OpenAI is planning to buy AI chips from Google, according to a report that landed on Tuesday. The Information first reported the plan, Reuters carried it, and neither company confirmed it. That unconfirmed status matters, but so does the direction of travel. The two companies that compete hardest in AI are now entangled as customer, distributor, and rival at the same time.
The report, cited by Reuters on October 27, said OpenAI plans to buy Google’s Tensor Processing Units, or TPUs, but disclosed no financial terms, no quantities, and no timeline. Google did not immediately respond to Reuters outside regular business hours, and OpenAI declined to comment. Treat this as a well sourced report, not a done deal, until one of the companies says so on the record.
Reported, not confirmed, and why the distinction matters
Chip deals of this kind usually leak before they are announced because they involve long supply negotiations, and because both sides have reasons to stay quiet. OpenAI does not want to signal how much compute it still needs, or from whom. Google does not want to preempt its own earnings call. Alphabet reports quarterly results later on Wednesday, October 28, and investors will be listening for any hint about TPU demand.
What is already on the record is substantial. Reuters reported in June 2025 that OpenAI had begun renting Google’s AI chips. Last week, OpenAI said its frontier models would be sold through Google Cloud. A purchase would be the next step on a path both companies have already walked, not a sudden reversal.
Three relationships at once
| Relationship | What it looks like | Status |
|---|---|---|
| Customer | OpenAI rents, and now reportedly plans to buy, Google TPUs for training and running models | Renting confirmed in 2025, purchase reported in October 2026 |
| Distributor | OpenAI models, including GPT-6 Astra, sold through Google Cloud’s Model Garden and Marketplace | Announced October 22, 2026 |
| Competitor | ChatGPT competes with Google Search and Gemini for the same users and the same questions | Ongoing |
That three way tangle is the story. A decade ago, a deal like this would have looked like surrender by one side. In 2026 it looks like arithmetic. Building frontier models needs more compute than any single supplier can comfortably provide, and selling models needs every distribution channel available.
Why OpenAI would want Google’s chips
OpenAI is one of the largest buyers of Nvidia’s graphics processing units, and it will stay one. TPUs are not a replacement for Nvidia in this story, they are a second source. Google’s TPUs have gained traction as a cheaper alternative to Nvidia’s chips, particularly for running models once they are trained, the work known as inference. Google began selling TPUs directly to some customers this year, and rival cloud providers are now lining up to offer them.
Reuters also put a number on the spending race behind all of this. Amazon, Microsoft, and Alphabet plan to spend as much as $600 billion combined this year on AI infrastructure, even as investors ask when that spending turns into returns. In a market spending at that scale, no lab wants a single point of failure in its chip supply, and no cloud provider wants to watch a rival’s chips become the cheaper default.
Oracle is making the same bet, in public
The OpenAI report did not arrive alone. On Tuesday, Oracle said it is expanding its cloud AI chip deals with Nvidia to multi-gigawatt scale with Nvidia’s newest Vera Rubin servers, while also offering Google’s TPUs for the first time, alongside chips from Cerebras and SambaNova. Oracle’s cloud chief, Mahesh Thiagarajan, put it plainly: different AI workloads have different compute needs, and customers want the flexibility to choose the right infrastructure for each one.
Cerebras chief executive Andrew Feldman told Reuters the Oracle deal was a massive expansion of his company’s tier one customer base. The pattern is consistent. Nvidia still dominates training, the work of building a model. Inference, the work of answering millions of user prompts cheaply, is more competitive and more diverse, and that is where TPUs, Cerebras, and others are gaining ground.
What this means for Nvidia, Google, and ChatGPT users
For Nvidia, this is not a breakup story, it is a diversification story, and diversification still stings. Every TPU that OpenAI buys is a GPU it did not buy, and every cloud that offers TPUs gives customers a reason to compare prices. Nvidia’s answer has been scale and speed, and its Vera Rubin systems are the centrepiece of that answer.
For Google, a direct chip sale to OpenAI would turn a competitor into a marquee TPU customer, and it would validate the decision to sell TPUs beyond its own data centers. Alphabet chief financial officer Anat Ashkenazi has previously said the company expects to begin recognising a small percentage of revenue from TPU sales agreements by the end of this year, with the vast majority expected in 2027. Wednesday’s earnings call is the next chance to see whether that timetable still holds.
For ChatGPT users, the practical effect is indirect but real. Cheaper, more plentiful inference compute is what keeps subscription prices and API prices from rising faster, and it is what makes features like computer use affordable to run at scale. Our October price list for a million tokens shows how quickly those costs are moving, and chip supply is a large part of why.
What to watch next
Four things will tell us whether this report becomes a real deal. First, any confirmation from OpenAI or Google, or a mention on Alphabet’s earnings call. Second, the scale, since a test purchase and a multi-year supply agreement are very different stories. Third, which TPU generation is involved, since that determines what workloads it can handle. Fourth, whether OpenAI’s models on Google Cloud, covered in our report on the Google Cloud deal, start to run on the same hardware they are sold through.
Until then, the safest summary is this. OpenAI and Google are competitors, partners, and now possibly customer and supplier, all at once. In the AI economy of 2026, that contradiction is not an exception. It is the business model.
Sources: Reuters, October 27 and 28, 2026, reporting The Information’s account of the planned purchase, Oracle’s October 27 chip announcement, and OpenAI’s October 22 Google Cloud announcement.
