OpenAI to Buy Google TPUs in First Big Break From Nvidia

What the OpenAI Google TPU deal actually changes

OpenAI is adding a second major chip supplier. That is the headline, not the price.

OpenAI has been one of Nvidia's biggest customers for years, buying GPUs to train and run the models behind ChatGPT. According to The Information, the company now plans to purchase Google's tensor processing units as well. Neither company has confirmed the deal, and the report leaves out financial terms, quantities and delivery schedules.

The move matters because it breaks a pattern. Until now, OpenAI's compute strategy was Nvidia first, with custom silicon as a future bet. Buying TPUs from Google turns a rival into a supplier and gives OpenAI a second source for the most expensive part of its business.

This is not a cold start. OpenAI began renting Google's AI chips in 2025, so the relationship already exists. The reported purchase would deepen it from rental to ownership.

Why Google is selling chips to its AI rival

Alphabet wants a return on its AI infrastructure spending, and TPU sales are one way to get it.

Google's TPUs were once internal hardware, used mainly for its own models and cloud services. Alphabet has since opened commercial sales to selected external customers. Finance chief Anat Ashkenazi has said the company expects to recognize a small portion of TPU sales revenue by the end of 2026, with most of it landing in 2027.

That timing is telling. TPU revenue is a 2027 story, not a 2026 one. For Alphabet, selling chips to OpenAI is less about this quarter's numbers and more about proving the TPU business can stand alongside its cloud and advertising arms.

The two companies compete fiercely in AI models, but they already have commercial ties. OpenAI recently announced its advanced models would be available through Google Cloud. Chips are the next layer of that uneasy partnership.

OpenAI's custom chip plan is still running

Buying TPUs does not replace OpenAI's own silicon. It adds a third leg.

In June 2026, OpenAI announced it had designed an AI chip with Broadcom and said it intended to deploy that processor before the end of 2026. That program is described as a multi-generation effort, not a one-off.

So OpenAI is now pursuing three paths at once: Nvidia GPUs, Google TPUs and its own Broadcom-built chips. That is a hedging strategy. Each supplier has different pricing, availability and performance trade-offs, and spreading demand reduces the risk of being stuck when one supply chain tightens.

The catch is complexity. Running models across three different chip architectures takes engineering work, and the software stack around TPUs is not the same as the one around Nvidia's CUDA. OpenAI has the talent to manage it, but it is not free.

The $600 billion question behind the chip shopping

AI infrastructure spending is enormous, and investors are starting to ask when it pays off.

Alphabet, Amazon and Microsoft are collectively expected to spend as much as $600 billion on AI infrastructure during 2026, according to the report. That number explains why chip supply deals are multiplying and why Nvidia's dominance is being tested.

For OpenAI, buying TPUs is a cost and supply move as much as a technical one. More suppliers means more leverage on price and delivery. For Google, it is a chance to turn a huge internal investment into an external business.

None of this is confirmed yet. The reported deal could change in size or scope, and neither company has commented publicly. But the direction is clear: the AI chip market is no longer a one-vendor story.

What it means for you

For readers in the US, UK and EU, the direct impact is indirect: more competition in AI chips could eventually mean lower prices for AI services, but that is a long-term effect, not a switch that flips this year. The reported OpenAI-Google deal has no consumer-facing price or launch date attached.

Businesses that buy AI compute through cloud providers may see more options over time as Google expands TPU sales to external customers. Alphabet has said most TPU sales revenue will be recognized in 2027, so any practical change in pricing or availability is likely to arrive gradually.

OpenAI buying Google TPUs would be a quiet but significant shift in how AI compute gets bought and sold. The terms are unknown, the deal is unconfirmed, and the revenue impact lands mostly in 2027. Still, the direction is unmistakable: Nvidia is no longer the only game in town.