Oracle (NYSE: ORCL) will offer Google's tensor processing units, the chips Google designs to train and run AI models, to customers of Oracle Cloud Infrastructure (OCI) starting in calendar 2027, the two companies said on October 27. Oracle will supply the power, cooling and data center networking, and the chips will sit inside its own infrastructure.
The companies said OCI will be the first hyperscale cloud outside Google to offer TPUs. A day later, on the third-quarter earnings call of Google's parent, Alphabet (Nasdaq: GOOGL), Google Cloud CEO Thomas Kurian said Google had entered partnerships to deliver TPU systems with Oracle and with Crux AI, "with more in the works."
What OCI customers get, and when
Customers will be able to choose TPUs inside OCI from 2027, with capacity ramping up over several years, according to the release. A post on Google Cloud's blog, co-written by Karan Batta, senior vice president of Oracle Cloud Infrastructure, gives the practical detail. The TPUs will connect to Oracle's databases and storage over Oracle's Acceleron network and sit alongside the GPUs in OCI, so companies can fine-tune and run models next to the enterprise data they already keep there, without moving it to another cloud.
Code carries over too, according to the post: teams using open tools, such as Google's TorchTPU stack for PyTorch, will be able to send workloads to TPUs on Google Cloud or on Oracle's infrastructure without rewriting model code. Batta told DCD that TPUs on OCI would look and feel to customers the way they do on Google Cloud, and that "when Google launches a new TPU, we will have it on day one."
Neither company has said which TPU generation OCI will carry first, which regions will get it, what it will cost, or what Oracle is paying or committing. Batta called it a "multi-year, multi-gigawatt" partnership, but Oracle declined to tell DCD the size of the first deployment. Oracle's Investor Day release on October 28, which set its new long-term targets, does not mention Google or TPUs.
Side by side, here is what the companies have published about TPUs on each cloud, as of October 29:
TPUs | Google Cloud | OCI |
|---|---|---|
Available | Now | From 2027 |
Chips | Ironwood and earlier; 8t and 8i "coming soon" | Not named |
Hosted in | Google's data centers | Oracle's data centers |
Prices | Public, per chip-hour | Not disclosed |
Sources: Google Cloud's TPU page and TPU price list; the October 27 release.
Google's chips, outside Google's cloud
The deal is the latest step in a change Google began this year. On April 29, Alphabet CEO Sundar Pichai said Google would start delivering TPUs to "a select group of customers in their own data centers," according to Alphabet's transcript of the call. In May, Google said Blackstone was making an initial $5 billion equity commitment to a new TPU cloud set up as a joint venture with Google, with Google supplying the chips, software and services. And on Alphabet's July 22 call, chief financial officer Anat Ashkenazi said Google had delivered TPU systems to customer data centers for the first time in the second quarter.
Kurian described the change on the October 28 call:
In the past, you could use TPUs in our cloud.Thomas Kurian, CEO of Google Cloud, on Alphabet's earnings call, October 28
Google now also places TPU systems in customers' own data centers and is "broadening availability to other providers," Kurian said, naming Oracle and Crux AI, which he described as a joint venture with Blackstone. Putting TPU systems in third-party facilities and customer data centers widens Google's market, he said, including to companies whose data cannot leave their own premises, and reduces its need to build facilities in advance to house the systems.
Alphabet said sales of TPU systems were "a meaningful contributor" to growth at Google Cloud, whose third-quarter revenue rose 103% from a year earlier to $30.7 billion, according to Alphabet's results release.
