Thinking Machines Lab与Google Cloud签署数十亿美元算力协议

ME News, April 22 (UTC+8): According to Dongcha Beating monitoring, Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has signed a multi-billion-dollar AI infrastructure agreement with Google Cloud. The deal includes a computing system powered by Nvidia’s latest GB300 chips, as well as supporting model training and deployment services. Thinking Machines Lab is among the first customers to use Google Cloud’s GB300 system. Google states that the system improves training and inference speed by 2x compared with the previous generation GPUs. This is the first time Thinking Machines Lab has signed an agreement with a cloud service provider. Previously, the company reached a partnership with Nvidia and received investment from it, but it had not used any third-party cloud platforms. The agreement is non-exclusive, meaning Thinking Machines Lab can use multiple cloud service providers in the future.

Google’s press release specifically mentioned support for the company’s reinforcement learning workloads. Its first product, Tinker, is built on a reinforcement learning architecture used to automatically generate customized frontier models. After leaving OpenAI in February 2025, Murati founded Thinking Machines Lab and completed a seed round of $2 billion, valuing the company at $12 billion. Recently, Google has been signing deals with AI developers intensively: earlier this month, Anthropic signed multi-GW TPU agreements with Google and Broadcom, and this week it also signed a training and deployment agreement with Amazon for up to 5GW. (Source: BlockBeats)