OpenAI is looking to shift from renting cloud capacity to directly purchasing Google’s custom AI processors in a strategic move to diversify its computing supply chains and reduce exclusive reliance on a single architecture.
Initial discussions to purchase Google processors directly without announced contracts
Characteristics of tensor processing units and their position against Nvidia’s dominance
The difference between renting cloud capacity and owning physical chips
The interplay of competition and partnership in the AI infrastructure market
Engineering and commercial dimensions of diversifying computing architectures at OpenAI
Methodological separation between cloud distribution agreements and hardware purchase deals
Frequently asked questions
Initial discussions to purchase Google processors directly without announced contracts
Technology publication The Information, citing a person familiar with the matter, reported in an article followed up by Reuters on October 28, 2026, that OpenAI is considering purchasing Google-designed artificial intelligence chips directly. OpenAI declined to comment on the report, while Google has not issued an official response as of publication time.
Neither report disclosed the potential deal price, timeline, shipment volume, or whether a purchase agreement has actually been signed. Therefore, the precise news relates to a purchasing plan under consideration rather than delivered equipment or a confirmed multi-billion-dollar contract.
Characteristics of tensor processing units and their position against Nvidia’s dominance
Tensor processing units developed by Google are processors specifically designed for machine learning workloads, capable of executing mathematical operations related to training artificial intelligence models and running inference, though the engineering architecture and software options determine practical compatibility with each model.
Nvidia graphics processing units continue to hold a central position in the global market, while companies such as AMD, Google, and Amazon offer various technical alternatives. Major buyers resort to hardware diversification to secure additional computing capacity, different economic terms, or specialized operational capabilities.
The difference between renting cloud capacity and owning physical chips
Reuters had reported in June 2025 that OpenAI had already begun renting Google artificial intelligence processors. Renting computing capacity via a cloud provider differs fundamentally from transferring physical chip ownership and purchasing them directly, making the new report an indicator of a potential shift in the commercial relationship between the two parties.
Last week, OpenAI also announced plans to make its flagship models available via the Google Cloud platform; however, that cloud service distribution agreement is entirely separate from the potential hardware purchase deal and the two arrangements should not be merged into a single completed transaction.
The interplay of competition and partnership in the AI infrastructure market
Google and OpenAI fiercely compete in consumer artificial intelligence products and search engines, but it is common in the technology sector for suppliers and customers to be competitors simultaneously where computing infrastructure, software distribution, and end applications intersect.
Entering into a direct purchase deal would diversify OpenAI’s supplies, but it would not automatically eliminate its heavy reliance on Nvidia systems or its ongoing work to develop custom accelerators in cooperation with Broadcom. Next steps will depend on an official purchase announcement, supplier disclosures, delivery schedules, and independent measurements of operating economics.
Engineering and commercial dimensions of diversifying computing architectures at OpenAI
OpenAI’s potential shift from renting Google tensor processing units via the cloud to studying direct purchase and ownership represents an important transformation in computing asset management. Owning physical hardware grants the company broader control over scheduling training and inference workloads and distributing them, but it requires massive capital investments and long-term operational commitments.
From an engineering perspective, running large language models on multiple processor architectures requires continuous adaptation of software frameworks and optimization tools to avoid performance bottlenecks, which explains OpenAI’s continued combination of Nvidia accelerators and custom chip projects with Broadcom alongside Google options.
Methodological separation between cloud distribution agreements and hardware purchase deals
Cloud computing analysts emphasize the necessity of not confusing OpenAI’s previous announcement regarding making its flagship models available via Google Cloud with the new report concerning the study of purchasing Google chips directly. The former represents a software distribution channel for cloud customers, whereas the latter relates to physical infrastructure ownership.
The report published on October 28 remains based on sources familiar with the matter without official disclosure of the deal value or supply schedule, making any financial estimations of the contract size premature until an official statement is issued by the two companies.
Frequently asked questions
Question: What type of chips is OpenAI considering purchasing from Google?
Answer: OpenAI is considering the direct purchase of tensor processing units designed by Google for machine learning and artificial intelligence workloads.
Question: How does OpenAI’s new direction differ from its previous dealings with Google chips?
Answer: In June 2025, OpenAI began renting Google processors via the cloud, whereas the new development concerns the potential purchase and direct ownership of the physical chips.
Question: Have OpenAI and Google signed a final purchase contract for these processors?
Answer: No, the reports did not disclose the signing of a final agreement or the determination of a price or volume for the deal, and OpenAI declined to comment.
Question: Will OpenAI abandon Nvidia processors if it purchases Google chips?
Answer: No, the potential move aims to diversify computing hardware sources alongside the continued use of Nvidia systems and the development of custom chips with Broadcom.
