Oracle expands cooperation with Nvidia AI chips and introduces Google TPU and other suppliers

Oracle (ORCL.US) announced an expansion of its cloud AI chip collaboration with NVIDIA (NVDA.US), planning to scale up the deployment of AI servers equipped with NVIDIA's next-generation Vera Rubin chips to several gigawatt levels. At the same time, it is introducing Google TPU chips, as well as AI inference chips from SambaNova Systems and Cerebras Systems, to diversify computing power supply. Oracle also plans to use its self-developed Acceleron network chips in some NVIDIA servers to replace certain NVIDIA network products. The company stated that different AI workloads have varying demands for computing power, and diversified chip supply helps customers select the appropriate infrastructure based on performance, throughput, and cost. NVIDIA holds a dominant position in the AI training chip market, but the AI inference market is more competitive. Cerebras CEO Andrew Feldman said the partnership with Oracle will significantly broaden its large customer base and that the company is entering other enterprises’ data centers on a large scale. Oracle has not yet disclosed the financing arrangements for this chip procurement. Last month, the company stated that customers are increasingly bearing the chip procurement costs and maintained its capital expenditure target for fiscal year 2027 at approximately $90 billion to $95 billion.