AI Pulse

IEA warns data-centre electricity use could double by 2030

AI Markets 233 min read

The International Energy Agency now assesses that global data-centre electricity consumption could double by 2030 as the AI boom accelerates. The constraint is shifting from chip supply to power and grid interconnection.

The structure behind the number

The key variable in data-centre power is not the number of servers but power density per rack. GPU clusters draw several times more power than conventional cloud hardware, plus the cooling overhead, so the same floor area produces a very different load curve. The IEA's doubling estimate comes mainly from that structural shift, not simply from more buildings.

Grids become the new lead time

Chip fabs can add capacity; electricity cannot be conjured. A high-voltage transmission line often takes five to ten years from planning to energisation, and transformer and interconnection equipment lead times are stretching. For hyperscalers, securing power is becoming a harder precondition than securing GPUs.

Who benefits, who is capped

Generation, transmission and electrical equipment supply chains benefit, as do regions able to offer long-term power contracts. New projects stuck in interconnection queues, and markets with already-high electricity prices, are constrained. This is part of why the geography of compute expansion is migrating toward power-rich regions.

What it means for markets

For two years the AI trade has been about chips and models. If power becomes a hard constraint, part of the investment narrative shifts toward energy infrastructure — one reason capital has been flowing into electrical equipment and utilities. A caution: energy projects have far longer payback periods than chips, and the two do not share a valuation logic.

Nothing here is investment advice.

Source: CNBC · IEA

Related coverage