🔍 Read the full analysis: OpenAI May Add Google TPUs To Its AI Chip Mix. What About Nvidia? on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
The Information and Bloomberg reported that OpenAI plans to buy Google’s AI chips, but neither OpenAI nor Google has confirmed a deal. The report comes as Google expands TPU sales beyond its own data centers and OpenAI builds relationships with several chip suppliers, including Nvidia, AMD, Broadcom, Cerebras and Amazon.
The Information and Bloomberg reported on October 27 and 28 that OpenAI plans to buy Google’s tensor processing units, or TPUs, adding a potential supplier to the company’s expanding mix of AI chips. Neither company has confirmed a purchase, and the price, chip quantity and delivery schedule remain unknown.
The Information first reported the plan, citing one person familiar with the matter; Bloomberg reported it the following day, also citing one person. OpenAI declined to comment to Reuters, which relayed the first report. Google did not comment. The available reporting does not establish whether the chips would run in Google data centers, OpenAI facilities or a third-party cloud.
There is evidence that OpenAI is preparing to work more closely with TPUs, though it does not prove a deal. A TPU compiler engineering role posted on OpenAI’s careers page on October 20 seeks experience optimizing workloads on the chips and says the hire would work on the company’s frontier models. A job posting indicates technical work, not a purchase order or deployment commitment.
Google has expanded access to TPUs beyond its own cloud. On its October 28 earnings call, the company said nearly 100 customers use TPUs and named Anthropic and Apple. It also cited partnerships with Oracle and Crux AI. Oracle has said it plans to offer Google TPUs through its cloud from 2027. Google previously said it would deliver TPU systems to selected customers’ own data centers.
Infographic 01 · Reported, documented, confirmed
OpenAI and Google’s TPUs: what we actually know
The purchase itself is a report. The pieces around it, Google selling its chips and OpenAI hiring for them, are on the record.
- OpenAI plans to buy Google’s AI chipsReported
The Information, 27 October; Bloomberg, 28 October. Each cites one person familiar with the matter. OpenAI declined to comment; Google did not comment.
- OpenAI is hiring to optimize its frontier models on TPUsPrimary document
“Software Engineer, TPU Compilers”, posted on OpenAI’s careers page on 20 October 2026. A sign of serious work, not a purchase order.
- Google sells TPU systems into customers’ own data centersConfirmed
Announced on 29 April, first deliveries in the second quarter, “a meaningful revenue ramp in 2027” (Alphabet’s CFO, 28 October).
- Oracle will offer Google TPUs in its cloudConfirmed
The first cloud outside Google to do so. Available in calendar year 2027, ramping over several years.
- Price, number of chips, TPU generation, delivery dates, locationNot known
Nothing published. Not even whether the chips would sit in Google’s data centers, OpenAI’s or a partner’s.
How OpenAI and Google got here
- Jun 2025OpenAI rents Google TPUs, calls it early testingReuters: no active plans to deploy them at scale
- 29 Apr 2026Google: TPUs for customers’ own data centersAlphabet Q1 earnings call
- 20 Oct 2026OpenAI posts a TPU compiler jobOpenAI careers page
- 21 Oct 2026OpenAI models coming to Google CloudOpenAI announcement, as reported by Reuters
- 27 Oct 2026The Information reports the purchase plan; Oracle announces TPUsReported · Oracle press release
- 28 Oct 2026Bloomberg reports it too; Alphabet’s call does not mention OpenAIReported · Alphabet Q3 call
“Reported” means the outlets’ own reporting, which I could read only in part (paywalls). “Confirmed” means the company’s own statement or filing.
Sources: The Information via Reuters, 27 Oct 2026, Bloomberg, 28 Oct 2026, as cited by The Asia Business Daily, OpenAI careers, posted 20 Oct 2026, Reuters, 30 Jun 2025, Alphabet Q1 call, 29 Apr 2026, Q3 call, 28 Oct 2026, Oracle, 27 Oct 2026.
Powered by Thorsten Meyer AI
OpenAI Adds Options Beyond Nvidia
If the reported purchase goes ahead, Google would become another possible source of computing capacity for OpenAI’s model development and services. That matters because demand for AI computing is high, and access to chips can affect how quickly companies build and operate their systems. A supplier with a different chip architecture could also give OpenAI more flexibility in matching workloads to available hardware.
The development could strengthen OpenAI’s position when negotiating with suppliers, including Nvidia. That is a potential effect, not a confirmed price or supply outcome: the reports disclose no purchase terms, and there is no evidence that OpenAI is ending or reducing its Nvidia relationship. Nvidia remains part of OpenAI’s chip mix, alongside several announced arrangements with other companies.
