OpenAI reportedly plans to buy Google TPUs for AI compute

OpenAI reportedly plans to purchase Google’s tensor processing units after renting Google AI chips in 2025.

Reuters, citing The Information, reported that OpenAI “plans to purchase artificial intelligence chips from Google.” Financial terms and other details remain undisclosed, so the proposal carries no disclosed order value, revenue figure or capacity commitment. The numbers are missing.

The planned deal would give OpenAI another source of AI computing as demand for infrastructure rises. OpenAI is among the biggest buyers of Nvidia’s GPUs, which it uses to train models and run inference workloads; buying Google chips would add a hardware option without signaling any quantified replacement of Nvidia processors.

What does the reported deal establish?

For now, it establishes intent and a deeper supplier relationship, not a completed, quantified sale. The report does not identify a TPU generation, order quantity, delivery schedule, unit price or total value. It therefore signals strategic diversification while leaving the commercial scale of the proposed purchase unresolved.

OpenAI began renting Google’s AI chips in 2025, according to earlier Reuters reporting. A purchase would mark another step into Google’s hardware ecosystem and expand on that rental relationship, although the source does not describe the products or deployment.

Why add Google’s TPUs?

Nvidia remains the dominant processor supplier described in the report, and OpenAI is among the biggest buyers of its GPUs for training and inference. Google’s TPUs handle machine-learning workloads and are emerging as an alternative, with developers exploring potential cost or performance advantages for specific AI tasks.

The potential gains remain unquantified. The source names no workload, benchmark or system configuration demonstrating lower TPU cost or higher performance, so OpenAI’s commercial rationale remains supply diversification and another route to capacity rather than a proven efficiency claim.

How does Google sell its TPUs?

Google has traditionally made TPUs available through its own cloud infrastructure, but it has started selling the chips directly to select customers. The proposed OpenAI purchase would align with that newer sales channel, potentially extending Google’s chip business beyond cloud-delivered computing as demand for AI computing grows.

For Google, direct sales could help capitalize on growing demand. For OpenAI, buying TPUs would represent another step in its engagement with Google’s hardware ecosystem, beyond the rentals that began in 2025.

What changes in the OpenAI-Google relationship?

The proposed transaction would deepen commercial ties between OpenAI and Alphabet, Google’s parent, even as the companies compete in AI products and search. More hardware cooperation would not resolve their product rivalry; it would add an infrastructure dimension to a relationship already shaped by competition.

If the plan becomes a purchase, OpenAI would gain a potential supplier outside Nvidia’s GPU ecosystem while Google would gain OpenAI, a major Nvidia GPU buyer, as a chip customer. Neither outcome is guaranteed because the reported deal has no disclosed scale, schedule or price.

What remains unknown?

The source does not support conclusions about TPU shipment volumes, process nodes, wafer sizes, capacity reservations, pricing or Nvidia displacement. It also offers no comparison of training throughput, inference performance or total cost of ownership, leaving the commercial case for a large OpenAI order untested.

That distinction matters because a purchasing plan can signal strategic intent without establishing how much compute OpenAI would receive. Google’s broader direct-sale move is more concrete: the company has already begun selling TPUs to select customers, whereas the OpenAI purchase remains a reported plan without disclosed terms.

What could happen next?

If the deal advances, Google could win a larger role in OpenAI’s AI infrastructure and OpenAI could secure an additional route to computing capacity. Whether that changes relative pricing, workload economics or competition with Nvidia depends on the product mix, order size and deployment schedule, none of which the cited reports disclose.

The immediate signal is strategic rather than numerical: OpenAI is seeking an alternative to Nvidia’s dominant GPUs, while Google is trying to turn its own TPUs into a broader merchant business. Disclosed order value, shipment volume and delivery timing will show whether either ambition becomes a material shift in supply or pricing.