The takeaway: "Data centres have hiked electricity prices on the public by $23 billion" hit Hacker News' front page overnight. The number is real — computed to the dollar ($23,100,955,341) by the independent watchdog of America's largest electricity market. It is also six months old, narrower than the headline, and an estimate its own authors label "correct as to direction and order of magnitude." Here's what it actually measures — and why the news that matters happened quietly on the same day the Fortune headline circulated: the market's newest capacity auction cleared at its price cap for the third year running, still short of the grid's reliability requirement.

Where the number comes from

The $23bn is not Fortune's finding, and it isn't from this week. Fortune's 14 July piece is a re-headlined version of a 6 July essay by University of Florida energy economist Theodore Kury (whose research centre, he discloses, is part-sponsored by Florida utilities and the state's consumer advocate, with editorial control retained). The figure itself was first published on 5 January 2026 by Monitoring Analytics, the independent market monitor for PJM — the grid operator serving 65 million people across 13 states and the District of Columbia — in its analysis of PJM's capacity auctions.

The monitor's method: re-run the last three annual capacity auctions in a counterfactual world with data-centre load stripped out of the demand forecast, and take the difference. Its conclusion, verbatim: "The total increase in costs to other PJM customers as a result of data center load in the last three BRAs was $23,100,955,341." The monitor is explicit about precision: "The quantitative results are estimates… Even in those cases, the quantitative results are correct as to direction and order of magnitude."

What it measures — and what it doesn't

Worth being precise, because "hiked electricity prices on the public" does a lot of work in that headline:

• It's the capacity market, not your bill. Capacity payments — what generators are paid to promise availability — are one component of wholesale costs, which are one component of retail bills. The $23bn has not "landed on households" as a line item; it's spread across three delivery years running to May 2028, across everything from homes to factories.
• It's one grid region. PJM is the largest US market, but it is not "the public" nationwide.
• The definition is contested — including by the monitor's own tables. The report's text calls it an increase in "costs to other PJM customers"; its headline table measures the increase in capacity-market revenues, and a portion of those capacity charges is paid by the data centres themselves. Critics of the viral framing argue the transfer to other customers is therefore somewhat smaller. Separately, the monitor's own narrower calculation — counting only data-centre load growth above what was already embedded in forecasts — comes out at $21.3bn.
• Attribution is genuinely multi-causal. Rising retail bills also reflect gas prices, grid equipment costs and weather. The monitor's position is that data-centre load growth is the primary driver of capacity-market conditions; PJM, the grid operator, frames the same facts as a general supply-demand imbalance. Both can be substantially true at once.

The part that's actually new: yesterday's auction

The reason to care about a six-month-old number this week arrived with no virality at all. On the same day the Fortune headline circulated, PJM announced the results of its newest capacity auction, for the 2028/29 delivery year: cleared at $325 per megawatt-day — the FERC-approved price cap, for a third consecutive auction (a 2.5% step down from the previous cap), procuring 138,318MW at a total cost of about $16.4bn, and still falling 6,831MW short of the one-event-in-ten-years reliability requirement. PJM's chief executive David Mills put it plainly: "These auction results show that demand for electricity continues to grow faster than electricity supply."

For the previous auction, the monitor attributed $6.5bn — 40% — of the $16.4bn cost to data-centre load. It has not yet published the equivalent analysis for yesterday's result, but its January report contains what reads like a standing forecast: "This total impact of data center loads will continue to grow until the issues associated with the additions of large data center loads are addressed in an effective manner." Three cap-clearing auctions in a row suggest the issues have not been addressed.

The scale behind the fight

The underlying demand story is not in dispute. The DOE-commissioned Lawrence Berkeley National Laboratory report put data centres at 4.4% of US electricity consumption in 2023 and projects 6.7–12% by 2028. What's in dispute is who pays for accommodating it: the monitor has proposed making large data centres bring their own matching generation; the industry's lobby fought the monitor's related federal complaint (regulators denied it in April); and states in the PJM footprint are already litigating the question — AEP Ohio's data-centre tariff (requiring large data centres to pay for most of their requested capacity whether used or not) is currently before the state's supreme court.

The view from the UK

Britain has no equivalent capacity-market transfer story — yet. What it has is a queue: contracted grid-connection offers for demand ballooned from 41GW in late 2024 to 125GW by mid-2025 — against an actual national peak of around 45GW — with data-centre projects a large share, and Ofgem has opened reform of how demand connections are allocated. The "who pays for the build-out" question that PJM's auctions answer with a price cap is arriving here as a planning and connections fight first. Our piece on New York's data-centre permit pause covers the policy end of the same pressure.

The honest summary

The viral number is real, sourced, and six months old. It measures a capacity-market cost shift in one (very large) American grid region, estimated by a credible independent monitor to order-of-magnitude precision — not a nationwide bill increase that has already hit households. And the quiet news is worse than the loud number: the auctions that generated it have now cleared at their price cap three years running, short of reliability targets, with the monitor's meter — by its own account — still running.