Two of the fiercest rivals in artificial intelligence may be about to become much closer business partners. OpenAI, the company behind ChatGPT, reportedly plans to buy Google's custom AI chips, known as Tensor Processing Units (TPUs), according to a report by The Information on Tuesday, 27 October 2026, carried by Reuters.
That single sentence matters more than it first appears. OpenAI is one of the largest buyers of Nvidia's graphics processing units (GPUs), the chips that have powered almost the entire AI boom. If the maker of ChatGPT starts buying hardware directly from Google — the very company whose Gemini models compete with ChatGPT — it would mark a real shift in how the AI industry buys its most important raw material: computing power.
Nothing here is confirmed by either company yet. Google did not immediately respond to Reuters' request for comment, and OpenAI declined to comment. The report did not disclose financial terms or other details, and it rests on a person familiar with the matter. This article therefore treats the purchase plan as reported, not confirmed — and explains why, even as a report, it tells us a lot about where AI computing is heading.
What has actually been reported
The core of the story is simple. The Information reported that OpenAI plans to buy Google's TPU chips directly, rather than only renting computing power through Google Cloud. Reuters reported the story on 27 October 2026, and it was widely republished on 28 October.
That distinction — buying versus renting — is the heart of the news. Reuters had already reported in June 2025 that OpenAI had begun renting Google's AI chips through the cloud. Renting gives a company flexibility: it can scale computing up or down and avoid a large upfront purchase. Buying chips outright is a bigger commitment. It suggests OpenAI wants a more permanent, predictable supply of TPU computing, and more direct control over how that hardware is used.
It also deepens a relationship that has been growing quickly despite the rivalry between the two companies. Last week, OpenAI said it would make its frontier AI models available through Google Cloud. The two companies now compete for users while selling each other services. That kind of "compete and cooperate" arrangement is becoming normal in AI, because no single company can supply everything it needs on its own.
The scale of spending behind all of this is enormous. Amazon, Microsoft and Google's parent company Alphabet plan to spend as much as $600 billion combined this year on AI infrastructure, according to Reuters. That spending sits inside a wider funding boom, as we explored in DeepSeek's $15B raise and the global AI funding race. Investors are asking when that spending will pay off.
From renting chips to buying them
To understand why buying matters, it helps to understand the two ways AI companies get computing power. The first is the cloud model: a company rents time on someone else's machines, paying for what it uses. This is flexible and fast, and it is how most AI start-ups begin. The second is ownership: a company buys the chips, installs them in data centres run by itself or its partners, and runs them for years. Ownership makes sense only at very large scale, when a company wants to lock in supply and cost for years.
OpenAI is now clearly at that scale. It trains very large models and then serves them to hundreds of millions of users, which means it needs computing for two different jobs: training, where a model learns from data, and inference, where a finished model answers questions. Inference is the job that grows every day with the user base, because every ChatGPT answer costs computing power.
Google's TPUs were designed for exactly this kind of large-scale work inside Google itself, powering search, YouTube recommendations and Gemini. Google began selling TPUs directly to some external customers this year, turning an internal tool into a product. If OpenAI becomes a direct buyer, it would be one of the strongest signs yet that TPUs have moved from a Google-only advantage to a genuine alternative in the open market.
Why TPUs appeal to OpenAI
There are three straightforward reasons why OpenAI might want Google's chips alongside Nvidia's GPUs.
The first is choice. When one supplier dominates a market, buyers have little bargaining power and little protection if supply runs short. Nvidia's GPUs are the industry standard, and OpenAI remains one of Nvidia's largest customers. But adding a second major supplier gives OpenAI more options on price and availability, and reduces the risk of depending on a single source for something so critical.
The second is cost. TPUs have gained traction as an alternative to Nvidia's chips partly because companies are looking for cheaper options, and for chips suited to running AI models efficiently. As AI spending soars, that search has become urgent. A chip designed for a specific kind of AI work can sometimes do that work more cheaply than a general-purpose GPU, even if the GPU is more flexible overall.
The third is fit. Different AI jobs suit different hardware. Training a brand-new frontier model and answering a billion short questions are not the same task, and the best chip for one is not always the best chip for the other. Having several types of hardware lets a company match each job to the machine that does it most efficiently.
