OpenAI Plans to Purchase Google TPUs in Bid to Diversify Chip Supply
nashnova research
OpenAI is planning to purchase Google's TPUs, directly challenging Nvidia's lock on top-tier AI customers — one of its most important buyers is now splitting orders with a rival.
What does this deal actually involve?
According to The Information, citing people familiar with the matter, OpenAI plans to buy Google's Tensor Processing Units (TPUs) — custom AI chips Google designed for training and running large models.
Specific financial terms and deal details remain unclear; the arrangement is still at the planning stage.
This means → Google's TPU business would land a heavyweight new client, while OpenAI further reduces its reliance on Nvidia for compute.
How tightly were OpenAI and Nvidia tied before?
OpenAI has been one of Nvidia's most important customers. Last August Nvidia announced over $100 billion in credit support for OpenAI's large data-center project in Ohio.
Nvidia CEO Jensen Huang said at the time that OpenAI had committed roughly $600 billion in spending on Nvidia servers through 2030.
In plain terms = the two were deeply bound together. OpenAI handing even a slice of its compute orders to Google sends a signal far larger than the dollar amount itself.
How far along is OpenAI's TPU effort?
OpenAI has posted a job listing for an engineer to "lead" optimization of its frontier models on TPUs, describing the role as carrying "broad technical ownership."
This reflects the fact that OpenAI's TPU optimization work is still in its early stages — the core engineer has not even been hired yet.
According to earlier reporting, OpenAI began experimenting with Google TPUs last year, but a spokesperson later said the company had "no specific plans to use TPUs." If this deal goes through, that stance will have materially shifted.
Is Google's TPU customer base expanding?
This week Oracle announced it would become the first major cloud provider to offer TPU access outside Google's own cloud, reportedly buying the chips directly and serving joint customers.
Anthropic, Apple, and Citadel Securities are all existing Google TPU users.
This means → Google's TPU is moving from an in-house tool to an externally sold product, and its client list increasingly reads like a roster of the AI industry's core players.
What does this mean for Nvidia?
OpenAI is also a major Oracle cloud client and last week announced it would make its models available on Google Cloud — its compute procurement is diversifying across the board.
In plain terms = Nvidia's GPUs remain the dominant force in AI training. But once your largest customer starts using a competitor's chips too, the monopoly has already begun to loosen.
This signals a broader shift in the AI compute market — from "Nvidia dominance" toward "multi-supplier coexistence" — a trend that puts long-term pressure on Nvidia's pricing power and bargaining position.
市场有风险，内容仅供研究参考，不构成投资建议。