US data centres could need 170 gigawatts (GW) more electricity by 2030 than the power grid can supply, according to TrendForce, the Taiwanese market research firm.
TrendForce forecasts that data centre power demand will reach 272.4GW in 2030, based on expected shipments of computing equipment.
It estimates the grid will be able to deliver only 100.7GW.
A gigawatt is enough to power several hundred thousand homes, so the gap is equivalent to the output of well over 100 large power stations.
About 124GW of the shortfall comes from demand running ahead of forecasts by Lawrence Berkeley National Laboratory (LBNL), the US government research centre.
TrendForce said that gap would have to be filled by behind-the-meter generation, meaning power plants built on site that bypass the grid, faster grid expansion, or by building data centres overseas.
The remaining 49GW or so reflects delays in connecting new capacity to the grid.
Those delays are concentrated in three regional grid operators: PJM, which covers 13 eastern states and Washington DC, ERCOT, which runs most of the Texas grid, and MISO, which serves the Midwest and parts of the South.
TrendForce does not expect the problem to bite immediately.
It said grid delays in 2026 and 2027 are too small to change how companies buy equipment.
The mismatch between power needs and supply only emerges after 2028, according to the firm.
Analysts said the timing supports their view that power shortages and local opposition to new data centres are not yet holding back spending.
They added that future constraints will depend largely on how companies and governments respond to the potential shortfall.
That puts the onus on utilities, regulators and technology companies to act before the end of the decade.
The "several hundred thousand homes" comparison and power-station equivalence are my additions for context, not from the note; cut if you'd rather keep strictly to source. Broker also needs naming.
