New agreements for "Oracle" with chip companies to expand AI capabilities

Oracle announced an expansion of its agreement with Nvidia to develop an AI computing architecture, with plans to make chips from Google and Cerebras available through its cloud services in 2027. The financial value of the agreement was not disclosed, and Oracle did not clarify how it will finance its new purchases.

Oracle Corporation announced on Tuesday an expansion of its agreement with Nvidia to develop a multi-gigawatt-scale AI computing architecture, alongside plans to make chips from competing companies, including Google and Cerebras, available through its cloud services during 2027. The two companies did not disclose the financial value of the expanded agreement, and Oracle did not clarify how it will finance its new chip purchases, amid its plans to increase its computing capacity dedicated to artificial intelligence. The expansion includes reliance on Nvidia's latest Vera Rubin servers, as well as the use of internally developed Oracle Axeleron networking chips, replacing some networking components provided by Nvidia. Oracle also intends to add processors from SambaNova, backed by Intel, as part of its efforts to expand options available to customers and run AI applications using technologies from different suppliers. These steps come at a time when Nvidia dominates the market for chips used in training AI models, while the inference market faces broader competition among companies developing specialized processors. Inference refers to running AI models after training to produce answers and execute user requests, a process different from the initial training phase that relies on processing large amounts of data. According to Reuters, Oracle maintained its capital expenditure target for fiscal year 2027 at around $90 to $95 billion, as announced during its results presentation.