The AI trade may cool without a dramatic crash, but the buildout will continue to reshape power networks, land markets, and delivery risk. Even if valuations drift, demand for compute and connectivity still feeds through to data centres, substations, and transmission upgrades. That means project timelines and permitting, not stock screens, will set the pace. The pipeline remains large and site specific.
Global data centre investment nearly doubled since 2022 and reached about $500 billion (C$670 billion) in 2024, according to the International Energy Agency. The same analysis projects that global data centre electricity use could more than double to around 945 terawatt hours by 2030, as AI workloads grow and new sites connect.
Those numbers point to a slower adjustment in spend, not a halt. They also suggest more pressure on local grids where clusters form, particularly around major cloud regions. In short, the cycle pivots from hype to hardware.
Load growth concentrates on utilities
Electricity needs are the binding constraint in most mature locations. Interconnection queues are long, and grid pockets with spare capacity are scarce. AI training demand remains bursty, yet the power contracts are long term. As one signal of pace,
“Thousands of data centres are set to be built in the next five years,” the IEA’s Fatih Birol said. Large sites pull in hundreds of megawatts, so the planning window extends over many years.
Under a non‑burst scenario, owners and operators face a different risk mix. Power purchase strategies must match shaping needs and curtailment risk. Cooling choices tie projects to water or to higher electrical intensity, which then affects transformer and switchgear schedules. These factors can push developers toward markets with faster siting, clearer tariff paths, and firm capacity options. AI demand becomes a siting filter rather than a blanket lift.
U.S. grid indicators already reflect this shift. PJM’s latest long‑term forecast points to a step‑change in peak demand through the 2030s, with summer peaks moving well above historic highs. The operator is expanding its planning horizon and clearing backlogs, yet bottlenecks remain in permitting and supply chains.
“This forecast captures the dramatic increases in future energy demand,” Aftab Khan said. New generation, storage, and transmission will have to firm up behind the meter and in regional plans.
For capital projects, the implication is steady work rather than a one‑way surge. Substations, high‑voltage lines, and synchronous condensers will see more bids where clusters deepen. Gas peakers and firm low‑carbon options may secure new roles that complement wind, solar, and storage portfolios.
Data centre developers, for their part, will weigh grid congestion charges against green power claims and delivery certainty. The result is a slower, more geographic rebalancing of spend.
What a soft landing means
A soft landing in AI markets would not erase committed pipelines. It would, however, shift focus from chip scarcity to grid readiness and lifecycle efficiency. Efficiency gains in software and hardware will offset part of demand, but not enough to remove the need for new capacity.
The IEA’s base case still shows electricity use more than doubling by 2030, and investment already near $500 billion (C$670 billion) in 2024, reinforcing the scale of works needed. Those works span land assembly, interconnection upgrades, and long‑lead equipment. The AI story is less about bubbles, and more about concrete, cable, and capacity.