NVIDIA reports early production results for DSX AI Factory Platform, demonstrating capability for scaled deployment.
NVIDIA on September 15, 2026, published early production results for its DSX AI factory platform, including a Lambda validation that achieved 24% more token throughput within a fixed power budget and a utility demand-response deployment running at its Eos AI factory. The announcements coincided with the opening of the AI Infra Summit in Santa Clara, running September 15–17 at the Santa Clara Convention Center with more than 8,000 expected attendees. Ian Buck, NVIDIA's vice president of hyperscale and high-performance computing, made AI factory efficiency the centerpiece of his keynote, "Advancing Infrastructure for the Era of Agentic AI," with Lambda's validation results released the same day.
Lambda's results represent the first validation of DSX MaxLPS on NVIDIA HGX B200 GPU Servers. The GPU cloud provider, serving more than 10,000 customers from AI-native startups to hyperscalers, ran the software on a five-rack, 19-node cluster. By running 19 nodes within the same power budget as 16 nodes at full power, Lambda achieved 24% more cluster-wide token throughput, rising from roughly 4 million to 5 million tokens per second, while performance per watt improved 23%. "With our proof of concept, we believe we've moved beyond the limitation of fixed power budgets," said Dave Ward, president of cloud services at Lambda. "NVIDIA DSX MaxLPS paves the way to reclaiming stranded capacity and converting it into real-world usage, with significantly more compute density in the same footprint."
DSX MaxLPS monitors GPU and rack-level power consumption and reallocates headroom across nodes based on workload type, recovering capacity that static provisioning leaves stranded. Training and inference draw power differently; the software optimizes allocation in AI factories running both. Based on NVIDIA's projections, DSX MaxLPS can enable up to 40% more GPU capacity for next-generation Vera Rubin NVL72 AI factories within the same megawatt power budget in suitable deployment environments.
NVIDIA announced the DSX platform at GTC Taipei on May 31, 2026, combining open-source software libraries, APIs, reference designs, NVIDIA computing platforms and partner technologies into a single platform for AI factory design, deployment and operations. The suite spans DSX Reference Design, DSX Sim, DSX OS, DSX MaxLPS, DSX Flex and DSX Exchange, covering validated architectures, simulation, open modular operations software, power management, grid-signal orchestration and secure data exchange. Cloud partners CoreWeave, Crusoe, Firmus, IREN, Lambda, Nebius, Nscale and Yotta Data Services have been deploying DSX Sim, DSX MaxLPS and DSX OS, while manufacturers including Dell Technologies, HPE, Lenovo and Supermicro build DSX-ready systems. NVIDIA founder and CEO Jensen Huang has characterized the power constraint directly: "A one-gigawatt factory will never become a two-gigawatt factory." DSX provides infrastructure builders a complete playbook to simulate, validate and operate AI factories.
The September 15 post indicates that DSX is incorporating 800 VDC power architecture into its reference designs, designed to reduce conversion complexity, improve power delivery efficiency and support denser accelerated computing racks. GB200 NVL72 racks running direct liquid cooling carry roughly 120 kW of heat that must be removed before that power reaches compute. NVIDIA positions DSX Sim for use before the first rack is installed, DSX OS and DSX Exchange once a factory is running, and DSX Reference Designs as a validated starting architecture.
On an August evening, as temperatures and air-conditioning loads spiked, Silicon Valley Power, the municipally owned utility of Santa Clara, sent a signal to an AI factory to adjust power consumption. Emerald AI's Conductor platform executed a predefined workload hierarchy: lowest-priority jobs yielded, high-priority inference continued, and power fell from four megawatts to three, automatically and without operator involvement. Emerald AI founder and CEO Varun Sivaram witnessed the moment on Zoom with roughly forty observers, including his San Francisco team, data center engineers and utility staff. His head of product, Mansi Shah, likened the moment to a SpaceX rocket launch. Silicon Valley Power has since sent more than 200 demand signals to the factory, with Conductor responding in under a minute every time.
The facility is NVIDIA's Eos AI factory in Santa Clara, operating as a participant in Silicon Valley Power's Flexible Load Interconnect Program—the first commercial grid utility program designed to treat AI factories as dispatchable resources. The Santa Clara installation predates DSX Flex itself; Emerald AI Conductor is integrating into DSX Flex as the platform matures. Silicon Valley Power and Emerald AI announced the pilot on April 21, 2026, with the first site operating at commercial, multi-megawatt scale at a data center running NVIDIA AI workloads on advanced GPUs. Nico Procos, Silicon Valley Power's electric utility director, said the pilot would evaluate practical tools to protect reliability and affordability while supporting flexible planning for future load growth.
The first dedicated DSX Flex commercial deployment will be a 96-megawatt Vera Rubin AI factory at NVIDIA's AI Factory Research Center in Manassas, Virginia. Emerald AI has indicated the Manassas project is planned in collaboration with Digital Realty, EPRI and the PJM Interconnection, scheduled for later in 2026.