OpenAI is reportedly preparing to buy artificial intelligence chips from Google, according to a report by The Information relayed by Reuters on October 27. If it happens, it would be one of the most telling deals of the AI infrastructure boom: the world’s best-known AI company buying the custom silicon of its biggest rival in AI products and search.

Here is the important caveat up front, because it shapes everything that follows. This is a reported plan, not a confirmed deal. The Information cited a person familiar with the matter; Reuters reported that OpenAI declined to comment and Google did not immediately respond. No price, volume, chip generation, or delivery timeline has been disclosed. Treat this article as an explainer of what the reported move would mean if it goes ahead — not as confirmation that it has.

Even with that caveat, the report matters. It is the clearest sign yet that the AI chip market is becoming a multi-vendor market, and that even the largest AI companies no longer want to depend on a single supplier. For businesses that buy AI services, that shift has practical consequences for cost, capacity, and negotiating power.

What was actually reported

The core of the story is simple. OpenAI plans to purchase AI chips from Google — specifically Google’s Tensor Processing Units, or TPUs, the custom processors Google designed to accelerate machine-learning workloads. The Information first reported the plan on October 27, US time, and Reuters picked it up the same day. Bloomberg was also reported to have cited a similar source.

Two details in the reporting deserve attention. First, The Information also reported that OpenAI is recruiting engineers to optimise its frontier models for TPUs. Adapting models to a different chip architecture is serious engineering work, and hiring for it suggests intent to run meaningful workloads on the hardware, not just to test it. Second, the report says nothing about terms. Until OpenAI or Google confirms the purchase and its scale, nobody outside the two companies knows whether this is a modest evaluation order or a large, multi-year supply arrangement.

That distinction is worth keeping in mind throughout. A reported plan to buy chips is not the same as chips arriving in data centres, and it is certainly not the same as ChatGPT getting faster or cheaper. Any immediate effect on the products businesses use every day is speculative.

From renting to buying: a relationship that moved fast

The reported purchase is the latest step in a relationship that has changed remarkably quickly. In June 2025, Reuters reported that OpenAI had begun renting Google’s AI chips through Google Cloud, hoping to bring down the cost of running its models. At the time, OpenAI said publicly that it was only testing Google’s TPUs at an early stage and had no plans to deploy them at large scale.

Roughly sixteen months later, the company is reportedly preparing to buy the chips outright. Renting capacity through a cloud provider and owning the hardware are different commitments: renting is flexible and reversible, while buying means planning data-centre space, power, cooling, and a software stack around a specific chip for years.

The commercial relationship between the two companies has deepened on other fronts too. Last week, OpenAI said it would make its frontier AI models available through Google Cloud — a deal covered in our earlier guide to OpenAI’s models arriving on Google Cloud. The two companies compete fiercely in AI products and search, yet they keep striking deals with each other. In the infrastructure era, rivalry and partnership now coexist: Google sells OpenAI compute while competing with it for users.

Why OpenAI wants a second chip supplier

OpenAI is already among the largest buyers of Nvidia’s graphics processing units, the chips that train and run most of today’s leading AI models. So why add Google’s TPUs?

The first reason is capacity. Demand for AI computing is surging faster than any single supplier can comfortably meet. Amazon, Microsoft, and Google’s parent Alphabet plan to spend as much as $600 billion combined this year on AI infrastructure, according to Reuters, and investors are increasingly asking when that spending will translate into returns. In a market where the binding constraint is often hardware availability rather than money, a second source of chips is a second source of capacity.

The second reason is cost and leverage. Google’s TPUs have gained traction as an alternative to Nvidia’s market-leading chips precisely because companies are looking for cheaper options. Even the credible threat of shifting some workloads to TPUs strengthens OpenAI’s negotiating position with every supplier, including Nvidia. It also gives OpenAI a fallback if one supplier’s roadmap slips or its prices rise.

The third reason is strategic independence. OpenAI is also developing its own custom AI chip with Broadcom, which Reuters has reported it plans to deploy by the end of 2026. A company that rents TPUs, buys TPUs, and builds its own chip is not betting on any single supplier — it is building a portfolio. That is the same logic behind Meta running Google’s TPUs in production: the AI chip war has gone multi-vendor, and the biggest buyers are the ones driving it there.

What Google gains from selling to a rival

For Google, selling TPUs to OpenAI would be a significant validation of a long-running strategy. Google designed TPUs for its own services first, then opened TPU capacity through Google Cloud, and this year began selling TPUs directly to some customers. Deals with customers including Anthropic have already fuelled demand for the chips.

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. An OpenAI purchase would fit that timeline — and it lands just as Alphabet reports its quarterly results, with investors watching for signs that heavy AI investment is driving growth in its cloud business.

There is a second, subtler benefit for Google. Every outside customer that optimises models for TPUs strengthens the TPU software ecosystem, which has historically been Nvidia’s greatest advantage. Nvidia’s lead is not just hardware; it is the CUDA software platform and the army of engineers trained on it. If OpenAI — whose models set the standard much of the industry follows — invests engineering effort in TPUs, that ecosystem gap narrows for everyone.

