What the report contains, and what it does not

Strip the story to its evidence and it is short. One person familiar with the matter told The Information that OpenAI plans to buy Google's TPUs. That is the claim. Around it sits a ring of absences: no contract value, no number of chips, no named TPU generation, no delivery schedule, no statement that an agreement has been signed. OpenAI declined to comment. Google had not responded when Reuters published.

That does not make the report worthless. The Information's sourcing on OpenAI's infrastructure has a track record, and "plans to buy" is a specific claim, narrower than a signed order and broader than idle interest. But readers should hold it at the strength it was published at. A plan, sourced once, is a statement about intention. A reported purchase plan can change before an agreement is signed or hardware is delivered. Without disclosed terms or confirmation from either company, the scale and likelihood of this particular transaction remain unknown.

Renting was the documented step

The renting half of this story is not in dispute. Reuters reported in June 2025 that OpenAI had begun renting Google's AI chips, and Data Center Dynamics' summary of that period's reporting describes it as OpenAI's first meaningful use of non-Nvidia chips, arranged through Google Cloud with the aim of lowering inference costs. The same reporting noted a limit: Google was not renting its most powerful TPUs to OpenAI.

That history is what gives the new claim its weight, if it holds. Renting capacity through a cloud is a procurement decision that can be unwound at renewal. A direct purchase would represent a different commercial commitment from renting TPU capacity through Google Cloud. Depending on the eventual arrangement, it could also require longer-term investments in deployment and software optimization. The report does not establish who would operate the hardware or where it would be installed. OpenAI already splits its compute across Microsoft, Oracle, CoreWeave and its own Stargate builds. A direct TPU purchase would add Google, its closest product rival in search and AI assistants, to the list of suppliers it depends on rather than merely rents from.

The one public signal is a job listing

OpenAI's own careers page lists a Software Engineer, TPU Compilers role in San Francisco. The description calls for optimizing frontier models across the full forward pass, from model code through kernels to hardware execution, and identifies quantization as one approach to improving inference efficiency. That is direct evidence of TPU engineering investment, not of a hardware purchase. AI Weekly also noted that OpenAI's hardware division is run by Richard Ho, who previously led development of Google's Cloud TPUs.

A compiler hire is not a purchase order. Companies staff for platforms they rent as well as platforms they own, and OpenAI has needed TPU optimization work since the 2025 rental began. The listing establishes that OpenAI is recruiting for TPU optimization involving its frontier models. It does not establish whether those models are running on purchased hardware, rented capacity or a future deployment. That is context for the report, not confirmation of it.

The timing does some of the talking

The report landed the evening before Alphabet's quarterly results, due after the close on Wednesday, October 28. Google's TPU business is one of the questions hanging over that print. Reuters notes Google began selling TPUs directly to some customers this year, a shift from a long period in which the chips were effectively reserved for Google's own services and cloud tenants. Earlier on October 27, Oracle announced it will offer Google TPUs on Oracle Cloud Infrastructure from 2027, the first hyperscaler outside Google to do so.

Read together, TPUs are becoming a product Google sells through other companies' data centers as well as its own. If confirmed, a direct OpenAI purchase would be a notable endorsement of Google's effort to sell TPUs beyond its own cloud. It would also illustrate how a major Nvidia customer is diversifying its hardware options, rather than necessarily replacing Nvidia GPUs. Which is precisely why the missing confirmation matters. The items that would settle this are concrete and checkable: any comment from either company on the October 28 Alphabet call, a disclosed contract value, a chip generation, a delivery window. Until one of those exists, the story measures OpenAI's appetite. It does not yet measure Google's revenue.

Sources

  1. [1] Reuters, "OpenAI plans to buy Google's AI chips, The Information reports," October 27, 2026 (wire read in full via Euronext syndication; the purchase claim, the absence of terms, both companies' responses, the June 2025 rental report and Alphabet's reporting date)Read source
  2. [2] OpenAI, "Software Engineer, TPU Compilers," job listing, San Francisco, published October 20, 2026 (read first-hand on OpenAI's careers board)Read source
  3. [3] The Information, "OpenAI Plans to Buy Google TPUs, Posts Job to 'Lead the Optimization' of Frontier Models on Them," October 27, 2026, reported by Ryan Barwick (original paywalled; headline as cited by Pivot News)Read source
  4. [4] AI Weekly, "OpenAI Plans to Buy Google TPUs, Hires for Compiler Role," October 28, 2026 (Richard Ho background)Read source
  5. [5] Data Center Dynamics, "OpenAI to use Google's TPUs — report" (2025 rental background: first meaningful non-Nvidia use, inference-cost aim, most powerful TPUs not rented)Read source
  6. [6] Finimize, "OpenAI Looks To Buy Google's TPUs For AI Workloads," October 28, 2026 (secondary account of the Reuters wire)Read source