With over half of renewable energy projects in the grid connection queue looking at connection dates well beyond 2030, national electricity grid infrastructure is increasingly unable to handle the demands of rapid electrification. Speeding up grid connections has been a common ask of the next government in the build up to the election.
A recent DNV report found global electricity demand will double by 2050 as reliance on fossil fuels decreases. It suggested that the pathway to a decarbonised energy system requires significant grid expansion, solutions for grid congestion and new business models to accommodate rising demand and generation from wind and solar.
Remi Eriksen, group president and CEO at DNV, said: “Deep digitalisation, including the application of AI, is crucial for managing the increased complexity of a renewable-dominated power system.”
The sudden entrance of generative artificial intelligence into public use has led to a massive hype cycle that the renewable energy industry has not evaded; it has meant an increased focus on digitalisation and data sharing, with the argument made that AI holds the answers to several problems facing the renewable energy sector, with grid connections oft raised as one of them.
In theory, AI could clear up the connections queue and create grid efficiencies—while, no doubt, generating a couple of articles on the topic. But this claim, which becomes a get-out-of-jail-free card for nearly every issue, with AI promising to take over and speed up mundane administrative tasks across industries, forgets the technology’s mechanics.
AI cannot solve the grid connection issue unless the data centres that facilitate it come online. That means they, too, must find a way to connect to the grid; the same barriers facing new solar and wind projects apply.
An electricity grid fit for AI
David Bloom, chairman of Kao Data, which historically operated successfully out of west London, told Bloomberg that the company is having to consider a change of location, facing waits beyond 2030 for connection.
Upon becoming president of Scottish construction industry trade association SELECT in June, Mike Stark said that the UK’s National Grid could struggle to satisfy the voracious energy needs of AI and the systems it supports. He questioned whether the UK’s current electrical infrastructure was fit for purpose in the face of the massive increase in predicted demand, particularly from the power-hungry data centres supporting AI.
Speaking exclusively to Current±, Stark described the struggle to facilitate the scale of grid connections needed.
He likened the current approach to stacking extension cables: “Not thinking about the fundamental source of everything; what about looking at the fundamental source of the grid, is it suitable to take this additional load?
“It’s adding and adding and adding on to something that may be, at source, struggling to meet the demands and might not be fit for purpose.”
Stark also added that carrying out the necessary upgrades is no mean feat. The skills shortage is “an absolutely massive and serious challenge” and, while education initiatives do exist, “in the last two years, the actual funding for apprenticeships has dropped dramatically”.
The added AI strain
The H100 chip manufactured by Nvidia has powered much of the generative AI boom (it is estimated that Nvidia supplies 94% of the chips used in the technology). According to a LinkedIn post by Microsoft’s Principal Electrical Engineer of Datacenter Technical Governance and Strategy, Paul Churnock, a single H100 GPU (graphics processing unit) running for 61% annual use consumes about 3,740kWh of electricity each year.
A data centre facilitating generative AI houses thousands of those chips, not only drawing electricity for computing but also to cool the immense heat generated by the technology. In fact, a large proportion of a data centre’s energy use goes to cooling the systems inside.
John Pettigrew, National Grid CEO, was recently quoted by Bloomberg UK saying the electricity demand from UK data centres will jump sixfold over the next 10-years due to AI’s power demands.
“Future growth in foundational technologies like artificial intelligence and quantum computing will mean larger-scale, energy-intensive computing infrastructure,” Pettigrew said in a speech shared to LinkedIn.
Big cloud services providers say they will power their data centres on green energy by 2030. This will also require more efficient energy use, with more effective chips providing computing power.
Using green energy to power data centres is one thing, but connecting a data centre to a renewable energy source is another. Stark pointed out that, while feeding back renewables into the grid is a go-to solution for balancing it, the current grid cannot handle that input and output level.
Furthermore, the distance between a renewable energy source and the data centre seeking to use its power can be an unforeseen difficulty. Locating data centres near renewables is one way that companies aim to ensure relative sustainability.
That is why places like Scotland and Ireland house proportionately large numbers of them. Ireland’s proliferating data centre landscape, with a vast number of sites housed in the country, is also partly due to its cooler climate, meaning that less power is needed to prevent the systems from overheating.
Current± has covered the impact of data centres in Ireland in some depth. It is estimated that in 2022 Irish data centres consumed roughly 5.3TWh of electricity – around 17% of the country’s electricity demand.
However, three years ago EirGrid put an effective moratorium on new data centres in Dublin set to last until after 2028. There are some reports that the Irish government may push to limit new data centre developments that don’t utilise carbon-neutral power sources.
In spite of that step, Cornwall Insight data showed that Irish wholesale power prices were a third higher than elsewhere in Europe. Sarah Nolan, senior modeler at Cornwall Insight said at the time that this was at least partially the result of data centre demands.
Irish Minister for the Environment Eamon Ryan has said new data centres will not be connected to the electricity or gas grids until they stop relying on fossil fuels and reduce their carbon emissions.
Using AI to solve the problem
It is worth noting that while all data centres make huge energy demands, those facilitating AI require an even higher amount.
Crucially, the amounts of energy and water used to keep AI running at the ever-larger scale that monetising the technology demands are unknown—but it is unlikely that businesses keep those numbers secret because they are so low.
Reducing that reliance on fossil fuels means effective build-out of renewable electricity sources and cooling methods, which tends to require grid efficiencies that are unachievable under the strain from AI development. Yet, the argument that AI will in fact solve that strain is also being made.
Used for research, some say that AI could provide solutions for nuclear energy generation, or feature in technologies that reduce emissions. In its most recent sustainability report, Microsoft claims: “New technologies, including generative AI, hold promise for new innovations that can help address the climate crisis.”
Other solutions might see the excess heat generated by AI data centres applied to other use cases. For example, Deep Green uses the heat generated by “high performance compute”, used for AI and machine learning, to warm swimming pools.
Last year, Mark Bjornsgaard, CEO at Deep Green told Current±: “Computers are brilliant sources of heat, they convert 97% of their electricity into heat. As we can’t necessarily build enough heat pumps fast enough, we actually need computers as heaters to solve the problem, even if that’s just in the short term.”
Whether or not the electricity grid is ready, AI is developing at scale. Perhaps the only way forward is to hope its application will accelerate renewables development and provide a solution to grid constraints – retroactively though it may be.
