Mira Murati’s Thinking Machines Lab Lands Billion-Dollar Google Cloud Deal

Mira Murati’s new AI startup has landed the one thing every frontier lab is desperate for: more compute.

Google has signed a new agreement to expand its cloud partnership with her AI startup, Thinking Machines Lab. The agreement effectively plugs the company into Google’s powerful “AI Hypercomputer” hardware, giving the young company the raw horsepower it needs to train its next generation of frontier models.

While the exact figure hasn’t been made public, sources familiar with the deal told TechCrunch the agreement is valued in the “single-digit billions.”

Thinking Machines, which Murati launched in early 2025 after her high-profile exit as OpenAI’s Chief Technology Officer, is already valued at $12 billion despite its relative youth.

Tapping into the ‘AI Hypercomputer’

Under the new deal, Thinking Machines Lab will gain massive access to Google’s “AI Hypercomputer” infrastructure. This isn’t just standard cloud storage; it’s a specialized stack of hardware and software designed specifically for the grueling task of training frontier models.

The startup will be one of the first to use Google’s A4X Max virtual machines, which run on Nvidia’s latest Blackwell (GB300) chips. In early tests, the lab saw training and serving speeds double compared to older technology.

“By leveraging A4X Max and the AI Hypercomputer integrated stack, Google Cloud got us running at record speed with the reliability we demand,” said Myle Ott, a founding researcher at Thinking Machines Lab, in a statement released by Google.

Building the Tinker engine

The partnership is designed to fuel Tinker, Thinking Machines Lab’s first product launched last Oct. Tinker is a tool meant to help organizations automate the creation of their own custom AI models.

The system relies heavily on reinforcement learning. However, reinforcement learning is incredibly hungry for data and processing power. To handle this, Thinking Machines Lab is tapping into Google’s deep toolbox, using services such as the Spanner database and Google Kubernetes Engine to keep its massive systems running smoothly.

Mark Lohmeyer, VP & GM of AI and Computing Infrastructure at Google Cloud, noted that the partnership aims to help the lab move even faster.

“Through this new agreement, and our deep partnership with NVIDIA, we’ll help Thinking Machines accelerate even further, using Google Cloud’s AI Hypercomputer, which brings together purpose-built hardware, open software and flexible consumption models in an optimized architecture,” Lohmeyer said.

Google’s move comes at a time when the compute wars are reaching a fever pitch. Just this week, Anthropic signed a massive deal with Amazon to secure 5 gigawatts of power. Google also recently inked a deal with Anthropic for its own custom chips.

While Thinking Machines Labs has already secured a “significant investment” from Nvidia and plans to use their next-generation Rubin chips in 2027, this Google deal gives them the immediate firepower they need to compete today.

Also read: Google Cloud is taking direct aim at Nvidia with new TPU chips built separately for training and inference, a move that shows how the AI infrastructure fight is expanding beyond GPUs.