OpenAI turns to Google Cloud for extra GPUs

When you're starving for compute, you buy GPUs from anyone who has them—even your fiercest rival.

OpenAI has officially added Google Cloud to its roster of infrastructure providers. The company plans to run workloads for both ChatGPT and its API across Google's data centers in the U.S., the U.K., the Netherlands, Norway, and Japan.

The expansion comes after Reuters first reported talks between the two companies in June. Google now joins a supplier roster that already includes Microsoft, Oracle, and CoreWeave.

Chasing every available cluster

OpenAI used to run exclusively on Microsoft Azure. But as user growth exploded and the models grew massive, Microsoft simply couldn't keep up with the demand.

The cracks in that exclusivity started showing months ago. Last year, Microsoft officially labeled OpenAI as a competitor, given that both companies sell rival AI tools and subscriptions. Then in January, the two renegotiated their deal: Microsoft surrendered its exclusivity in exchange for a right of first refusal whenever OpenAI needs more computing power. Microsoft also retained exclusive rights to host OpenAI's APIs for third-party developers.

Since then, OpenAI has been scrambling for chips anywhere it can find them. It tapped Oracle Cloud Infrastructure last year, teaming up with Microsoft on data center buildouts linked to the White House's $500 billion Stargate initiative. In March, OpenAI locked in a massive five-year, $12 billion compute deal with CoreWeave.

Even that wasn't enough. By April, CEO Sam Altman took to X with an urgent plea: "if anyone has GPU capacity in 100k chunks we can get asap please call!"

Google Cloud answered the call.

Why it matters

For end users and developers, this should mean fewer capacity hiccups, less downtime, and better latency for ChatGPT and API calls, especially across Europe and Japan where Google's local infrastructure is coming online. If you rely on OpenAI's models for production workloads, diversifying the underlying metal is pure upside.

For the broader industry, it shatters the illusion of neat AI alliances. We used to think of this as Microsoft-OpenAI versus Google-Gemini. But cloud infrastructure is a brutal commodity game. Google Cloud, which still trails AWS and Azure by market share, now hosts both of Microsoft's biggest AI headaches: Anthropic and OpenAI.

We think OpenAI's decision to hand millions of dollars to its arch-rival proves just how desperate frontier AI labs are for compute. Pragmatic? Absolutely. But it also shows that as the model wars rage on, the hyperscalers selling power, cooling, and silicon hold all the leverage.