Oracle expands Nvidia deal and adds Google, Cerebras chips

By Marcus Bennett

Oracle will expand its Nvidia cloud AI chip deal to “multi-gigawatt scale,” the company said on Tuesday, Oct. 27, without disclosing the contract value. The rollout centers on Nvidia's newest generation of “Vera Rubin” AI servers.

Oracle and Nvidia both framed the announcement around capacity rather than contract value. Nvidia CEO Jensen Huang has previously said that Vera Rubin systems can generate up to $40 billion in revenue per gigawatt for the chipmaker.

What changes in Oracle's AI chip portfolio?

Oracle will offer Google's tensor processing unit (TPU) chips for the first time. It also plans to add chips from Intel-backed SambaNova Systems and Cerebras Systems.

The additions broaden Oracle's line-up across three suppliers:

• Google TPU chips, offered through Oracle for the first time.
• SambaNova chips, from a company backed by Intel.
• Cerebras chips, aimed at the inference market.

Oracle's Nvidia-based systems will skip some Nvidia networking chips. The company will use its own in-house “Acceleron” networking chips instead.

That substitution changes the supplier mix inside each system. It also gives Oracle a proprietary component it controls as it assembles AI infrastructure for cloud customers.

The result is a mixed hardware strategy. Customers can buy Nvidia's latest generation, Google accelerators, or chips from two alternative suppliers through the same cloud platform.

Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, said: “Different AI workloads have different compute needs.” He said the expanded choice lets customers select infrastructure based on performance, throughput and economics for each workload.

Why does Oracle add inference chips?

Nvidia dominates the market for training AI systems. Inference, where AI systems respond to requests from human users, has a more competitive and diverse supplier base.

Oracle is therefore extending its Nvidia capacity while adding options in the segment with more competing suppliers. SambaNova and Cerebras both make chips for inference work.

Cerebras CEO Andrew Feldman said: “That's a massive expansion of our customer base of tier-one customers.” Feldman also said Cerebras systems will run at large scale in other people's data centers.

Oracle's agreement thus reaches beyond the infrastructure the company operates directly. The arrangement gives Cerebras access to large customers through Oracle's cloud, while Oracle adds capacity from a chipmaker outside its Nvidia and Google line.

How will Oracle finance the chip purchases?

Oracle plans to hold an investor day this week. The company did not disclose how it will finance the chip purchases.

During an earnings call last month, Oracle said customers are increasingly footing the bill for the chips. That points to a model in which some purchase costs shift to the customers who use the capacity.

Oracle maintained its capital spending target of about $90 billion to $95 billion for its 2027 fiscal year. The target remains unchanged despite the expanded chip sourcing plans.

Oracle's fiscal 2027 budget therefore remains the clearest financial guide in the announcement. The absence of a financing plan leaves the balance between company capital and customer-funded purchases unresolved.

What does this change for cloud customers?

Customers gain more hardware choices for AI workloads through Oracle's cloud infrastructure. The menu now includes Nvidia Vera Rubin systems, Google TPUs, SambaNova chips and Cerebras chips, alongside Oracle's Acceleron networking.

Buyers can match hardware to workload requirements. They can also place Cerebras deployments in other organizations' data centers, according to Feldman.

For Oracle, the expansion raises the scale of its Nvidia commitment while diversifying suppliers. The commercial value of the Nvidia agreement and the financing plan for the chip purchases remain undisclosed.