AI is no longer just a software story. Andrew Bailey has pushed its power problem into the centre of economic policy.
Bank of England Governor Andrew Bailey has put a sharper name on the problem behind the AI boom: rationing. Speaking at an event in Kirkcaldy, Scotland on Friday, Bailey warned that artificial intelligence may soon be capable of doing more large-scale work than energy systems can physically support.
That is a different kind of warning from the usual arguments about jobs, safety or model accuracy. It asks a harder question. If there is not enough electricity to run every high-value AI workload, who gets priority?
As The Next Web reported, Bailey framed the issue as a social choice between uses such as healthcare breakthroughs, defence and other major sectors. That matters because central bankers do not usually talk this way about technology unless it has started to look like a macroeconomic constraint. Power supply is no longer a back-office issue for data centre developers. It is becoming part of the conversation about productivity, inflation, investment and financial stability.
The timing is important. The International Energy Agency has said data centres used about 415 terawatt-hours of electricity in 2024, roughly 1.5% of global power consumption, and that use could more than double to around 945 terawatt-hours by 2030. AI is the main driver of that increase, alongside wider digital demand. A typical AI-focused data centre can use as much electricity as 100,000 households, while the largest projects under construction can use far more.
For investors, Bailey's comments change the frame. The AI trade has often been treated as a race for chips, models and cloud capacity. But the physical bottleneck may be the grid. If compute cannot be connected quickly enough, projected AI revenue may arrive later than expected, and some data centre assets may be worth less than planned.
This is where the central bank angle matters. Energy shortages can feed into prices, capital spending and regional growth. If power demand from data centres pushes up network costs or forces expensive generation upgrades, households and businesses may feel it through bills. If governments decide that hospitals, defence systems or public infrastructure should come before commercial AI workloads, the market will have to price that political risk.
The IEA has also warned that around 20% of planned data centre projects could face delays if grid risks are not addressed. Transmission projects can take years to permit and build, while lead times for transformers and cables have lengthened. That is a serious mismatch for an industry where model cycles move in months and capital budgets are being written as if capacity will arrive on command.
Bailey is not saying AI should be stopped. He is saying the technology may run ahead of the infrastructure that makes it useful. A model that can discover a drug, automate a trading desk or manage a defence system still needs electricity, cooling, land, grid approvals and reliable backup power. The intelligence may be digital. The constraint is physical.
What rationing could actually mean
Rationing does not have to mean a minister deciding which startup gets a plug. In practice, it could begin through pricing. Data centres that need constant power in constrained areas may pay more for grid access. Projects that can shift workloads away from peak demand may get faster connections. Governments may also create priority rules for critical sectors, especially where AI supports healthcare, public safety or national security.
There are already signs of how this could work. National Grid said in March that a UK trial with Emerald AI, EPRI, Nebius and NVIDIA cut electricity demand from a 96-GPU NVIDIA Blackwell Ultra cluster by more than a third in under a minute without disrupting critical workloads. Over five days, the trial tested more than 200 simulated grid events, showing that AI infrastructure can respond to power conditions rather than behave like a fixed load.
That kind of flexibility may become a competitive advantage. A startup that can train or run inference when power is cheaper, pause non-urgent jobs during grid stress, or prove that its workloads support critical services will have a stronger case with customers, regulators and infrastructure partners. The companies that ignore energy will find themselves selling a product whose unit economics depend on a resource they do not control.
There is also a strategic opening here. AI can help the energy system itself, from grid forecasting to predictive maintenance and better use of transmission capacity. The IEA estimates that AI tools could help unlock up to 175 gigawatts of transmission capacity without new power lines if applied effectively. That does not erase the demand problem, but it shows why the best answer is not simply to slow AI down.
For founders, the lesson is practical. Build energy awareness into the product now. Know where workloads run, how much power they consume, whether they can be delayed, and how much value each unit of compute actually creates. The next phase of AI will not reward raw usage alone. It will reward useful work per watt.
Bailey's warning is not a niche comment about data centres. It is a sign that AI has reached the point where its limits are moving from the lab into the grid, the Treasury and the central bank. The winners will be the companies that treat power as a core input, not a footnote.
Also read: OpenAI is bringing Lockdown Mode to ChatGPT users as security risks grow • Zcash faces a trust test after its Orchard bug shakes ZEC • Apple will use WWDC 2026 to prove Siri can still matter
