OpenAI Plans to Buy Google AI Chips to Further Diversify Compute Strategy as Nvidia Faces New Competition

TradingKey - OpenAI is seeking to expand its sources of AI computing power. According to a Tuesday report from The Information, citing people familiar with the matter, the ChatGPT developer plans to procure artificial intelligence chips from Google (GOOGL) to meet its model training and operational needs and reduce its reliance on a single chip vendor. Currently, the two sides have not finalized the terms of the potential deal, and the procurement scale and financial details have not been disclosed.

OpenAI declined to comment, while Google did not immediately respond to media inquiries.

If the plan materializes, Google's Tensor Processing Units (TPUs) will further enter OpenAI's computing ecosystem. Although OpenAI and Google compete in AI models and search, their commercial ties are expanding. Last week, OpenAI announced that it would offer its frontier AI models through Google Cloud, and this potential chip procurement plan indicates that their collaboration could extend further from cloud services to underlying computing infrastructure.

As the scale of AI models continues to expand, tech companies' demand for computing resources is growing rapidly, making chip supply and computing costs critical factors in industry competition.

OpenAI has been a major buyer of Nvidia (NVDA) GPUs, which are widely used for AI model training and inference. However, as model scale, inference traffic, and data center construction costs continue to rise, relying solely on a single chip and cloud platform could not only constrain computing power supply, but also weaken OpenAI's bargaining power on pricing and lead times.

The growing attention on Google's TPUs stems from their positioning in the AI computing market. Unlike Nvidia's GPUs, which are offered to a broad range of customers, TPUs are application-specific chips designed by Google for specific AI tasks. As companies seek to lower computing costs and diversify supply chain risks, TPUs are increasingly becoming an alternative considered by some customers. Google also began selling TPUs directly to select customers this year, further expanding the commercial potential of the business.

For Google, rising demand for TPUs is expected to create new growth avenues for its cloud business and chip sales. A number of AI companies, including Anthropic, have already partnered with Google, further boosting market focus on its TPU products.

Alphabet Chief Financial Officer Anat Ashkenazi previously stated that the company expects to begin recognizing revenue from certain TPU sales agreements before the end of 2026, with the majority of related revenue expected in 2027. As Alphabet prepares to release its quarterly results, investors will also focus on whether AI-related demand can continue to drive Google Cloud's growth and when infrastructure investments will convert into more noticeable financial returns.

OpenAI is not relying solely on external chip vendors; the company is collaborating with Broadcom (AVGO) to develop custom AI processors and plans to deploy them later this year to provide computing support for its AI models. Reports indicate that the processor, codenamed "Jalapeno," is part of OpenAI's multi-generational custom chip strategy. By simultaneously utilizing Nvidia GPUs and Google TPUs alongside developing its own custom chips, OpenAI stands to increase its flexibility in allocating computing power, though differences remain across chips in software ecosystems, model adaptation, and deployment efficiency.