LONDON — Tech giants must be allowed to burn more fossil fuels if the U.K. is to become a global AI leader, British ministers were told this summer.
The warning was raised at a meeting in late June between Technology Secretary Peter Kyle, Energy Secretary Ed Miliband and executives from leading U.S. tech firms.
The suggestion — that on-site gas fuel cells could provide an “interim measure” to get around lengthy waits for a connection to the electricity grid — exposes the tensions the U.K. faces as it tries to be both a climate leader and an “AI maker.”
Achieving the latter relies on rapidly increasing the number of energy-intensive data centers for AI on British soil.
But Kyle has said the government is “clear-eyed … on the need to make sure we can power this golden era for British AI through responsible, sustainable energy sources.”
High costs, long delays
The AI Energy Council, made up of tech firms, energy suppliers and regulators, is tasked with advising the government on how the U.K. can marry its AI and climate goals.
Its focus includes “ensuring the U.K.’s energy system is ready to support the country’s AI and compute infrastructure” while “promoting sustainability and the use of renewable energy solutions.”
But members have warned ministers that the U.K.’s comparatively high energy costs and long wait times to access the grid risk undermining AI investment, according to readouts seen by POLITICO.
At an inaugural meeting of the Council in April, attendees raised “critical challenges associated with grid connections — including in relation to recent reforms — and the infrastructure necessary for AI Growth Zones,” the government’s flagship plan to boost AI capacity by easing planning requirements in designated areas. Bids must be able to guarantee access to at least 500MW of power by 2030 to be designated.
Ministers were further told that “high wholesale electricity prices, long lead times for grid connections, and energy pricing are challenges to data center investment across the U.K.” at a second meeting in June, days after the government promised new measures to speed up grid connections in its Industrial Strategy.
A spokesperson for the Department for Energy Security and Net Zero said its plan for the U.K. to reach clean power by 2030 will enable the development of energy-intensive data centers, alongside work with regulator Ofgem and network companies “to reform the outdated connections process and speed up delivery of new infrastructure.”
However, attendees of the June meeting said AI demand is likely to “rely heavily” on reprioritizing existing grid connections until 2030, suggesting promised reforms and investment are not expected to alleviate constraints significantly before then.
Quick fix?
As countries race to adapt their electricity grids for the demands of AI, locating data centers close to new and existing gas facilities has been suggested as a quick way to access energy.
“Temporary on-site generation, including natural gas fuel cells, was raised as an interim measure to meet power needs during grid connection delays,” according to the June readout, noting the “limitations of batteries for backup.”
DESNZ said the government is considering a range of options, but indicated it is not planning to authorize on-site gas cells. A spokesperson repeatedly declined to rule it out on the record.
However, a recent report by the Tony Blair Institute warned that “bridging measures” will be needed to meet the U.K.’s short-term demand for AI infrastructure until adequate clean energy, including nuclear, is brought online.
It urged the government to permit developers to site data centers close to both existing gas plants and purpose-built “modular” gas fuel cells.
The government should “avoid long-term lock-in to gas” by refraining from constructing a new fleet of plants and repurposing gas assets to support “grid flexibility” once data centers are connected to renewable energy, it added.
Gas pipeline operators are “actively pursuing” enquiries from data centers, the Future Energy Networks, which represents the U.K.’s gas transmission network, told The Times in April.
The U.K.’s National Energy Systems Operator (NESO) has said the U.K. will require up to 35GW of spare gas capacity even if the U.K. hits the government’s target for a clean power system by 2030.
Access granted
AI Energy Council members include Microsoft, Google and AWS, as well as American data center operator Equinix and Canadian asset manager Brookfield.
It is not clear from the readout which attendee raised gas fuel cells. The point was not raised by Google, the company said. Microsoft and AWS declined to comment.
Equinix said it is interested in fuel cells, which it has already deployed at sites in the U.S., for their “energy reliability and sustainability,” as well as other “onsite gas to power solutions.”
The suggestion of permitting the use of gas comes amid scrutiny of both the environmental cost of the U.K.’s AI ambitions and its reliance on largely American firms.
Despite DSIT warning in a report last month about the risks of relying on global hyperscalers to provide the U.K.’s AI infrastructure, only one British tech firm, ARM, has attended the Council.
A DSIT spokesperson said attendees were selected “for their unrivaled expertise” operating some of the world’s largest data centers, but the AI Energy Council expected to draw on a wider range of stakeholders in the future.
The U.K.’s largest energy providers and officials from DSIT, DESNZ and regulators also attend the Council, with participants emphasizing the need for “improved collaboration” and “coordination” between the government, energy and tech sectors.
DESNZ said its clean power plans are consistent with forecasts for data center demand, but attendees have also stressed the importance of refining estimates of AI’s energy demand.
NESO estimates that energy demand from data centers could reach 7 percent of the U.K.’s total by 2030, but acknowledged at June’s meeting that its models “don’t fully incorporate location and grid connection dynamics.”
Impact on the grid is likely to be affected by the balance of demand between energy-intensive AI training, which can take place in remote locations with abundant, cheaper energy, and the use of those models known as “inference,” which must be located closer to demand hotspots.
DSIT expects the U.K.’s demand for AI computing capacity to increase seven-fold to 13.6GW in 2035, with inference accounting for up to 70 percent of demand by that point.
“Building a shared industry view on grid constraints, latency trade-offs, and siting challenges was identified as a critical next step,” at the minutes from the June meeting read. The council is due to meet again in the fall.
The post UK’s AI ambitions clash with its climate goals appeared first on Politico.
