Ian Buck, NVIDIA's vice president of hyperscale and high-performance computing, spoke Tuesday at the AI Infra Summit, an event held at the Santa Clara Convention Center whose attendance has more than doubled year over year, growing from roughly 3,500 to over 8,000 participants. That surge reflects the industry's intensifying focus on infrastructure built to support generative AI workloads at scale.
A central theme of his remarks was the concept of the AI factory, a framing NVIDIA has been promoting for several quarters to describe data centers optimized specifically for token production rather than general-purpose computing. Buck detailed progress on the Vera Rubin and DSX platforms, positioned as successors to the current Blackwell architecture, highlighting improvements in tokens generated per watt of energy consumed — a metric the company is increasingly pushing as a standard benchmark for comparing AI infrastructure efficiency.
This emphasis on energy efficiency comes as power consumption at AI-focused data centers has become a growing concern for operators, utilities, and regulators alike. By framing efficiency gains in terms of watts rather than raw compute output, NVIDIA is signaling that upcoming chip generations aim to deliver more useful output without a proportional rise in energy costs for operators running these facilities.
While precise performance figures for Vera Rubin remain only partially disclosed, the presentation reinforces NVIDIA's broader strategy: positioning its platforms not merely as compute accelerators, but as components of infrastructure designed end-to-end for the industrial scale of generative AI deployment.