Data centre load has become a consumer price issue

For about two years the link between data centre construction and household electricity bills was a plausible mechanism without much evidence behind it. That is no longer the case. US household electricity prices now sit roughly 32 percent above their 2019 level, with larger increases in data centre intensive regions, and the pathway from load to tariff is documented at the level of individual auctions and rate cases.

The mechanism, in order

It runs in four steps and each one is observable.

1. Load arrives faster than generation. A single hyperscale facility can draw as much power as 100,000 households, and the largest projects are an order of magnitude beyond that. Interconnection and generation take years; a data centre takes months.
2. Capacity markets reprice. In the PJM interconnection, data centre load is associated with a 9.3 billion dollar increase in the 2025-26 capacity auction, implying on the order of 18 dollars a month on bills in western Maryland and 16 dollars in Ohio.
3. Utilities file for base rate increases. In Virginia, Dominion proposed its first base rate increase since 1992, adding roughly 8.51 dollars a month for a typical household in 2026 and a further 2.00 in 2027, alongside a resource plan of nearly 27 gigawatts of new generation by 2039.
4. Regulated cost recovery spreads the bill. Transmission and generation investment is recovered through tariffs across the rate base, which means households pay a share of infrastructure built to serve a load they do not consume.

Why it is a macro number and not just a bill

Electricity is a small share of the consumption basket and a large share of the fixed cost of living for low-income households. That asymmetry is what makes it macroeconomically interesting. Estimates put the drag on real consumer spending growth at roughly 0.2 percentage points in 2026 and 2027, with the net effect on GDP growth closer to 0.1 points once higher utility capital expenditure is netted against it.

Those numbers look small. They are not small relative to the growth rate they are subtracted from, and they are concentrated: larger for lower-income households, and larger in the regions hosting the facilities. An average of 0.2 across the distribution implies something considerably worse at the bottom of it.

The political economy is moving faster than the analysis

Policy has already begun responding. There have been moves to require technology firms to contract for new generation capacity on long-dated terms, and a commitment framework under which operators bear the full cost of the infrastructure upgrades their facilities require. Whether those hold is a separate question. The relevant observation for forecasting is that cost allocation, not total cost, is now the live variable.

That matters commercially in a specific way. If the allocation shifts toward operators, the effective cost of compute rises and some marginal projects do not proceed, which slows the load growth that caused the problem. If it does not shift, household bills continue to absorb it and the political pressure compounds. Both paths are stable in the short run and neither is stable over a five-year horizon.

The exposure nobody hedges

Energy-intensive manufacturers in data centre dense regions face the full power price increase with none of the offsetting revenue. Aluminium, steel, cement, chemicals: same grid, same tariff, no AI upside. This group is rarely modelled as exposed to the compute buildout, and it is the group with the least ability to pass costs through. We would treat regional electricity price forecasts as a first-order input to any industrial siting or margin analysis in the affected interconnections.

Sources

• CEPR, powering the digital economy, on household prices 32 percent above 2019
• Goldman Sachs, macroeconomic spillovers from AI electricity demand
• Belfer Center, AI, data centers and the US electric grid
• PJM capacity market impact, data centre growth projections
• Consumer Reports, AI data centres and electricity bills