AI NewsInfrastructureAnnouncement
Oracle will sell Google TPUs on Oracle Cloud Infrastructure
Oracle Cloud Infrastructure will host Google TPUs. It is the first time Google TPUs will be sold outside Google.

Image: Oracle
Why it mattersA team on OCI can soon pick TPUs for training or inference without moving its workload to Google Cloud. One more supplier means more price pressure on Nvidia graphics cards.
The only place to rent Google's Tensor Processing Units today is Google Cloud. That changes next year. Oracle and Google announced at Oracle AI World in Las Vegas that Oracle Cloud Infrastructure will offer Google TPUs. Oracle says it will be the first large cloud provider outside Google to sell them.
The service becomes available in calendar year 2027. Oracle says it will ramp up over several years to match customer demand. Oracle will provide the power, cooling and data-centre networking. Google will supply the chips. There is no earlier date on the press release and no capacity number.
What Oracle customers get
Google says its eighth-generation TPUs come in two shapes. The TPU 8t is for training. The TPU 8i is for inference. Google says the 8t gives almost three times the compute per pod of the previous generation. Google also says it cuts frontier model development cycles from months to weeks. Both figures are Google's own. There is no outside benchmark.
Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, said customers want to pair each workload with the right hardware. In plain terms, an Oracle customer can keep its systems on OCI. It can also train a model on TPUs if that is the cheaper choice. Amin Vahdat, Google's senior vice president for AI and infrastructure, said the deal gives more organisations access to TPUs. Both quotes are from the Oracle press release.
Google's own description of the chip is unchanged from its Google Cloud marketing. TPUs are built for dense models, mixture-of-experts models and long-context inference. The language in the Oracle release says nothing new about the hardware itself.
A second supplier of AI chips at scale
Nvidia's graphics cards are the default option for most AI compute today. Google TPUs on a second cloud give buyers a second supplier at scale. The Information reported that OpenAI also plans to buy Google TPUs, citing a person familiar with the matter. OpenAI has been one of Nvidia's biggest customers. The Oracle press release does not name OpenAI.
For teams building AI products, this will show up as more compute options at renewal time. A buyer can now ask Oracle for a quote on TPUs. Before this deal, that quote only came from Google. More sellers gives the buyer more room to bargain on price.
What is not clear yet
The press release gives no price. There is no public figure for how many TPUs OCI will have. There is no date more precise than "calendar year 2027" and no benchmark against Nvidia hardware. Oracle also did not say which TPU generation the launch will carry at the start. Any speed or price claim about TPUs on OCI should wait for the first customers to run their own numbers.
Source
Oracle press release. The Information reporting on OpenAI buying Google TPUs.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

