Nvidia Unveils Vera Rubin Hardware for AI Factories

CEO Jensen Huang announced new hardware and infrastructure strategies during the GTC Conference in Berlin, introducing server racks designed to transform data centers into AI factories. The new systems feature 72 next-generation Vera Rubin GPUs and 256 custom language processing unit chips.

By integrating Vera family CPUs, Nvidia intends to expand its monetization of AI compute applications, with a specific focus on AI inference to lower energy costs and increase efficiency. Huang noted that the sales opportunity per gigawatt of compute has increased to $40 billion with the Vera Rubin family, compared to $25 billion with the previous Blackwell architecture.

Huang estimated that establishing a 1-gigawatt AI factory would require an investment of approximately $50 billion but could generate $40 billion in annual revenue. He warned that a three-month delay in the deployment of these systems could result in $10 billion in lost potential sales.