OpenAI Plans to Buy Google AI Chips, Report Says

ByAIERA Blog Editorial Team

Key Takeaways

  • OpenAI reportedly plans to buy Google's Tensor Processing Units, according to a report by The Information, carried by Reuters and other outlets on 27 October 2026.
  • The purchase is reported, not confirmed: OpenAI and Google have not commented on the report, and no deal size, price or delivery schedule has been disclosed.
  • OpenAI already rents Google's AI chips through Google Cloud, a relationship that began in 2025, so a purchase would deepen an existing hardware connection rather than start one.
  • The reported move fits a wider strategy of diversifying beyond Nvidia, alongside OpenAI's own chip project with Broadcom, codenamed Jalapeno.
  • A deal would also matter for Google, giving its TPU business a flagship outside buyer as it begins selling the chips directly to select customers.
  • The timing is notable because it comes days after OpenAI's models became available on Google Cloud, making Google the third major cloud to offer them.

What is reported, and what is not

On 27 October 2026, The Information reported, citing sources, that OpenAI plans to buy Google's AI chips, known as Tensor Processing Units or TPUs. Bloomberg reported the same development, describing it as part of OpenAI's broader push to expand its computing capacity. Reuters carried the report onward, and it was covered the same day by outlets including the Economic Times, the Korea Herald and TipRanks.

Three things are important about the status of this story. First, it is a report based on unnamed sources, not an announcement by either company. Second, neither OpenAI nor Google has provided specific comment on the reports. Third, the specific purchase amount, contract value and delivery schedule have not been disclosed. Any article that presents this as a done deal would be overstating the evidence. What can be said is that a credible technology outlet reports the plan, that a second major outlet has matched it, and that the surrounding facts, from job listings to an existing rental relationship, point in the same direction.

Those surrounding facts are worth laying out because they are confirmed independently of the purchase report. OpenAI has been renting Google's TPUs through Google Cloud since 2025, with some coverage dating the rental relationship to June 2025. OpenAI recently posted job listings seeking engineers to optimise its models for TPUs, with experience in either training or inference welcome. The Information interpreted those listings as a sign that OpenAI's TPU optimisation work is still in its early stages. None of this proves a purchase, but it shows the two companies are already technically engaged on hardware, not merely discussing it in the abstract.

Why TPUs matter as an alternative to Nvidia

Google began developing TPUs more than a decade ago, initially to improve the speed and efficiency of its search engine. The chips are specialised processors for machine learning workloads, and they have since emerged as the most serious alternative to Nvidia's graphics processing units for AI work. Google's own models run on this hardware, including the Gemini 4 Argon family, whose pricing and availability we covered earlier. For training, Nvidia's GPUs remain dominant. For inference, the work of running already trained models to answer user questions, TPUs are seen as potentially improving cost efficiency, which is where much of the industry's growth now sits.

OpenAI is one of Nvidia's biggest customers, using its GPUs at large scale to train models and to run ChatGPT. That dependence has two costs. The first is price: Nvidia's most advanced chips are expensive and in high demand, and every major AI company is competing for the same supply. The second is leverage: a buyer with one supplier has limited room to negotiate, and limited protection if supply tightens. Adding a second hardware supplier addresses both problems at once, even if Nvidia remains the primary supplier.

The reported TPU purchase is therefore best understood as diversification, not replacement. TipRanks noted that OpenAI's use of Google's TPUs does not mean it will stop buying Nvidia chips. As its AI business grows, OpenAI needs more computing power, and Nvidia alone may not be able to meet all of its needs. Using chips from different suppliers could help OpenAI meet growing demand while giving it more room in pricing discussions.

That pattern is visible across the industry. Anthropic, OpenAI's rival, also taps TPUs through a Broadcom arrangement, according to the Korea Herald's coverage. Oracle is bringing Google TPUs to its cloud from 2027. The TPU is moving from a Google-only technology to a product with outside customers, and OpenAI would be the most prominent of them.

