OpenAI Reportedly Plans to Buy Google's TPU Chips. Neither Company Has Confirmed It.

OpenAI plans to buy Google's AI chips, The Information reported on Oct. 27, 2026, and Reuters carried the report. Neither company has confirmed it. OpenAI declined to comment, and Google did not immediately respond to Reuters. The report, citing a person familiar with the matter, disclosed no financial terms and no other details.

That makes this a signal to watch, not an announced deal. Below is what is known around it, and what each piece would mean if it holds.

What Is Reported

The core claim is simple: OpenAI intends to purchase Google's tensor processing units, or TPUs, the custom chips Google designs for AI workloads. We have no figures for quantity, price, timing or where the chips would be installed, and the source is anonymous.

The report is not a first contact between the companies. Reuters reported in June 2025 that OpenAI had begun renting Google's chips. Last week, OpenAI said its frontier models would be available through Google Cloud, which we covered in OpenAI models on Google Cloud. Renting capacity and buying chips are different arrangements: renting is a service, while owning means OpenAI would run the hardware itself, in its own or a partner's data centers.

The Surrounding Facts

These points come from Reuters, as relayed by TrendForce.

• Google sells TPUs directly now. Reuters says Google began offering TPUs for purchase to select customers this year, and that deals with customers including Anthropic have fueled demand.
• Google expects most of the revenue in 2027. Alphabet's CFO, Anat Ashkenazi, said the company expects to recognize a small share of TPU sales revenue by the end of 2026 and most in 2027, according to Reuters.
• OpenAI is already a very large Nvidia buyer. In June, OpenAI unveiled a custom AI chip designed with Broadcom and said it planned to deploy it by the end of 2026.
• Oracle announced TPUs for 2027. On Oct. 27, Oracle said it will bring Google TPUs to Oracle Cloud Infrastructure in 2027. Oracle senior vice president Karan Batta told Data Center Dynamics it is a multi-year, multi-gigawatt partnership, not an experiment. That would be the first time TPUs are offered through a cloud platform other than Google's own, according to a report TrendForce cites.
• Oracle is not leaving Nvidia. Reuters reported the same day that Oracle is expanding its Nvidia deal to deploy Vera Rubin systems at multi-gigawatt scale, and is also adding accelerators from Cerebras and SambaNova.

What It Would Mean If True

This section is analysis, not reporting.

• OpenAI would be buying from three directions. Nvidia GPUs, a custom Broadcom chip and Google TPUs would give it more than one source of compute. That is a hedge against supply shortages and price power at any single supplier.
• Nvidia faces more competition, not a collapse. Oracle's own plans show new TPU capacity arriving alongside, not instead of, large Nvidia deployments. TrendForce frames the news as adding pressure on Nvidia but gives no market-share numbers, so the size of the effect is unknown.
• Google becomes a chip seller as well as a cloud. If TPUs are sold outside Google's data centers, Google competes with Nvidia as a hardware vendor and, at the same time, with OpenAI as a model maker. The two companies already compete in AI products and search while doing more business with each other.
• Software is the hidden cost. Chips only help if models run well on them. Moving a large model to different hardware takes engineering work. Our model comparison shows the end-user side of these choices: price and capability can differ widely between models, and hardware cost is part of why.

What to Watch

• A confirmation or denial from OpenAI or Google, and any size or schedule.
• Whether the chips would be installed in OpenAI's own sites or at a partner such as Oracle.
• Alphabet's next earnings commentary on TPU sales, which the CFO has said will mostly land in 2027.
• OpenAI's custom chip, which it said would begin deployment by the end of 2026.

What We Cannot Conclude

The central report rests on one anonymous source and has been repeated, not independently confirmed. We do not know the number of chips, the price, the delivery dates or whether OpenAI would use them for training, serving models, or both. Another summary we read says Oracle would buy the TPUs directly for a customer shared with Google, but that customer is not named and we could not confirm the claim, so we did not rely on it.

Why It Matters

• Compute is the constraint. The largest AI labs are now sourcing chips from rivals' product lines.
• Cloud lines are blurring. Oracle, Google and OpenAI are all cooperating where they also compete.
• Costs flow downstream. More chip supply and competition tends to affect what model providers charge, though the timing and size of any effect are unknown.
• Treat it as a report. It is a credible outlet with an anonymous source, and a pending confirmation.