NVIDIA DSX Delivers 24% More AI Token Throughput Within Fixed Power Budgets
Lambda validation and grid-flexibility deployment show measurable AI factory efficiency gains.
AI-generated editorial illustration.
NVIDIA announced results from GPU cloud provider Lambda validating its DSX MaxLPS software on HGX B200 servers, released at the AI Infra Summit on September 15, 2026. Running 19 nodes within the same power budget as 16 nodes at full power, Lambda reported 24% more cluster-wide token throughput and 23% better performance per watt. NVIDIA projects DSX MaxLPS could enable up to 40% more GPU capacity for Vera Rubin NVL72 factories within the same power budget, the company claims. Separately, NVIDIA's Eos AI factory has completed more than 200 automated demand-response events for Silicon Valley Power using Emerald AI's Conductor platform, reducing load from four megawatts to three without interrupting priority workloads. The first dedicated DSX Flex commercial deployment is planned for a 96-megawatt facility in Manassas, Virginia.
Source: NVIDIA News. Original announcement: September 15, 2026.