NVIDIA CEO Jensen Huang speaks at the GTC Taipei keynote on June 1. (Screenshot from livestream)
NVIDIA unveiled a new generation of computing hardware and software at its GTC Taipei keynote on Monday, aiming to generate more output from fixed power budgets and positioning energy efficiency as a key driver of AI infrastructure economics.
The announcements included the Vera CPU, the Vera Rubin multi-rack system, and the DSX AI factory platform. While each targets a different layer of the AI stack, they share a common objective of helping operators maximize AI output within increasingly constrained electricity supplies.
With individual AI factories expected to cost between USD 50 billion and USD 100 billion, operators are increasingly focused on maximizing output from fixed power allocations. Every additional token generated from the same amount of electricity translates directly into revenue.
“If your factory has one gigawatt, it will not have more,” CEO Jensen Huang said during the keynote. “Performance per watt is your revenues.”
Unlock the full article to explore three key takeaways:
- NVIDIA's new DSX platform aims to improve AI factory efficiency by reducing unused power capacity, deploying more GPUs within existing power limits, introducing 45°C hot liquid cooling, and enabling facilities to adjust electricity consumption in response to grid conditions.
- The Vera CPU is designed for agentic AI workloads, where tasks such as code execution, database access, and tool use increasingly create bottlenecks that limit GPU utilization and overall AI factory productivity.
- NVIDIA projects that 100 GW of AI factory capacity will come online globally before the end of the decade, making power management and energy efficiency increasingly important to the economics of AI infrastructure investment.
