Emerald AI, NVIDIA and Nscale Cut AI Server Power Use 25% in Iceland Pilot

Emerald AI has demonstrated a 25% reduction in AI server rack power consumption during a pilot conducted with NVIDIA and Nscale at Nscale's AI computing facility in Iceland. Announced at NVIDIA GTC Berlin on 21 October 2026, the demonstration used Emerald Conductor software integrated with NVIDIA DSX power-management technology to adjust computing power consumption in response to a simulated utility demand signal. The systems maintained the reduced power level for 30 minutes before returning to normal operation, demonstrating how AI infrastructure could respond to electricity grid constraints.

 

The pilot addresses an escalating challenge facing Europe's burgeoning artificial intelligence infrastructure: securing adequate grid capacity to accommodate the growing energy demands of data centers. Typically, electricity networks are designed to manage peak demand, leaving excess capacity during off-peak times. By strategically reducing electricity consumption during peak demand, AI computing facilities could optimize existing grid infrastructure and potentially gain access to connections without requiring extensive upgrades to generation or transmission. However, the speed of obtaining faster grid connections would depend on utility requirements, local network conditions, and the terms of operational agreements.

 

Read more: Electric Plant Raises $11.2 Million to Launch Bombadil AI Model for Plant Health

 

The demonstration evaluated a 30-minute flexibility event consistent with EPRI Flex MOSAIC Class B parameters for grid-responsive data center operations. It also marked the first full integration of Emerald Conductor with NVIDIA DSX, a technology designed to optimize computing performance within available power limits. The test builds on six earlier Emerald AI demonstrations with NVIDIA and follows Nscale's previous validation of NVIDIA DSX MaxLPS for improving AI throughput under fixed power constraints. Together, these developments aim to establish how AI workloads can be managed dynamically to support both computing requirements and electricity grid flexibility.

 

Dr. Varun Sivaram, Founder and CEO of Emerald AI, said, "This pilot shows that AI computing can reduce its power use on request, which is what utilities need to see before they can connect AI factories faster." The findings suggest that flexible power management could help data center operators reduce peak electricity demand while supporting grid reliability. Although the demonstration achieved a measurable reduction in server rack power consumption, the announcement does not report verified carbon emissions savings or confirm that the pilot has resulted in accelerated grid connections.

 

Source: Emerald AI