On October 28th, Oracle (ORCL.N) unveiled its plans to broaden its partnership with NVIDIA (NVDA.O) in the realm of cloud-based AI chips. The company intends to ramp up the deployment scale of AI servers, which are equipped with NVIDIA's cutting-edge Vera Rubin chips, reaching the gigawatt level. Simultaneously, Oracle will also incorporate Google's TPU chips, along with AI inference chips from SambaNova Systems and Cerebras Systems, into its lineup, thereby diversifying its computing power offerings. Furthermore, Oracle has plans to integrate its self-developed Acceleron network chips into select NVIDIA servers, potentially replacing some of NVIDIA's network products.

Oracle explained that, given the diverse computing power requirements for different AI workloads, a varied chip supply will empower customers to select the most suitable infrastructure, taking into account factors such as performance, throughput, and cost. While NVIDIA holds a dominant position in the AI training chip market, the AI inference market presents a more competitive landscape with a wider array of players.

Cerebras CEO Andrew Feldman remarked that the collaboration with Oracle would substantially broaden its customer base, particularly among large enterprises. He also mentioned that Cerebras is making significant strides in penetrating the data centers of other businesses. Although Oracle has not yet disclosed the financing details for this chip procurement initiative, it did mention last month that customers are increasingly shouldering the cost of chip procurement. Nevertheless, the company remains committed to its capital expenditure target, estimated at approximately $90 billion to $95 billion for the fiscal year 2027.