Google’s custom AI chips are beginning to evolve from a cloud-only capability into a commercial hardware business, executives said during its quarterly earnings call.
Some lucky customers took delivery of their own Tensor Processing Units (TPUs), custom chips developed by Google for AI applications, in the second quarter, company executives disclosed during a call to discuss its latest financial results on Wednesday.
The disclosure comes at a time when enterprises are grappling with a global shortage of high-end GPUs that is driving up costs and slowing AI deployment timelines, and even those who have GPUs don’t always have electricity to operate them.
“We delivered to customer data centers for the first time in Q2,” said Anat Ashkenazi, CFO of Google’s parent Alphabet, adding that the company had begun to recognize a small amount of revenue from TPU orders, although it expected to recognize more in 2027.
Ashkenazi did not name the customers or describe the commercial arrangements, leaving open questions about whether the systems are being deployed by large enterprises, governments, sovereign AI initiatives, or other cloud providers.
Until recently, Google’s TPU strategy centered on giving customers access to the chips through Google Cloud, while using the same silicon internally to power its own AI models and services. But during the company’s last earnings call in April, just days after it released two new TPU models, CEO Sundar Pichai said Google would consider selling TPUs to AI labs, capital markets firms, and high-performance computing applications.
On Wednesday’s call, Pichai said Google will scale up TPU sales “based on the opportunities we see and the demand we see, commensurate with the constraints that exist and the allocation needs we have for frontier model development.”
But Google is still keeping the bulk of its TPUs for itself, either for internal use or to rent out through Google Cloud Platform.
“Our first priority is making sure we are allocating what we need to compete at the frontier in terms of AGI development,” Pichai said on Wednesday’s call. “We are using both TPUs and GPUs mainly for serving our models.”
