OpenAI's hiring for Google chip engineers suggests a real TPU purchase is under way
OpenAI is reportedly moving from talk to action on Google's AI chips, which could loosen Nvidia's hold on one of its biggest customers.
The short version
- The Information reports OpenAI plans to buy Google's TPUs, its in-house AI chips.
- OpenAI is hiring engineers with TPU training and inference experience.
- Deal value and scale are not public, and neither company responded to the report.
- Google now rents TPUs to Anthropic, Apple and Citadel Securities, and sells to data center operators.
- Oracle will become the first major cloud provider to rent TPUs to customers.
Why it matters
AI progress depends heavily on who can get enough chips. If a top customer like OpenAI builds a real second source, Nvidia's pricing power and the supply squeeze may ease over time. That is a possibility, not a promise, so it should not shape your budget yet.
What was reported
As technews.tw reported, citing The Information and unnamed sources, OpenAI plans to buy tensor processing units (TPUs), the AI chips Google has developed over several years. The size and value of the deal have not been disclosed, and neither company answered the report.
The clue that the plan is concrete is a job posting. OpenAI is looking for engineers to lead the work of tuning its most advanced models to run on TPUs, and it wants people with experience in training, inference (running a finished model) and real-world workloads.
From renting to rebuilding
Hiring specialists suggests more than renting computing time. Software built for one chip family usually has to be adapted before it runs well on another, and that takes skilled people.
It also reverses an earlier position. When OpenAI was previously reported to be using TPUs, Nvidia stepped in and agreed a new arrangement, and an OpenAI spokesperson said there were no particular plans to use them.
Nvidia's strong position
Nvidia is hardly being dropped. It agreed to provide credit support worth well over a hundred billion dollars for a planned OpenAI data center in Ohio that will use only Nvidia chips. Its chief executive, Jensen Huang, has said OpenAI committed to buy about six hundred billion dollars of Nvidia servers by 2030.
Those commitments should cushion Nvidia in the short term. Still, Nvidia's GPUs are expensive and in tight supply, and customers looking for a second supplier is a trend that looks hard to reverse.
Google pushes TPUs outward
Google used to offer TPUs mainly through its own cloud, with customers such as Anthropic, Apple and Citadel Securities. It has recently begun selling directly to data center operators.
On the same day, Google said Oracle will be the first major cloud provider to rent TPUs to customers. Oracle is reportedly buying the chips directly. OpenAI is a major Oracle cloud customer, and it said last week it will serve its models through Google Cloud.
What to watch
- Will OpenAI or Google confirm the purchase and disclose its size?
- Does OpenAI's Nvidia commitment change as TPU work grows?
- Do more cloud providers follow Oracle in offering TPUs?
Sources and links
- Reportingtechnews.tw: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by technews.tw · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.