OpenAI plans to purchase Google’s tensor processing units (TPUs) to reduce its reliance on Nvidia, according to a report by The Information citing a person familiar with the matter. Neither company has confirmed the transaction, and specific details regarding price, volume, and delivery timelines remain undisclosed. This move follows reports from June 2025 that OpenAI had begun renting Google’s TPUs for early testing, though the company previously stated it had no plans for large-scale deployment at that time. A recent job listing in San Francisco indicates OpenAI is hiring an engineer to optimize its frontier models for TPU architecture, signaling active preparation for hardware integration.
The potential acquisition aligns with OpenAI’s broader strategy to diversify its compute infrastructure. In September 2025, Nvidia signed a letter of intent to invest up to $100 billion in OpenAI, contingent on deploying at least 10 gigawatts of Nvidia systems. However, Nvidia ultimately contributed $30 billion to OpenAI’s funding round, significantly below the initial proposal. Concurrently, AMD granted OpenAI the right to purchase up to 160 million shares, roughly 10% of the company, tied to the deployment of 6 gigawatts of AMD chips. OpenAI also unveiled Jalapeño, its first custom chip built with Broadcom, designed for inference tasks with initial deployment planned for the end of 2026.
Google has been expanding its TPU offerings, unveiling two eighth-generation units in April: the 8t for training and the 8i for inference. Alphabet’s CFO expects most revenue from TPU system sales to materialize in 2027, noting that Google began delivering these systems to customers’ data centers in the second quarter. Meanwhile, competitor Anthropic secured an agreement with Google and Broadcom in April for multiple gigawatts of next-generation TPU capacity, expected to come online starting in 2027.
OpenAI's reported shift toward purchasing Google TPUs represents a strategic effort to mitigate supply chain concentration risks associated with Nvidia. By moving from rental agreements to ownership, OpenAI seeks greater control over its computational resources and cost structures. The existence of specialized engineering roles focused on TPU optimization underscores the technical complexity involved in migrating frontier models away from Nvidia's CUDA ecosystem, suggesting this is not merely a procurement decision but a significant architectural pivot.
This development highlights intensifying competition among chipmakers to secure anchor tenants for their AI hardware platforms. While Nvidia remains dominant, its reduced investment commitment compared to the initial letter of intent may have prompted OpenAI to accelerate diversification. The parallel moves by Anthropic to secure Google TPU capacity indicate a broader industry trend where leading AI labs are hedging against single-vendor dependency. For Google, securing OpenAI as a direct customer would validate its TPU roadmap and potentially disrupt Nvidia's market share in high-performance AI training clusters.
