Thinking Machines Lab Inc. is expanding its use of Google LLC’s cloud platform under a newly announced partnership, reportedly valued in the single-digit billions. The artificial intelligence startup, led by former OpenAI CTO Mira Murati, offers a cloud service called Tinker that allows developers to customize open-source large language models.
As part of the deal, Thinking Machines will migrate some workloads to Google Cloud’s A4X Max instances. These instances are optimized for AI models, featuring four Nvidia Blackwell Ultra graphics processing units per virtual machine, supported by 72-core CPUs.
The infrastructure utilizes Nvidia's liquid-cooled GB300 NVL72 appliances, offering extensive memory and high internal bandwidth for data transfer. Thinking Machines is among the first customers to leverage this advanced NVL72 infrastructure.
This partnership follows a recent significant investment in Thinking Machines by Nvidia, which also includes plans for the startup to purchase billions of dollars worth of Nvidia hardware, including Rubin, the successor to the Blackwell Ultra graphics card.