Emerald AI and Nvidia cut Iceland AI rack power 25% in pilot

AI server racks at Nscale's data center in Iceland cut their power consumption by approximately 25 percent during a pilot run by Emerald AI and Nvidia, with the reduction held for 30 minutes before operations returned to normal.

The companies announced the results during Nvidia's GTC event in Berlin this week. Emerald AI's Emerald Conductor platform ran together with Nvidia's DSX power-management technology to deliver the reduction. The test marked the first time Emerald Conductor was fully integrated with DSX, which Nvidia designed to help AI data centers extract as much compute as possible from every watt of energy.

Varun Sivaram, founder and CEO of Emerald AI, said access to power now defines the limits of Europe's AI expansion. He said: "Access to power has become the main constraint on Europe's AI build-out, and the fastest way to relieve it is to make AI factories flexible enough to use the grid capacity that already exists, which also keeps energy affordable for local communities."

He added: "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."

What does the pilot change for grid operators?

Emerald Conductor acts as a mediator between the grid and data centers, orchestrating AI workloads in real time. Data centers can dynamically adjust their energy consumption this way, supporting grid stability while keeping AI compute performance at acceptable levels.

The Iceland test carries weight for utilities because it demonstrates demand response on request. If AI facilities can shed load when grids are strained, utilities gain a basis for connecting those facilities faster — the core of Sivaram's argument.

The pilot is the seventh Emerald AI has undertaken. The full list of tests:

• Three in the US: Santa Clara, Phoenix, and Chicago
• One in the UK, in partnership with the National Grid
• The Iceland pilot with Nvidia and Nscale, announced at GTC Berlin

In plain terms: the company now has seven field demonstrations across three countries.

Where does the commercial rollout stand?

Emerald AI announced its first commercial project in October. The orchestration software is set for deployment at Nvidia's Aurora data center in Manassas, Virginia — a 96MW facility currently under construction.

The Iceland pilot strengthens that pipeline. The 25 percent reduction, sustained for 30 minutes, gives the Aurora deployment a validated reference point.

How is the company funded?

Emerald AI raised $150 million in Series A funding in August, its fourth funding round. Earlier raises:

• $25 million in seed extension funding in February
• $24.5 million raised the previous July
• An $18 million boost to that round in October

The company is a member of the Nvidia Inception program for startups. Kate Brandt, Google's chief sustainability officer, backs the firm.

The takeaway: a 25 percent power cut on request, demonstrated in production hardware, is now Emerald AI's pitch to utilities across Europe.