Innovation is at the heart of what we do at National Grid and a recent UK first trial, demonstrating how AI can help manage data centre demand, is just another example of this.
National Grid Partners portfolio company, Emerald AI is showing how AI-powered data centres can enhance the UK’s electricity grid and speed up the connection of new data centres to the grid.
Together with collaborators on the project, EPRI, Nebius and NVIDIA, the team used Emerald’s technology in a live trial in December 2025 at a data centre in London. This dynamically adjusted the energy consumption of Nebius’ data centre in real time using NVIDIA graphics processing units (GPU)s. The aim: to optimise energy use during peak demand, unlocking capacity and speeding up connections to the network.
The importance of power-flexible data centres
Currently, most large data centres have fixed, “always on” demand. As AI adoption grows and more facilities seek to connect to the grid, that fixed-demand model risks increasing network constraints and lengthening connection times.
Demonstrating that data centres can flex their power usage shows how they can ease constraints, unlock faster and higher-capacity connections, and support both AI growth and low-carbon power on the grid.
A live trial of AI grid technology in the UK
The technology tested in the London trial was Emerald AI’s software, called Emerald Conductor. This managed a cluster of 96 NVIDIA Blackwell Ultra GPUs to respond to real-time grid signals. The trial lasted five days during December 2025 and more than 200 real-time simulated grid events were sent to the site to test responses across a range of scenarios.
The objective was to validate that AI data centres can adjust power consumption quickly and reliably while critical AI workloads continue to run. So how did it work out?
Reducing power demand by a third in under 60 seconds
Results were impressive; the platform cut electricity demand by more than a third in under a minute in multiple tests, with reductions up to 40%, all while workloads continued as normal.
The test did not stop there; it also looked at emergency response. In simulated lightning strikes or plant failures, the system shed around 30% of load in under 40 seconds to help prevent blackouts.
Another benefit was peak smoothing: the software successfully reacted to sudden demand spikes (for example during major broadcast events), demonstrating its ability to counterbalance consumer electricity surges.
The trial also followed load-reduction requests for up to 10 hours, supporting grid operation during prolonged stress periods such as low wind or extreme heat.
The data showed:
- Reliability: Emerald Conductor met requested power adjustments in every simulated event in the trial.
- Scale potential: as the UK prepares for over 6 GW of data centre deployments by 2030, the partners estimate that power-flexible AI data centres could make more than 2 GW of capacity available back to the grid when needed.
- Practical value: the trial demonstrates a pathway for AI infrastructure to move from a potential source of constraint to a controllable grid asset that helps manage peaks and improve use of existing network capacity.
Widespread potential positive impacts
Emerald’s tech can bring a number of benefits, including:
- Improved grid resilience: flexible demand helps the system manage peak events and emergency conditions more effectively.
- A reduced need for upgrades: by unlocking capacity through demand flexibility, the approach can reduce the requirement for costly and time-consuming grid reinforcement.
- Support for low-carbon power: enabling more flexible demand makes it easier to integrate variable renewables and manage periods of low generation.
- Economic and growth benefits: faster and higher-capacity connections for data centres can support the UK’s AI ecosystem while managing impacts on consumers.
Steve Smith, our Group Chief Strategy and Regulation Officer, said the demonstration shows data centres can actively support the grid by flexing demand in real time, easing pressure at peak moments and unlocking much-needed capacity.
Dr Varun Sivaram, Founder and CEO of Emerald AI, meanwhile, highlighted that the trial validated fast emergency curtailment and sustained, precise peak reduction across dozens of realistic AI workloads and that the same approach can scale to larger AI facilities.
Looking ahead
The London trial is intended as a blueprint for larger use of Emerald.ai’s technology in the near future. These include Aurora – a near-100 MW power-flexible AI factory in Virginia, US. Emerald is also integrating these capabilities into the NVIDIA Omniverse DSX Blueprint for very large “gigascale” AI factories to promote “grid-aware by design” infrastructure.
Overall, this initial trial has demonstrated that AI data centres can be dynamic, controllable resources for the electricity system. By enabling real-time, reliable demand flexibility without compromising critical workloads, this approach can support faster connection times, help manage peaks, facilitate renewable integration, and unlock capacity needed for the UK’s AI and low-carbon ambitions.
Find out more about our investments in innovation and projects to accelerate decarbonisation and reduce costs for consumers in our innovation hub.