Oracle Expands NVIDIA Partnership and Adds Google TPU and Cerebras Chips

On October 28, Oracle said on Tuesday local time that it will expand its cloud AI chip partnership with Nvidia, while further advancing supply diversification by bringing in products from Nvidia competitors, including chips from Google, owned by Alphabet, and Cerebras Systems.

Oracle and Nvidia said Oracle will deploy Nvidia's latest-generation Vera Rubin AI servers at scale, with the deployment scale reaching "multiple gigawatts." The two sides did not disclose the specific amount of the deal. However, Nvidia CEO Jensen Huang previously said that each gigawatt of Vera Rubin system deployment could bring Nvidia up to $40 billion (note: approximately RMB 267.35 billion at current exchange rates) in revenue.

Oracle is also pushing supply diversification. The company said that even systems based on Nvidia chips will abandon some of Nvidia's networking chips in favor of Oracle's self-developed Acceleron networking chips.

In addition, Oracle also announced that it will for the first time offer Google's Tensor Processing Unit (TPU) chips to customers. Oracle will become the first hyperscale cloud provider other than Google to offer Google TPUs. The two sides said the service is expected to launch in 2027 and gradually expand in scale over multiple years, supporting large-model training, inference and agent applications. Google's eighth-generation TPUs include TPU 8t for training and TPU 8i for inference.

Mahesh Thiagarajan, president of Oracle's cloud infrastructure division, said in a statement: "Different AI workloads have different computing resource requirements." He noted that offering more choices allows customers to flexibly select appropriate infrastructure based on the specific performance, throughput and cost-efficiency requirements of various AI workloads.

In addition to Google chips, Oracle also plans to offer chips from Intel-backed SambaNova Systems and Cerebras Systems. Both companies focus on AI inference chips, the computing stage used when an AI system generates a response after receiving a request from a human user.

Cerebras CEO Andrew Feldman, speaking to Reuters about the partnership with Oracle, said: "This will significantly expand our tier-one customer base." He also said: "Our products are entering other companies' data centers through large-scale deployments."

Oracle plans to hold an investor day this week but has not yet disclosed how it will fund the procurement of these chips.

On last month's earnings call, Oracle said customers are increasingly bearing chip procurement costs. The company expects related spending to reach approximately $90 billion to $95 billion (approximately RMB 601.538 billion to RMB 634.956 billion at current exchange rates) in fiscal 2027.