From 2015 to 2024, Every Doubling of Data Centre Capacity Was Followed by a 3.5% Fall in Average Retail Electricity Prices — About 6% at the State Level. Data Centres Were Making Power Cheaper. What Changes That Is Not the Buildout Itself but Whether the AI Demand Behind It Shows Up.

A working paper from the Electric Power Research Institute found that until at least 2024, data centre operations did the opposite of what households feared. Using Federal Energy Regulatory Commission data and EIA retail revenue figures from 2015 to 2024, researchers identified a causal relationship running the other way: for every doubling of data centre capacity, average retail electricity prices fell 3.5%, and roughly 6% at the statewide level.

The mechanism is specific to how electricity is priced. Unlike soybeans or gasoline, where the price reflects the cost of production, electricity prices are based on cost recovery — how much is consumed. Fixed costs are divided across kilowatt hours, so more kilowatt hours mean a larger denominator and lower unit prices, an effect compounded as additional and more efficient generators come online to serve the load. That arithmetic holds only while the consumption arrives. EPRI researcher Asa Watten put the failure case directly: if the grid builds capacity expecting substantial data centre demand and that demand does not materialise, the fixed costs get spread across fewer people, which is the opposite of the dynamic that produced the savings. The evidence that the trend may already be turning is accumulating — PJM has attributed a projected $6.3 billion increase in consumer electricity costs over three years mostly to data centre demand, Virginia's residential prices have risen more than 13% in the past year, and Goldman Sachs has projected the buildout adding 6% to electricity costs between 2026 and 2027 with a further 3% by 2028. McKinsey puts total data centre construction spending on track for $7 trillion by 2030.

For the investor, this reframes the AI capex debate in a way the usual bubble framing misses. The question has generally been whether hyperscalers earn a return on their infrastructure; the EPRI finding shows the same uncertainty determines whether households see lower bills or higher ones, through the identical variable — realised demand against built capacity. Mark Cuban's warning that many data centres will end up as pickleball courts rests on a related argument, that AI will become cheaper to run through efficiency gains and much of the capacity will prove unnecessary. What makes the electricity channel worth watching is that it is measured continuously and publicly, in utilisation and retail rates, rather than quarterly in earnings. If the historical relationship inverts in the data, that is an early read on the demand question — visible in a utility bill before it is visible in a balance sheet.

Sources — Fortune, July 26, 2026 · Electric Power Research Institute working paper, 2026 · McKinsey, 2026

[email protected] is on ObliqueFront.com  list because you opted in before the crowd caught on.

If we go quiet — look in promotions, updates, or wherever your inbox files things it hasn't figured out yet.

Privacy Policy

Unsubscribe. We'll part ways cleanly.

If something's off, you can reach us here

© 2026 Alpha One Marketers LLC. All rights reserved.

📍 254 Chapman Rd Ste 208 Newark, Delaware 19702.