The same week in AI chips: Oracle opens the door wider
The OpenAI report did not arrive in isolation. On the very same day, 27 October, Oracle announced that Google's TPUs will be available on Oracle Cloud Infrastructure (OCI) starting in calendar year 2027, ramping over several years. Oracle said it will be the first hyperscaler outside Google to offer Google TPUs, and it will supply the data-centre power, cooling and networking needed to host them.
Oracle's announcement went further. Alongside Google's chips, Oracle plans to offer chips from Cerebras Systems and SambaNova Systems for inference work, while expanding its Nvidia Vera Rubin systems to multi-gigawatt scale. Oracle's cloud president, Mahesh Thiagarajan, told CNBC that the new chips are not a pivot away from Nvidia, calling Nvidia's chips the best and most fungible. The message was clear: customers want a menu, not a single dish. That menu approach is spreading beyond the biggest clouds, as shown by GlobalFoundries' FDX Fusion chips that will power physical AI.
There is also a third leg to OpenAI's hardware strategy. In June 2026, OpenAI unveiled its first custom AI chip, designed with Broadcom and called Jalapeño, built for running language models, with deployment planned by the end of this year as part of a multi-generation chip programme. So OpenAI's emerging picture has three layers: Nvidia GPUs, its own Broadcom-built chip, and now possibly Google TPUs bought directly.
What this means for Nvidia
It is tempting to read every TPU story as bad news for Nvidia, but that would be too simple. Nvidia still dominates AI training, its chips remain the most flexible and widely supported, and both OpenAI and Oracle are still expanding their Nvidia purchases. Nvidia is also pushing into new kinds of AI work itself, as seen in Nvidia's Flora robot-video labelling system.
The real pressure on Nvidia is subtler. For years, Nvidia could charge a premium because there was no serious alternative at scale. As Google TPUs, custom chips from Broadcom, and specialist inference chips from Cerebras and others become easier to buy, large customers gain real alternatives for at least some of their work. That does not dethrone Nvidia, but it does give buyers leverage in negotiations, and it may slow the pace at which Nvidia can raise prices. Cerebras shares rose about 10% after Oracle's announcement, while the wider story was framed as diversification rather than desertion.
What this means for Google
For Google, the reported OpenAI interest validates a long, expensive bet. Google has designed its own AI chips for years, mainly for its own services. Selling them to others, including direct rivals, is a newer strategy, and it only works if outside buyers trust the hardware, the software tools around it, and the support that comes with it.
The financial payoff will not be instant. Alphabet's 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. Google itself is therefore telling investors to judge the TPU business on next year's numbers, not this year's.
What it could mean for everyday users
Most readers will never buy a TPU or a GPU. So why should this story matter outside the chip industry? Because computing cost is, in the end, what decides how good, how fast and how cheap AI services can be.
Cheaper, more secure computing gives OpenAI room to hold prices steady and invest in better models, while stronger chip competition should bring faster hardware and lower costs over time.
What to watch next
Several things will show whether this report turns into a real shift. First, confirmation: will OpenAI or Google confirm a purchase, and on what scale? A small test purchase and a large multi-year supply deal are very different stories, and the report gives no figures. Second, Alphabet's quarterly results, due later on Wednesday 28 October, and any comment on TPU demand. Third, deployment: how quickly do OpenAI's models actually run well on TPUs? Moving a large model between chip types is real engineering work, not a simple swap. Fourth, Nvidia's response on pricing and new products.
The bottom line
The reported plan for OpenAI to buy Google's TPU chips is, for now, just that: a report, based on one person familiar with the matter, with no terms disclosed and no comment from either company. It should not be treated as a done deal.
But even as a report, it captures the defining hardware story of late 2026. The AI industry is moving from a world with one dominant chip supplier to a world with several: Nvidia GPUs, Google TPUs, OpenAI's own Broadcom-built chip, and a growing set of specialist inference chips, all available through clouds like Oracle's.
If that trend continues, the winners will not only be the chipmakers. They will be the users, who get faster, cheaper and more reliable AI services built on a computing market that finally has real competition. That is why a short report about a chip purchase deserves more attention than its length suggests — and why the next confirmation, or denial, will be worth watching closely.