Google lets its TPU chips out of its own cloud as Oracle adds them to OCI from 2027

For the first time, Google will let another major cloud provider offer its in-house AI chips. Oracle and Google announced on Tuesday that Google's Tensor Processing Units (TPUs) will become available in Oracle Cloud Infrastructure (OCI). According to the companies, it is the first time TPUs will be offered on a hyperscale cloud platform outside Google. The chips arrive in 2027.

The news was announced at Oracle's AI World conference in Las Vegas, in a joint press release published at 17:07 Oslo time (12:07 ET) on 27 October. On the same day, Oracle also launched an expanded deal with OpenAI covering ChatGPT and Oracle's business applications.

Google's chips move into Oracle

TPUs are purpose-built accelerators that Google has developed for more than a decade to train and run AI models. According to the press release, OCI will be the first hyperscale cloud outside Google to offer them. Under the deal, Oracle will provide power, cooling and datacenter networking in its own facilities, while customers choose TPUs inside OCI alongside the other chip types Oracle offers.

The companies say the TPUs will support training and post-training of large models, including mixture-of-experts models, as well as production inference for reasoning, agentic and long-context applications.

The press release refers to Google's eighth TPU generation, which is split in two: TPU 8t for training and TPU 8i for inference. Google says TPU 8t delivers nearly three times the compute performance per pod of the previous generation. The release does not explicitly state which generations will be deployed in OCI.

Available from 2027, ramping over several years

TPUs on OCI will be available in calendar year 2027 and will ramp over a multi-year period to meet customer needs, the companies write. The announcement does not disclose pricing, capacity, which OCI regions will get the chips, or whether they will be available in Europe and Norway.

"Different AI workloads have different compute needs," Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, says in the press release. He says customers gain access to multiple leading AI compute architectures in one cloud and can choose infrastructure based on the performance and economics of each workload.

Amin Vahdat, chief technologist for AI and infrastructure at Google, says the next generation of AI will require a diverse range of compute, and that the deal gives more organizations access to TPU technology.

The race for AI chips intensifies

The deal is a clear sign of how scarce and sought-after AI compute has become. Oracle is building out OCI with several chip suppliers, and Google gets its TPUs in front of customers who do not use Google Cloud.

The companies do not present TPUs as a replacement for other chips, but as an additional option in OCI. How pricing will compare with the alternatives has not been disclosed.

What it means for leaders

For Norwegian organizations that train or run their own models, several chip types in the same cloud could mean a stronger negotiating position and less dependence on a single AI hardware supplier. The practical effect comes no earlier than 2027, however, and without disclosed pricing or European availability it is too early to plan around the offer.

What leaders can do now is ask their own cloud provider which AI chips are available in which regions, and what it costs to move workloads between them. Code written for one chip type does not always run equally well on another.

Sources and media

• Oracle: Oracle Brings Google TPUs to Oracle Cloud Infrastructure (press release)
• Google Cloud: Oracle Brings Google TPUs to Oracle Cloud Infrastructure
• Oracle Cloud on X about Google TPUs in OCI

The press release was published at 17:07 Oslo time on 27 October 2026, and Oracle Cloud's post on X followed at 17:40 Oslo time. All information in this article comes from these sources.

Illustration: AI-generated with Grok Bot / hogby.ai