and Google Cloud to power advanced model workloads. - The deal is described as a multi‑billion‑dollar agreement centred on Nvidia GB300 chips and large cloud capacity. - This underscores continued concentration of premium AI compute and higher costs for platforms that need frontier acceleration (techcrunch.com).
Thinking Machines Lab has signed a new agreement to expand its use of Google Cloud for training and running advanced artificial intelligence models. (techcrunch.com) Google Cloud announced the deal on April 22 at its Cloud Next conference in Las Vegas, saying Thinking Machines will use A4X Max virtual machines with Nvidia GB300 graphics processors and other cloud services. (googlecloudpresscorner.com, googlecloudevents.com) TechCrunch reported the agreement is worth single-digit billions of dollars, citing people familiar with the matter; Google and Thinking Machines did not disclose a dollar amount in the announcement. (techcrunch.com, googlecloudpresscorner.com) The hardware at the center of the deal is the kind used to train large models by spreading work across many chips at once. Google said Thinking Machines will be among the first Google Cloud customers to use Nvidia GB300 NVL72 systems through its A4X Max offering. (cloud.google.com, googlecloudpresscorner.com) Google said A4X Max is built around 72 Blackwell Ultra graphics processors and 36 Grace central processors linked to act as one large computing system. In early testing, Google said Thinking Machines saw training and serving speed increase by 2 times versus prior-generation graphics processors. (cloud.google.com, googlecloudpresscorner.com) Thinking Machines was founded by former OpenAI chief technology officer Mira Murati in 2025, and Reuters reported in July 2025 that the startup raised about $2 billion at a $12 billion valuation in a round led by Andreessen Horowitz. (reuters.com) The company says it is building artificial intelligence research and products, and its website now also describes a long-term gigawatt-scale partnership with Nvidia. That gives the startup ties to both the leading chip supplier and one of the few cloud providers with enough capacity to deliver the newest systems at scale. (thinkingmachines.ai, blogs.nvidia.com) Google is using the deal to show that its cloud unit can win large model-training customers even as it also promotes its own chips and Gemini models. Alphabet chief executive Sundar Pichai said on April 22 that just over half of the company’s machine learning compute investment in 2026 is expected to go to the Cloud business. (blog.google) The agreement also shows how a small group of companies still controls the most sought-after artificial intelligence computing: Nvidia supplies the chips, and hyperscale clouds such as Google decide who gets large blocks of capacity. For startups trying to build frontier models, access to those systems is now a financing question as much as an engineering one. (techcrunch.com, cloud.google.com) For Google, the pitch is that its newest Nvidia-based machines are ready now. For Thinking Machines, the test is whether buying that much compute turns a well-funded lab into a company with models and products to show for it. (googlecloudpresscorner.com, techcrunch.com)
