TechCrunch recently broke the news that Thinking Machines Lab—founded by former OpenAI COO Mira Murati—has struck a multi-billion-dollar deal with Google Cloud to expand its access to AI infrastructure. Sources familiar with the matter put the deal’s value in the single-digit billions, granting the startup access to Google Cloud’s systems powered by Nvidia’s latest GB300 Grace Blackwell chips, plus a suite of cloud services including storage, Kubernetes container orchestration, and Spanner distributed database.

At the heart of the agreement are Nvidia’s GB300 chips, which integrate the company’s Grace CPU and Blackwell GPU into a unified platform optimized for large-scale AI tasks like training large language models (LLMs). Compared to earlier generations, the GB300 boosts computational efficiency per watt by up to 2x—a critical metric for cutting the energy costs and environmental impact of long model training cycles. For Thinking Machines, this access means faster training of more complex models, a key competitive edge in the fast-evolving AI landscape.

For Google Cloud, the partnership fits into a larger strategy to build a robust AI ecosystem. In recent weeks, Google also signed an agreement with Anthropic to provide multiple gigawatts of its custom Tensor Processing Unit (TPU) capacity—signaling a dual approach that leverages both proprietary chips (TPUs) and industry-standard Nvidia GPUs to cater to diverse developer needs. This move aims to challenge Amazon Web Services (AWS), which has long dominated the cloud market and recently secured a 5-gigawatt capacity deal with Anthropic for its Claude model.

Led by Murati, who played an instrumental role in launching ChatGPT during her tenure at OpenAI, Thinking Machines is poised to develop next-generation AI models. The Google Cloud partnership eliminates the need for the startup to invest in building its own data centers, allowing it to focus on core model innovation. Integrating Google’s cloud services—like Spanner for handling large datasets and Kubernetes for managing containerized workloads—will streamline the end-to-end process from data preprocessing to model deployment.

The deal mirrors a wider industry shift: AI startups are increasingly relying on cloud providers for their infrastructure needs. According to Gartner, global AI infrastructure spending is projected to reach $100 billion by 2025, growing at a 35% compound annual growth rate (CAGR). This growth is driven by the rising computational demands of LLMs and generative AI applications, which require massive scale and efficiency.

Competition in the space remains fierce. AWS, Google Cloud’s main rival, recently announced a $10 billion investment to expand its AI data center capacity by 2025, aiming to meet surging demand from startups and enterprises. Microsoft Azure has also ramped up its AI infrastructure offerings, including partnerships with Nvidia and its own custom AI chips. These moves underscore the intense race among cloud providers to capture a larger share of the fast-growing AI infrastructure market.

The Google Cloud-Thinking Machines deal is a win-win: Google strengthens its position in the AI cloud sector, while Thinking Machines gains the resources to accelerate its AI development. As the AI race heats up, cloud providers will continue to compete for high-profile clients, driving further innovation in infrastructure and services.