National Grid has invested in an AI start-up that can intelligently manage data centres’ electricity usage as it tries to accommodate a wave of new projects seeking to connect to its power network.
Data centre owners could ultimately be offered the option of connecting to the grid sooner if they agree to temporarily dial down power usage during periods of peak demand, such as by using the software of Emerald AI.
Server farms have generally been regarded as inflexible consumers, wishing to maintain full power supplies at all times. Emerald AI’s platform, Emerald Conductor, uses artificial intelligence to assess which workloads at the data centre can be slowed, paused or moved to another data centre to reduce power consumption temporarily.
The technology is designed to allow data centres to secure grid connections without waiting for new cables in areas where cabling constraints would otherwise risk their usage causing local electricity shortages at peak times.
A demonstration of the technology at a UK data centre is planned for later this year.
Ministers are expected to welcome a raft of AI data centre investments by US companies this week during President Trump’s state visit.
The government says it wants to be the world leader in AI, which relies on vast amounts of computing power from data centres, using huge amounts of electricity.
Data centre developers have been frustrated by difficulties securing grid connections because parts of Britain’s electricity network do not have the cabling capacity to cope with their demand at peak times.
• AI forecast to fuel doubling in data centre electricity demand by 2030
National Grid, which operates the high-voltage electricity cables in England and Wales, has 13.5 gigawatts of data centres seeking to connect to its network, the equivalent power demand of four new nuclear plants like Hinkley Point C.
The FTSE 100 energy group said it had signed a strategic partnership with Emerald AI that would “help position the UK as a global leader in AI and reduce the need for additional infrastructure by giving National Grid more confidence that data centres can lower their energy use when required”.
National Grid has also invested an undisclosed sum in Emerald AI, the start-up founded last year by Varun Sivaram, a 36-year-old former Orsted executive, who has already raised $24.5 million in seed funding from backers including Nvidia’s venture capital arm, Jeff Dean, Google’s chief scientist, and the former US secretary of state John Kerry.
Sivaram told The Times: “This is the biggest opportunity I’ve ever seen. In the US alone, 100 gigawatts of data centres could connect to the electricity grid today with today’s generation and today’s transmission, if we had just modest flexibility.” They would only need to reduce their load by an average of 25 per cent for two hours at a time or less.
Varun Sivaram founded Emerald AI last year
VALERIANO DI DOMENICO/WORLD ECONOMIC FORUM
He said Emerald Conductor was “an AI for AI”, assessing where power savings could be made across the numerous activities taking place in the data centre. “When you send a query to ChatGPT, that answer is going to come to you after 1.5 seconds and so if I add 10 milliseconds because I had to move a query from Dallas to Denver, you may not even notice. Sometimes I will slow things down in one data centre, or pause something for an hour, or move an operation from one location to another location.”
He said managing this was “computationally intractable for humans to try and figure out”.
Steve Smith, chief strategy and regulation officer at National Grid, said: “As the UK’s digital economy grows, unlocking new ways to flexibly manage energy use is essential for connecting more data centres to our network efficiently. This groundbreaking trial with Emerald AI demonstrates how innovative technologies can help optimise the grid, enable increased investment in advanced computing, and deliver real benefits to the wider UK economy.”