Do Data Centers Increase Or Decrease Your Electric Bill?
As this debate rages, the Compute Heat Rate can tell you what happens next.
If you follow energy policy right now, you have probably seen the debate: Do data centers raise electricity prices, or lower them?
Depending on which study you read, you can find credible evidence for either answer. A recent E3 report called “Understanding the Drivers of Rising Electricity Rates and the Role of Data Centers” has been oft cited in recent weeks by the industry, and basically says data centers aren’t to blame for rising costs. Academic research published this spring in Environmental Research Letters says data centers could increase power costs by up to 57% in certain regions by 2030. There are many other examples on both sides.
Both sides are citing real data. Both sides are making defensible arguments. So what’s the deal?
The Denominator Argument: Why Prices Go Down
Electricity pricing involves a basic fraction. The numerator is fixed costs: the cost of building and maintaining power plants, transmission lines, substations, and distribution networks. These costs exist whether you consume a single kilowatt-hour or not. The denominator is the total volume of electricity consumed across all customers sharing that infrastructure.
Fixed Costs (Numerator) / Total Consumption (Denominator) = Cost Per Unit
When a data center plugs into an existing grid that has spare capacity, it increases the denominator without meaningfully changing the numerator. The same power plants, the same transmission lines, the same substations now serve more customers. Total fixed costs get divided across more megawatt-hours. Everyone’s per-unit share of those fixed costs declines.
So when existing infrastructure has slack, new demand absorbs that slack and spreads costs more broadly. Rates go down, or at least rise more slowly than they would have otherwise.
The historical data over the past five years reflects a period when this dynamic dominated. There was excess generation capacity in many regions. Transmission networks were not yet congested by new data center load. The infrastructure had room, and data centers filled it.
If the analysis stops here, the conclusion is straightforward: data centers are good for electricity prices. Several recent reports funded by industry groups have stopped here.
But the analysis cannot stop here, because the conditions that made it true are changing.
The Numerator Problem: Why Prices Go Up
The denominator story works when existing infrastructure can absorb new demand. When it can’t, the story inverts.
Once excess capacity is consumed, serving new demand requires building new infrastructure: new power plants, new transmission lines, new substations, new distribution feeders. And new infrastructure is dramatically more expensive than the legacy assets currently serving the grid.
The existing generation fleet was largely built decades ago, when construction costs, permitting timelines, and capital costs were all lower. A combined cycle gas plant built in 2005 might have cost $700 per kilowatt of capacity. The same plant built today costs $1,200 to $1,500 per kW. Solar was cheaper a year ago than it is today, and permitting timelines for both generation and transmission have stretched from years to decades in some jurisdictions.
PJM Interconnection, the largest wholesale electricity market in the United States, is already showing the early signs of this. Capacity auction prices hit the administrative price cap of $333/MW-day in consecutive auctions. Wholesale power costs in the market surged 76% in early 2026. The system fell 6,623 MW short of its reliability target for the first time in its history, and PJM’s own load forecast attributes the vast majority of incremental peak demand growth to data centers.
The Variable Nobody Is Measuring
The transition between Phase 1 (using excess slack in the system) and Phase 2 (building more), explains where we are today. But there is a deeper structural issue that neither the pro-data-center nor anti-data-center camps have fully grappled with.
Every forward curve model, every production cost model, and every rate impact study used across the electricity industry was built on a foundational assumption: that demand responds to price. When electricity gets expensive, industrial consumers curtail. Aluminum smelters shut down at $60 to $80/MWh. Steel mills reduce output at $80 to $120/MWh. Chemical plants cycle back at $100 to $160/MWh. This demand-side brake is the self-correcting mechanism that has kept wholesale prices within a bounded range for decades.
AI data centers do not voluntarily engage that brake. The compute revenue generated per megawatt-hour of electricity consumed is so far above any plausible wholesale electricity price that curtailment is not a rational economic decision.
And even if the economics somehow favored curtailment, the contractual structure prevents it independently. Hyperscale SLAs guarantee 99.95% to 99.999% uptime. This creates a demand class that is simultaneously massive in scale and non-curtailable at any historically relevant price.
The analytical framework for measuring this variable is the Compute Heat Rate (CHR)1: the maximum electricity price at which a given AI compute workload remains profitable, expressed in the same units ($/MWh) as every other electricity market metric. Where the gas heat rate tells you the floor (the price below which generators will not dispatch), the CHR tells you the ceiling (the price above which AI demand would curtail). The blended CHR across deployed commercial workloads is approximately $6,350/MWh, roughly 127 times above the traditional industrial curtailment threshold.
If your rate impact study does not include demand-side price tolerance as an input, it cannot tell you whether Phase 2 will behave like previous supply-constrained periods (where high prices triggered demand destruction and prices mean-reverted) or whether it will behave differently (where a significant share of demand persists regardless of price, preventing the historical self-correction). That distinction is the entire analytical ballgame.
What This Means for You
If you are a corporate energy buyer, a utility planner, a regulator, or an industrial consumer sharing a grid node with growing data center load, the actionable question is not which study is correct. The question is which phase your market is in.
If your local grid still has excess generation capacity and uncongested transmission, you are likely still in Phase 1, and the denominator argument holds. Enjoy it while it lasts.
If your market is showing capacity auction price spikes, reserve margin shortfalls, interconnection queue backlogs, and data center demand growth outpacing supply additions, you are entering Phase 2. And Phase 2 is governed by new-build economics, not legacy fleet economics.
In Phase 2, the relevant question is whether your current hedging strategy, your forward price assumptions, and your procurement approach account for the possibility that the fastest-growing demand class on your grid can absorb electricity prices 10 to 100 times higher than you can, and has no economic or contractual reason to stop.
Hans Royal is the originator of the Compute Heat Rate™ (CHR) framework. All views are his own and do not represent those of any employer or affiliated organization.
Royal, Hans, “The Compute Heat Rate: Quantifying AI-Driven Electricity Price Tolerance and Its Implications for Wholesale Market Repricing” (February 28, 2026). Available at SSRN: http://dx.doi.org/10.2139/ssrn.6322318