The OpenAI and Google relationship is already deepening

The reported chip purchase does not exist in isolation. On 26 October, one day before the report, OpenAI struck a partnership with Google allowing its AI models to run on Google Cloud. With that deal, OpenAI's GPT models became the second model family, after Anthropic's Claude, to be available across all three major cloud platforms: Microsoft Azure, Amazon Web Services and Google Cloud. OpenAI has also been changing how its frontier models are built, including ending live internet training for frontier models, which shows how fast its training and deployment practices are moving at the same time as its hardware plans.

That sequence matters. First, OpenAI's models arrive on Google's cloud. Then, days later, reports emerge that OpenAI wants to buy Google's chips outright. The two steps point in the same direction: a commercial relationship that is widening from software distribution into hardware procurement. The companies still compete fiercely in consumer AI products and in search, and the Economic Times noted that the proposed purchase would deepen commercial ties between OpenAI and Google's parent Alphabet even as the two compete. Cooperation and competition are running side by side, which is increasingly normal in AI infrastructure.

There is also a third leg to OpenAI's hardware strategy. The company is developing its own AI chip with Broadcom, codenamed Jalapeno, and plans to begin using it later in 2026. That project is led by Richard Ho, who previously worked on TPU development at Google. Taken together, OpenAI's hardware picture has three parts: keep buying Nvidia GPUs, buy Google TPUs, and build a custom chip with Broadcom. Each part reduces dependence on the others. None of them, on its own, replaces Nvidia.

What it means for Google's chip business

For Google, an OpenAI purchase would be a significant boost for the TPU business as it challenges Nvidia's GPUs. Google has traditionally made its TPUs available through its own cloud infrastructure, but it has started selling the chips directly to select customers. Selling to OpenAI, one of Google's main AI competitors, would show that the TPU business can stand on its own as a product, not just as an internal Google technology.

The financial timing, however, is slower than the headlines suggest. Alphabet chief financial officer Anat Ashkenazi has said that some revenue from TPU sales contracts would begin to be recognised from the end of 2026, with most of it recognised in 2027. Multi-year supply deals can be signed today, but revenue is typically recorded as chips are delivered and contract milestones are met. Investors should therefore not expect a TPU windfall in Google's next quarterly report, even if the OpenAI deal is confirmed quickly.

The wider context is the sheer scale of AI infrastructure spending. Amazon, Microsoft and Alphabet together plan to invest about $600 billion in AI infrastructure in 2026, according to coverage of the sector's capital plans. Against that total, a single chip purchase, whatever its size, is a small part of the picture. Its importance is strategic rather than financial: it signals that the TPU has arrived as a credible second source for the world's most demanding AI buyer.

What to watch next

The first thing to watch is confirmation. If OpenAI or Google confirms the purchase, the story moves from reported to real, and the disclosed size and timing will show how serious the commitment is. If both companies stay silent, the report remains plausible but unproven, and readers should treat it that way.

The second is the job listings. OpenAI's postings for TPU optimisation engineers are public evidence of intent. If those roles fill and the work advances, it supports the purchase story. If the listings disappear without a deal, that would cut the other way.

The third is delivery. TPU supply, like GPU supply, is constrained, and Google must balance its own internal demand, its cloud customers and any direct sales. When chips actually ship to OpenAI, and at what scale, will determine whether this is a meaningful diversification or a symbolic one.

The fourth is Nvidia's response. Nvidia is also a major investor in OpenAI, and its Blackwell chips and upcoming Vera Rubin platform are in high demand. If OpenAI buys fewer Nvidia chips, other cloud providers and AI companies could use the available supply. Nvidia's position is not threatened by a single reported purchase, but a pattern of diversification by its biggest customers would be.

For readers, the cleanest summary is this. OpenAI reportedly wants to buy Google's AI chips. The report is credible and corroborated by a second outlet, but it is not confirmed, and no terms are known. The move fits everything else OpenAI is doing on hardware: renting TPUs already, hiring TPU engineers, building its own chip with Broadcom, and putting its models on Google Cloud. Whether the purchase happens, and how big it is, is the next chapter. The direction of travel is already clear.

Sources

Similar Posts