For ChatGPT users, there is no reported immediate change. The source material says many of the announced chip arrangements are expected to take effect in 2027 or later, and it is not clear whether any Google TPUs would be used to serve ChatGPT. No price reduction, speed improvement or other user-facing change has been announced.
Products Worth Considering
Apple Mac mini Desktop Computer with M4 chip, 10‑core CPU and GPU, 24GB Unified Memory, 512GB SSD, Gigabit Ethernet
Apple MacBook Air 13-inch with M5 chip, 16GB memory, 1TB SSD, and 12MP Center Stage camera (Midnight)
Apple MacBook Air 15-inch (2026) with M5 chip, 16GB memory, 512GB SSD, Silver
As an affiliate, we earn on qualifying purchases.
Google Opens TPU Sales to Customers
Google developed TPUs for its own data centers and made them available to cloud customers through Google Cloud. This year, it began moving toward direct hardware sales as well. On its April 29 earnings call, chief executive Sundar Pichai said Google would begin delivering TPUs to selected customers in their own data centers. The company has since described customer demand and named commercial partners.
OpenAI had already explored the technology before the October reports. Reuters reported in June 2025 that OpenAI had started renting Google TPUs through Google Cloud. OpenAI told Reuters at the time that it was in early testing and had no active plan to deploy them at scale. On October 21, 2026, OpenAI also said its frontier models would be available through Google Cloud. Those steps show a business relationship, but do not confirm a new chip purchase.
Nor would a Google deal mark OpenAI’s first move beyond Nvidia. The source material lists announced arrangements with AMD, Broadcom, Cerebras and Amazon, as well as Nvidia. It gives figures including 6 gigawatts for AMD, 10 gigawatts of custom chips for Broadcom, 750 megawatts for Cerebras and about 2 gigawatts of Amazon Trainium. These are reported commitments or plans, not evidence that all the capacity has been delivered or is in use.
“As TPU demand grows from AI labs, capital markets firms, and high performance computing applications, we will begin to deliver TPUs to a select group of customers in their own data centers in a hardware configuration.”
— Sundar Pichai, Google chief executive, on the company’s April 29, 2026 earnings call
Deal Terms and Deployment Remain Unknown
No purchase has been confirmed by either company. The two reports each cite one person familiar with the matter; the source material does not provide documentary evidence or a public statement confirming an agreement. OpenAI declined to comment, while Google did not comment.
Key details have not been reported: the number and generation of TPUs, the cost, delivery dates, and where the chips would operate. It is also unclear what work they might run, whether the reported plan has reached final terms, or how much capacity OpenAI would ultimately use. The careers listing is evidence of TPU-related engineering work, but cannot answer those questions.
Claims that this would be OpenAI’s first move away from Nvidia are inaccurate based on the existing announced arrangements described above. At the same time, announced agreements do not show how much computing capacity each supplier will deliver or how the chips will be divided among OpenAI’s workloads.
Watch for Terms and Delivery Dates
The next clear evidence would be a statement from OpenAI or Google confirming an agreement, or further reporting that specifies its scale, cost and timeline. Until then, the October stories remain reports rather than an announced deal.
Google’s planned Oracle cloud offering in 2027 and its direct customer deployments will provide further evidence of how broadly TPUs are being sold. For OpenAI, the practical test will be whether the reported chips are delivered and used at scale. No change to ChatGPT has been announced, and any effect on capacity, service performance or prices remains uncertain.
Key Questions
Has OpenAI confirmed that it will buy Google TPUs?
No. The Information and Bloomberg reported the plan, but neither OpenAI nor Google confirmed a purchase in the source material. OpenAI declined to comment to Reuters, and Google did not comment.
What are Google TPUs?
TPUs, or tensor processing units, are AI chips designed by Google. The company developed them for its own data centers and also makes them available through Google Cloud; it has begun offering systems to selected customers outside its facilities.
Would a Google deal mean OpenAI is leaving Nvidia?
There is no indication that OpenAI is ending its Nvidia relationship. The source material lists other announced chip arrangements with AMD, Broadcom, Cerebras and Amazon, so the reported Google plan would add to a broader supplier mix rather than represent OpenAI’s first non-Nvidia move.
Will Google TPUs change ChatGPT for users?
No user-facing change has been announced. The reports do not say whether OpenAI would use the chips for ChatGPT, and the deal’s timing and scale are unknown. The source material says many of the related chip plans are for 2027 or later.
Source: ThorstenMeyerAI.com
