Oracle Cloud to Offer Google TPU and Inference Chips from OpenAI, Cerebras, and Others

Oracle is broadening OCI’s AI infrastructure with Google TPUs and other chips while deepening its Nvidia partnership to give customers more workload choice.Wylie Wong,Chips and Hardware Writer

October 28, 20264 Min ReadOracle is broadening OCI’s AI infrastructure with Google TPUs, Cerebras, OpenAI, SambaNova, and Positron AI chips alongside Nvidia systems.Getty Images

Oracle is broadening the range of AI chips available through Oracle Cloud Infrastructure, saying it plans to deploy Google’s Tensor Processing Units (TPUs) as well as inference chips from OpenAI, Cerebras, SambaNova, and Positron AI. The move gives OCI customers more options for running AI workloads beyond Nvidia GPUs.

Oracle said Tuesday, Oct. 27, that it will become the first hyperscaler to offer TPUs outside Google Cloud.

The announcement, made at Oracle AI World in Las Vegas, came alongside an expansion of Oracle’s partnership with Nvidia. Oracle said Nvidia’s Vera Rubin NVL72 systems and Vera CPUs are now generally available on OCI. The company also said it plans to expand OCI Superclusters to multi-gigawatt scale using Nvidia Vera Rubin NVL72 systems.

Industry analysts said Oracle’s adoption of the additional AI accelerators gives customers more flexibility to match infrastructure to specific workload requirements. While Oracle Cloud has relied primarily on Nvidia GPUs, it also has some AMD GPU deployments, according to analysts.

“Oracle has been a leader in AI factory development using Nvidia systems, and their move today suggests that they’re aligning with the industry direction towards heterogeneous computing,” said Brendan Burke, research director at Futurum Group. “Oracle is working closely with partners like OpenAI to select the best-fit compute, including Cerebras for fast inference.”

“The usage of alternative accelerators suggests that there are unique advantages to each, and that it’s worthwhile to offer choice to customers overall,” Burke added.

Matt Kimball, vice president and principal analyst at Moor Insights & Strategy, said the announcement is an aggressive expansion by Oracle Cloud to support multiple AI chip architectures.

“This is Oracle trying to be the open silicon cloud – something for everybody,” Kimball said.

Burke noted that the challenge for Oracle will be proving that the new AI accelerators are competitive with Nvidia-based offerings and showing long-term commitment to them rather than treating them as limited experiments.

Oracle said Google TPUs for training and inference will be available on OCI in 2027, with adoption ramping over a multi-year period. Oracle also said it will deploy the Cerebras Wafer-Scale Engine for high-speed AI inference in 2027, while Positron AI’s Positron Atlas inference chip is now generally available on OCI.

Oracle also said Tuesday that it will deploy OpenAI’s Jalapeño inference accelerators, developed in partnership with Broadcom. OpenAI and Broadcom previously said they expect initial deployment of Jalapeño by the end of 2026, but Oracle did not specify when the chip will be available on OCI.

SambaNova also announced Tuesday that it is collaborating with Oracle to deploy its SN50 inference chips in OCI, though no availability date was given in the press release. The SN50 is SambaNova’s fifth-generation Reconfigurable Dataflow Unit chip, designed for agentic AI workloads.

SambaNova asserts that the chip delivers up to five times the speed of competing chips at a total cost of ownership that is three times lower. The company also said the SN50 draws an average of 20 kW per rack, which is low enough to scale in existing air-cooled data centers.

Burke said Google TPUs are highly competitive processors that have shown significant generational improvements.

With its eighth-generation TPU family, Google introduced two versions of the processor: TPU 8t for training and TPU 8i for inference. In a company blog post, Google said both versions deliver up to twice the performance per watt. Google also said TPU 8t provides up to 2.7x better performance per dollar, while TPU 8i provides up to 80% better performance per dollar than the previous generation.

Hyperscalers such as Google and AWS have found strategic value in building custom AI chips for their own infrastructure and cloud services. Now they also appear to see an opportunity to commercialize those chips more broadly.

AWS has reportedly considered selling its Trainium AI chip for use in third-party data centers. Oracle’s announcement suggests Google also sees a market for TPUs beyond its own cloud platform.

After weighing whether to reserve its compute capacity for internal teams or make it more broadly available, Google appears to have shifted toward external deployment, Burke said. He attributed that move in part to continued tightness in the AI compute market.

“There's higher value in making it available than using it all internally,” he said.

Kimball said Google’s decision to make TPUs available on OCI allows customers to use them without migrating their data to Google Cloud.

“This is about Google understanding the value that it’s delivering in its TPU and trying to bring that value to customers, regardless of where their data lives,” Kimball said. “Oracle Cloud has a lot of customer data and a lot of AI use cases being deployed in it, so it only makes sense that you would want to bring TPUs to where the data is.”

Wylie Wong is a journalist and freelance writer specializing in technology, business and sports. He previously worked at CNET, Computerworld and CRN and loves covering and learning about the advances and ever-changing dynamics of the technology industry. On the sports front, Wylie is co-author of Giants: Where Have You Gone, a where-are-they-now book on former San Francisco Giants. He previously launched and wrote a Giants blog for the San Jose Mercury News, and in recent years, has enjoyed writing about the intersection of technology and sports.See more from Wylie Wong

Aug 21, 2026

Oct 29, 2026

Data Center World POWER

Join the entire data center ecosystem September 21 – 23, 2026, at the Gaylord Texan Resort & Convention Center in Dallas, TX. Learn More