Oracle’s announcement adds another data point. CoinCentral’s coverage of the report noted that Oracle said on Tuesday that TPUs will be available through its cloud infrastructure, widening the channels through which customers can access Google’s chips beyond Google Cloud itself.

GPUs, TPUs, and custom chips: how the options compare

Option Strengths Trade-offs
Nvidia GPUs Mature software ecosystem, broadest model support, largest talent pool High demand and pricing power, supply constraints, single-supplier dependence
Google TPUs Purpose-built for AI workloads, competitive cost, growing cloud availability Smaller software ecosystem, models need optimisation for the architecture
Custom chips (e.g. OpenAI–Broadcom) Tailored to one company’s workloads, long-term cost and control Enormous upfront investment, years to design, no fallback if the design disappoints

The table explains why the biggest AI companies are pursuing all three at once. No single option wins on every dimension, so the rational strategy is a portfolio — and that portfolio logic is now spreading from the AI labs to the enterprises that buy from them.

What this means for businesses that buy AI

Most businesses will never buy a TPU or a GPU directly. But the chip decisions of OpenAI, Google, Meta, and Anthropic shape the prices, availability, and reliability of the AI services you do buy. Here is how to read the reported deal through that lens.

  • Expect capacity to stay tight, but diversify in your favour. More suppliers and more chip types mean the industry’s total capacity grows faster than any single vendor could manage. Over time, that should ease the shortages and waitlists that have frustrated AI buyers — though not overnight, and not evenly.
  • Do not expect an immediate price cut. Nothing in the reporting suggests ChatGPT or any other product gets cheaper tomorrow. Chip purchases take time to deploy, and model optimisation takes longer. Treat any claim of instant savings as speculation.
  • Watch for model behaviour differences. The same model running on different chips can behave slightly differently in speed and cost, even if the outputs match. If you run AI workloads at scale, ask your provider which hardware your workloads run on and whether that will change.
  • Negotiate with the multi-vendor shift in mind. The fact that frontier labs now buy from multiple chipmakers weakens the idea that any one provider’s stack is the only serious option. When renewing AI contracts, compare providers on measured performance and total cost for your workloads, not on brand alone.
  • Build portable AI workflows. The practical lesson of the chip wars is to avoid deep dependence on one provider’s proprietary features where an open or portable alternative exists. Teams that can move workloads between providers will benefit most as competition intensifies. Our WebReOn Tools collection and WebReOn Chat are useful starting points for experimenting with AI utilities without locking yourself into a single vendor’s ecosystem.

What to watch next

Four milestones will show whether this reported plan becomes a real shift or fades as a rumour.

  1. Confirmation and terms. A statement from OpenAI or Google, or a filing that discloses scale, would turn a report into a measurable commitment. Silence, on the other hand, would leave the story as an unconfirmed plan.
  2. Alphabet’s earnings. Alphabet reports quarterly results on Wednesday, October 28. Investors will look for commentary on TPU demand and cloud growth; any mention of new chip customers will be closely read, even if no names are given.
  3. OpenAI’s Broadcom chip. If OpenAI deploys its own custom chip by the end of 2026 as reported, the company will be running three chip strategies at once — Nvidia GPUs, Google TPUs, and its own silicon. How it splits workloads between them will reveal which workloads each chip genuinely suits.
  4. Signs of price or capacity movement. The real test for businesses is whether AI service prices, rate limits, or availability change in 2027. If TPU capacity helps OpenAI serve more users at lower cost, some of that benefit should eventually reach customers — but the timing and size are unknown.

Frequently asked questions

Has OpenAI actually bought Google’s chips?

Not confirmed. The Information reported a plan to buy, citing a person familiar with the matter, and Reuters relayed it. OpenAI declined to comment and Google did not immediately respond. Until either company confirms, treat it as a reported plan.

What are Google’s TPUs?

Tensor Processing Units are Google’s custom-designed chips for accelerating machine-learning workloads. Google has used them internally for years, offered them through Google Cloud, and this year began selling them directly to some customers.

Will this make ChatGPT faster or cheaper?

There is no confirmed evidence of an immediate effect. Any impact would depend on the hardware deployed, how well OpenAI’s models are optimised for TPUs, infrastructure costs, and how OpenAI chooses to use the processors.

Is OpenAI dropping Nvidia?

No. Nothing in the reporting suggests OpenAI will stop using Nvidia GPUs. The reported move is about adding a second supplier, not replacing the first — the same diversification logic behind its custom chip programme with Broadcom.

Why does this matter if I just use AI tools?

Because chip supply shapes what AI costs and how reliably it is available. A multi-vendor chip market should, over time, mean more capacity, more competition, and more choice for the businesses buying AI services — even if the benefits take a year or more to arrive.

Sources

Conclusion

The reported OpenAI–Google chip deal is best read as a signal, not a finished transaction. It signals that the AI industry’s biggest buyer wants a second source of the hardware its business depends on, that Google’s TPUs have earned a seat at the table alongside Nvidia’s GPUs, and that rivalry no longer prevents rivals from doing business together. For businesses, the action is not to react to an unconfirmed report — it is to prepare for the multi-vendor world the report points to: portable workflows, measured provider comparisons, and contracts that leave room to move. The chip war is no longer a single-supplier story. The companies that plan for that will be the ones that benefit from it.