The unprecedented growth in AI applications across the world is accelerating the use of high-performance servers and thus, the demands for power in data centres. In fact, these facilities are fast becoming the most power-hungry components of the digital economy. The International Energy Agency (IEA) estimates that against the backdrop of AI expansion, global data-centre electricity use could more than double by 2030, to around 945 TWh. Over the last five years electricity consumption has grown at 12% per year.
While data centres can be built and operational within two or three years to meet the demands of AI, as the IEA points out, it takes considerably longer for energy infrastructure to be updated. In the meantime, AI-driven power needs are creating issues of supply which are upending existing power systems and the wider energy transition. In some parts of the U.S. the demand from AI data centres is already outstripping the supply of generation. Power operators are having frenzied discussions to establish how to build generation faster, with soaring costs for consumers in the meantime.
As many regions of the world seek to double down on climate goals, AI is presenting new challenges to already strained systems that are trying to ensure uninterrupted power while balancing the cost of delivering net zero and the cost of falling short on policy goals. Consequently, existential conversations about AI’s contribution to economic and social progress spark questions over how its development can be met affordably and cleanly.
Why data centres present new challenges
The difficulty is not solely about the increase in electricity demand; it’s also about the nature of the demand. AI workloads can be unpredictable; one moment power use is steady, and the next an AI training run sends demand soaring by tens of megawatts. These sudden spikes—called “burst” loads—are hard to predict and put pressure on the grid. If the grid can’t respond instantly, the blackout risk rises.
AI also cannot afford a single second of downtime. Data centre cooling systems must run continuously, and their computing loads are relentless. Every backup needs a backup.
Adding to this challenge, data centres often cluster in the same places such as Frankfurt, London, or Chicago which puts extra pressure on local networks. Generative-AI is already straining energy grids, with data centres consuming as much as 30% of grid capacity in some parts of the US.
In many areas, energy grid operators have paused new connections until upgrades are made. Developers have been forced to find quick fixes on-site such as batteries, or diesel, just to keep running. These stop-gap solutions may keep the lights on, but they are insufficient in the long term.
It would be easy to think that balancing the demands of AI with the drive to decarbonise are at odds. But what if we reframed the question? Could AI be the much-needed catalyst that society needs to deliver on its climate goals and expedite the supporting clean technologies to get us there?
Managing the electrical revolution
Energy grids around the world are becoming more decentralised, with power coming from huge, distributed networks across the grid. This requires sophisticated management from grid operators to respond in real-time to demand and ensure long-term strategic planning. AI has a role to play here in helping to optimise energy systems by forecasting demand, integrating renewables, and improving grid efficiency. In time, it should be able to predict maintenance, reduce downtime and emissions contributing to grid management.
In fact, effective grid management is indicative of a well-functioning, modern and advancing society. Like the energy transition, however, it is a system in evolution.
The power outage in Spain and Portugal exemplified this. At the time of the outage, Spain had installed renewable generation capacity more than four times the immediate load requirement, yet the system lacked sufficient rotational inertia from conventional generators, leading to instability. The system services missed should have come from more traditional forms of inertia on the system – e.g. from rotating equipment from gas turbine power generation.
This was a clear example of where the ability to predict what the grid required and integrate it, was left wanting.
As energy management systems advance, AI algorithms will be able to analyse vast datasets from smart grids, help to manage battery and long-term energy storage and carbon tracking and will enhance grid resilience, particularly crucial for integrating renewables. It’s clear that the more digital our infrastructure becomes, the more intelligent our energy systems need to be.
Delivering technology solutions without compromising the transition
Adapting existing power infrastructure is how we can deliver more capacity for the AI power surge, while maintaining grid stability to avoid blackouts. Shifting from coal fired power plants to gas will cut emissions by up to 65%, ensuring progress on climate goals.
But it is gas turbines that will allow us to move to the next stage, providing the flexibility and responsiveness that’s needed to stabilise the grid. They can be brought online quickly, offer high ramp rates, and, when designed with future fuels in mind, such as hydrogen, can serve as a crucial stepping-stone in the energy transition.
For energy players, the takeaway is to build optionality that will support the ever-increasing demands of AI. The market clearly recognises gas not just as a bridge, but as a foundation that can evolve with the energy transition. If we invest now in systems that can run on hydrogen blends, even if full-scale supply isn’t yet available, it allows us to leverage today’s transitional technologies while shaping tomorrow’s low-carbon ecosystem. What we need is a well-balanced energy system that combines renewables, storage, and dispatchable power, like gas and hydrogen.
Grasping the opportunity of AI
As AI expands, transparency will be key. Sophisticated energy management systems must align strategic planning with that of data centres to build future capability. Furthermore, data centres must take responsibility for power, and water usage, a requirement that the EU will seek in time through regulation.
At the same time, the rules that govern AI will need to change, providing certainty and boundaries which will enable AI to advance. For now, it’s for us to embrace the technological depth and challenge that AI can bring and work to secure an energy future that can realise a technology-neutral mix, from renewables and hydrogen to intelligent grids and flexible thermal systems.
