Like protein-folding for electricity grids? AI is deciding where electrons go
But do some energy-guzzling forms of AI outweigh those benefits?
Did you notice that, as the US President arrived on British soil – to be swaddled in as much pageantry as the British state could muster – quite a lot of people were talking about AI?
Microsoft’s chief executive claimed investment in AI could boost Britain’s stuttering economy. And the UK and US also agreed on a new “prosperity deal” worth $42 billion (£31 billion) to forge stronger ties over technologies such as AI, quantum computing, and nuclear energy, an economic component of the state visit. (Part of multiple investment pledges worth a total of £150 billion). A smartly presented bottle of wine hastily handed over to dinner hosts in the hallway.
It’s no surprise that AI is a key part of this exchange – AI is the future. I’m sure you’ve heard that by now. Coincidentally, National Grid announced this week that it is partnering with US-based firm Emerald AI to bring electricity grid-managing solutions to Great Britain. In short, AI is going to help make it possible for the UK to, ahem, build more AI. Allow me to explain.
Data centre boom
Electricity grids are struggling as energy-gobbling data centres spring up. Those data centres are largely earmarked for AI because, well, the algorithms need somewhere to live, basically. Many see it as ruinous because the grid can barely accommodate this new AI rush, particularly for job-stealing, word- and image-spewing generative AI. The carbon emissions associated with such developments are huge.
But National Grid has an answer, it says: “In many cases, there’s room on the existing grid to connect new data centres, if they can temporarily dial down energy usage during periods of peak demand.” This is right where AI comes in, at least, a different form of AI better referred to as machine learning or intelligent control. Apparently, it’s going to help manage how and when data centres suck up the energy they need.
In principle, this sort of thing is possible, for example, by pre-cooling data centres when electricity is plentiful, or by carefully scheduling non-urgent computation. National Grid says it will soon demonstrate its own version of this with a system capable of “changing computing activity when the grid is under pressure”.
I have no doubt that non-generative AI and other technologies are going to help improve data centre efficiency. But there’s arguably an even bigger prize here in the form of smart grids that use AI to manage the flow of electricity more intelligently than ever before. Smarter heating and cooling, for instance, might drastically reduce energy demand from buildings. “Adopting artificial intelligence could reduce energy consumption and carbon emissions by approximately 8% to 19% in 2050,” one 2024 study explained.
Grids are getting more complicated. Renewables and electrification – heat pumps and electric vehicles (EVs), for example – mean that operating sprawling electricity grids is only going to get harder. AI, admittedly, has a really important role to play in all that. A sort of analogue, perhaps, to the revolution it has brought about in protein science, since AI can work out how to fold incredibly complex molecules and potentially help discover new, life-saving drugs.
Being able to predict with greater accuracy when renewable energy will be available, along with how demand will fluctuate, makes it easier to turn the grid green. You don’t need to rely on lumbering fossil fuel plants providing a “base load” so much.
Machine control
AI-driven management of energy storage systems, or maintenance schedules, could also help. Plus, the International Energy Agency said last year that AI tools could unlock as much as 175 gigawatts of transmission capacity: “AI-based fault detection can help rapidly identify and precisely pinpoint grid faults, reducing outage durations by 30-50%. Remote sensors and AI-based management can increase the capacity of transmission lines.” The energy sector is not yet making enough use of AI, the IEA says.
Among those currently trying to bring AI technology to grids is the US Department of Energy. Really, it’s beginning to pop up everywhere. You can go very fine-grained with this stuff. Recently, I covered an experimental machine learning-based system that could predictively pre-heat people’s water heaters, and help stabilise the grid in the process.
I was about to say that AI technology of this kind is not generative in nature. But, actually, the US National Renewable Energy Laboratory (NREL) is working on an electricity grid management system that uses generative AI to “deliver reliable information and decision-making support across a range of possible applications within power systems”.
Do grid operators need an AI companion? I’m not sure. But I’m keen to hear more about applications of AI in the energy system generally. In the coming months, I’m hoping to cover some genuinely transformative examples of AI for grid management on this newsletter. (If you have an example that’s worth shouting about, get in touch!)
In any case, the AI industry as a whole rightly faces scrutiny because of serious ethical concerns around generative AI and soaring energy demands, which threaten to push up electricity prices, creating a chilling effect on electrification. Plus, we can’t ignore the fact that some data centre projects are hardening reliance on fossil fuels. All that said, I do think that there’s a really interesting march of data-crunching technology afoot in the middle of all this.
The truth is, electricity grids really are going to have to get smarter. And we are going to need to work with algorithms, of one kind or another, to make that happen.
Further reading on this week’s story
Among the articles you should read on AI’s energy consumption are this one from Wired – about the fact that many AI companies are not yet being fully transparent on this point – and this thought-provoking one on the energy costs of ChatGPT queries. But, overall, remember that several analyses show that the rapid expansion of AI is currently prolonging fossil fuel use.
A 2023 study describes an app called MERLIN, for managing grid-interactive domestic appliances.
Finally, a paper published just this week explores how the considerable cost of reinforcing electricity grids in preparation for an electric future could be reduced through a combination of improved appliance efficiency, lower thermal demand and better coordination of when devices run.
Update: Added a note about electricity prices.
Thanks for reading! If you enjoyed this story, don’t forget to share it with your friends and colleagues. You can also subscribe to The Reengineer and follow me on Bluesky.