One of NVIDIA's most important customers is placing orders with its competitor.
On October 27, according to The Information citing people familiar with the matter, OpenAI plans to purchase Google's Tensor Processing Unit (TPU) chips.
This represents a significant breakthrough for Google's chip business. Google TPUs have long been viewed as a potential challenger to NVIDIA GPUs in the AI computing power market, and OpenAI is one of NVIDIA's most valuable customers.
Meanwhile, OpenAI has posted a job listing seeking an engineer to "lead" the optimization of its most advanced frontier models on TPUs — indicating that related plans have entered a substantive advancement phase.
NVIDIA and OpenAI: Deep Ties Worth Hundreds of Billions of Dollars
OpenAI and NVIDIA share an extremely close relationship.
In August of this year, NVIDIA agreed to provide over $100 billion in credit support for OpenAI's planned large-scale data center project in Ohio, which will exclusively use NVIDIA chip architecture.
NVIDIA CEO Jensen Huang wrote in a blog at the time that OpenAI has committed to spending approximately $600 billion on NVIDIA servers by 2030.
However, this has not prevented OpenAI from seeking diversified supply.
Not the First Attempt, But This Time Going Further
According to a report by The Information last year, OpenAI had already begun using Google TPUs to reduce its dependence on NVIDIA chips. Upon the news breaking, Jensen Huang quickly took action, and NVIDIA and OpenAI subsequently reached a new agreement. An OpenAI spokesperson stated at the time that the company "currently has no plans to specifically use TPUs."
Now, that statement has become a thing of the past.
The job listing posted by OpenAI this time indicates that the related work is still at an early stage. The job description states that the position will have "broad technical leadership" over the entire optimization effort, and notes that "experience in TPU training or inference is welcome, along with a solid foundation in optimizing real workloads on TPUs."
Google TPU: From Internal Use to External Expansion
Google has been developing TPUs for over a decade and claims they are more efficient than NVIDIA GPUs.
For years, Google has rented TPUs to external customers through its own cloud infrastructure, with users including Anthropic (which has used TPUs to train models), Apple, and trading firms like Citadel Securities, as well as numerous startups. Recently, Google began selling TPUs directly to other data center operators, initially targeting emerging cloud service providers such as Fluidstack.
On the same day, Google announced that Oracle will become the first major cloud service provider to rent TPUs to customers. According to people familiar with the matter, Oracle is purchasing these chips directly, and the end user will be a major shared customer of both companies, though the specific identity remains unclear.
Notably, OpenAI itself is an important cloud customer of Oracle, having just announced last week that it will offer its model services through Google Cloud, and had previously been renting NVIDIA chips through Google Cloud.
Comments