Oracle ( Oracle ) said on Tuesday local time that it will expand its cloud-based AI chip collaboration with NVIDIA ( Nvidia ), and at the same time further promote supply diversification by introducing products from NVIDIA's competitors, including those from Google, a subsidiary of Google's parent company Alphabet, as well as chips from Cerebras Systems.

Oracle and NVIDIA announced that Oracle will deploy NVIDIA's latest generation of Vera Rubin AI servers on a large scale, with the deployment reaching "gigawatt levels." The two parties did not disclose the specific amount of this transaction. However, NVIDIA CEO Jensen Huang previously stated that each gigawatt of Vera Rubin system deployment could generate up to $40 billion (Note from the editor: current exchange rate is approximately 267.35 billion RMB) in revenue for NVIDIA.

Oracle is also pushing for diversification in supply. The company stated that even systems based on NVIDIA chips will abandon the use of some of NVIDIA's networking chips and switch to Oracle's own independently developed Acceleron networking chips.

In addition, Oracle also announced that it will for the first time provide customers with Google's Tensor Processing Unit (TPU) chips. Oracle will become the first hyperscale cloud service provider other than Google to offer Google's (TPU). Both parties stated that this service is expected to be launched in 2027 and will gradually scale up over the years to support large model training, inference, and agent applications. Google's eighth-generation (TPU) includes the (TPU) 8T for training and the (TPU) 8I for inference.

Mahesh Tiagarajan, President of Oracle's Cloud Infrastructure Division ( Mahesh Thiagarajan ), stated in a declaration: "Different AI workloads have different requirements for computing resources." He pointed out that offering more options allows customers to flexibly choose the appropriate infrastructure based on the specific requirements of various AI workloads in terms of performance, throughput, and cost-effectiveness.

In addition to Google's chips, Oracle also plans to provide chips that support SambaNova Systems and Cerebras Systems. Both companies focus on AI inference chips, which are the computational components used by AI systems to generate responses after receiving requests from human users.

Currently, NVIDIA holds a dominant position in the AI model training market, but the competitive landscape in the AI inference market is more diversified, with a larger number of participants.

Cerebras CEO Andrew Feldman ( Andrew Feldman ) said in an interview with Reuters regarding the collaboration with Oracle: "This will significantly expand our base of primary customers." He also stated, "Our products are being deployed on a large scale in the data centers of other companies."

Oracle plans to hold an Investor Day event this week, but has not yet disclosed how it will fund the procurement of these chips.

At last month's financial report conference call, Oracle stated that customers are increasingly bearing the costs of chip procurement. The company expects that related expenditures for the fiscal year 2027 will reach approximately $90 billion to $95 billion (equivalent to about 601.538 billion to 634.956 billion yuan at current exchange rates).