Oracle Will Offer Google’s AI Chips From 2027

LAS VEGAS, Oct. 27 — OracleOracle and Google said on Tuesday that Google’s custom AI chips, its Tensor Processing Units, or TPUs, will be available through Oracle Cloud Infrastructure in calendar 2027, “ramping over a multi-year period to meet customer needs.” Oracle’s release called it “the first time Google TPUs will be offered on a hyperscale cloud platform outside Google.”

Oracle will provide “the power, cooling, and datacenter networking” to run the chips at scale inside its own infrastructure. Neither company gave a price or a number of chips, or said who will own them.

The arrangement puts Google’s chip on offer in another company’s cloud, beside the Nvidia systems Oracle already rents out, and lets Google reach Oracle’s customers without supplying the buildings or the power. Google rents out TPUs on its own cloud, and has begun selling whole systems: Alphabet said in its report for the second quarter that it had signed “a limited number of agreements to supply TPU systems to customers who require or provide on-premises infrastructure for specialized, high-scale workloads,” and had begun recognizing revenue from them, with “the significant majority” to be recognized in 2027.The TPUGoogle’s Tensor Processing Units are custom accelerators for AI work; Google says it built its first more than ten years ago. The eighth generation comes in two designs, the TPU 8t for training models and the TPU 8i for running them. The announcements do not say which Oracle will offer.

“Even as TPU adoption on Google Cloud grows, the global demand for AI compute exceeds any single cloud provider’s footprint,” the two companies wrote in a joint post on Google Cloud’s blog.

“Different AI workloads have different compute needs,” Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, said in the release. “Bringing Google TPUs to OCI gives customers access to multiple leading AI compute architectures in one cloud.” For Google, Amin Vahdat, its senior vice president and chief technologist for AI and infrastructure, said: “The next generation of AI will require a diverse range of compute.”

The post leaned on software. It pointed to TorchTPU, Google’s open-source software for running PyTorch, a widely used AI toolkit, on TPUs, and said such portability lets teams “assign each workload to the accelerator that delivers the lowest cost per token or fastest time to train, without rewriting model code.” That is the companies’ claim.

Nvidia, Too

Nvidia had news of its own at the conference: it and Oracle announced plans to expand Oracle’s Superclusters with Nvidia’s Vera Rubin NVL72 systems “to multi-gigawatt scale,” and said those systems and Nvidia’s Vera CPUs were now generally available on Oracle’s cloud. They gave no gigawatt figure or timetable for the expansion.The NVL72Nvidia’s rack-scale system: each rack connects 72 GPUs through Nvidia’s NVLink, and Oracle’s network links racks across hundreds of thousands of GPUs, the release says. Oracle already runs Nvidia’s GB200 and GB300 NVL72 systems.

The release cited OpenAI, which it said has trained and served GPT-6 Astra on an OCI Supercluster in Abilene, Tex., with more than 100,000 Nvidia Blackwell GPUs. The same day, OpenAI and Oracle widened their partnership: Oracle is developing Oracle AI Workspace to bring its applications and data into ChatGPT, and is building new products on OpenAI’s models, among them a security service that uses “OpenAI cyber models.” Oracle’s customers in the United States have been able to buy OpenAI’s API, Codex and ChatGPT Work through Oracle Marketplace since August.

Oracle’s shares closed Tuesday at $130.53, up 3.55 percent, and Alphabet’s Class A shares at $320.58, up 1.19 percent.ORCLNYSE130.67▼8.89−6.4%Oracle · Friday’s session

Who Pays

Oracle supplies the sites, and its build-out is costly. In its fiscal first quarter, which ended Aug. 31, Oracle spent $28.5 billion on capital expenditures, its free cash flow, a measure it reports outside generally accepted accounting principles, was negative $5.4 billion, and it sold $20 billion of common stock as part of what it calls its capital investment program. Alphabet, for its part, warned investors in July that supplying TPU systems “may increase our costs and operational complexity.”

Oracle holds an investor day at 11:55 a.m. Eastern today. Alphabet reports its third quarter after the market closes, with a call at 4:30 p.m., on a night when Microsoft reports too.

Daily Cortex is written with AI models, including Anthropic’s Claude, and fact-checked with OpenAI’s GPT models.