Emerald AI and Nvidia cut AI server power use by 25 percent in Iceland pilot

Results from the Iceland pilot

Emerald AI recently shared the results of a pilot program conducted at an Nscale data center in Iceland. The project involved a collaboration between Emerald AI, Nvidia, and Nscale to test new methods for managing energy in high-performance computing environments. The companies unveiled these findings during Nvidia’s GTC event held in Berlin this week.

During the test, the Emerald Conductor platform worked alongside Nvidia’s DSX power-management technology. The goal of this integration was to determine how effectively the two systems could coordinate to lower the electricity demands of AI server racks. The results showed that the combined technology was able to reduce power usage by approximately 25 percent.

The pilot demonstrated that the system could maintain this lower power state for a duration of 30 minutes. After this period, the servers were able to return to their normal operating levels. This test represented the seventh pilot for Emerald AI, following previous trials in the United States and the United Kingdom.

How the orchestration works

The Emerald Conductor platform functions as a mediator between the electrical grid and the data center. It is designed to orchestrate AI workloads in real time, allowing for a dynamic adjustment of how much energy the facility consumes at any given moment. This mediation helps the data center maintain a balance between high-performance computing and the limitations of the local power supply.

By using this orchestration, data centers can respond to the needs of the utility provider while still performing necessary tasks. The system ensures that AI compute performance remains at an acceptable level while the facility scales its energy intake up or down. This flexibility is intended to support overall grid stability, which is often strained by the massive power requirements of modern AI factories.

The specific integration with Nvidia’s DSX technology is a key component of this approach. DSX is a power-management technology created by Nvidia to help data centers extract the maximum amount of compute power from every watt of energy used. Together, these tools allow for a more granular control over the energy profile of an AI server rack.

Addressing the power bottleneck

The companies are positioning this technology as a solution to one of the biggest hurdles in the current AI build-out. Varun Sivaram, the founder and CEO of Emerald AI, noted that access to power has become the primary constraint for building AI infrastructure in Europe. He argued that the fastest way to solve this is by making AI factories flexible enough to use the grid capacity that is already available.

By making AI factories more adaptable, the technology aims to keep energy costs affordable for local communities. Sivaram stated that the pilot shows AI computing can reduce its power use on request. This ability to adjust consumption is what utilities need to see before they can connect AI factories faster.

This dynamic adjustment is intended to alleviate the pressure on existing infrastructure. By proving that these facilities can voluntarily throttle their power intake during peak times, developers can potentially bypass some of the regulatory and physical hurdles that currently delay the construction of new data centers.

Funding and commercial rollout

Emerald AI is moving toward a wider commercial release of its orchestration software. The company announced that its first commercial project will begin in October at Nvidia’s Aurora data center in Manassas, Virginia. This facility is currently under construction and has a capacity of 96MW. The deployment will mark a significant step in moving the technology from a pilot phase to a production environment.

To support this growth, the firm recently secured $150 million in Series A funding. This latest round of investment follows several previous funding stages, including a $25 million seed extension in February and a $24.5 million raise in July. These funds are intended to help the company scale its operations and expand its reach in the data center market.

The company also maintains several high-profile partnerships and endorsements. It is a member of the Nvidia Inception program, which supports startups, and it receives backing from Kate Brandt, who serves as the chief sustainability officer at Google. These relationships highlight the interest from major tech players in finding sustainable ways to power the next generation of AI.

The path forward for AI infrastructure

The success of the Iceland pilot provides a blueprint for how AI data centers might operate in the future. Instead of being static consumers of electricity, these facilities could become dynamic participants in the energy grid. This shift is necessary as the demand for AI compute continues to outpace the construction of new power plants and transmission lines.

As more data centers are built, the integration of orchestration software like Emerald Conductor will likely become a standard requirement. It allows developers and operators to maximize their hardware's utility without overwhelming the local infrastructure. This balance is essential for the long-term viability of large-scale AI training and inference.

Moving forward, the industry will watch how these systems perform at scale in diverse geographic regions. The upcoming deployment in Virginia will provide critical data on how the software handles a commercial-grade workload. These results will determine how quickly utilities might begin approving new AI projects based on the promise of flexible power consumption.