Why would OpenAI buy chips from Google when it already has Nvidia?

A rival's chips in your data centres sounds odd. Here is why it might be the most practical move in the AI race.

Reported by The Information on 27 October: OpenAI plans to buy Google's TPUs, the AI chips Google designs. Not just rent them, which it has done since 2025, but buy. It is even hiring an engineer to lead the work of tuning its frontier models to run on them. Neither company has confirmed the plan.

Think about how strange that is. Google and OpenAI compete head to head in consumer AI. And the same day, Oracle announced it would become the first big cloud provider outside Google to offer TPUs.

Is Nvidia not enough anymore?

So is Nvidia failing OpenAI, or is this about saving on power and the cost of training? Tempting stories, but the reporting does not say either of those.

What it does say is that OpenAI's relationship with Nvidia stays deep. This looks like a hedge, not a breakup. If you spend that much on compute, depending on one supplier is a risk you can feel in your bones, in prices, in delivery times, in what you can negotiate.

There is a technical price for it, too. Models and software tuned for one kind of chip do not simply run on another. Hiring someone to bridge that gap shows the work is real and still at an early stage.

The takeaway: the AI hardware market used to be one company and a lot of customers. It is slowly turning into several suppliers, and the labs that fight each other on models now share the same chips.