PJM Data Centers Added $11.11/MWh to Wholesale Power in H1 2026

Data center load added $11.11 per megawatt-hour to PJM wholesale power costs in the first half of 2026, equal to 9.7% of the total, according to Monitoring Analytics, the grid's independent market monitor. The figure appears in the firm's 2026 State of the Market reporting and in a PR Newswire summary circulating again on November 7, 2026.

This is not a bill line on your home electric account. It is a capacity-market attribution for the country's largest wholesale grid, serving roughly 67 million people across 13 states and Washington, D.C. — the same footprint where Google and Voltus expanded a 1 GW virtual power plant and where regulators recently rejected a curtail-first data center plan.

TL;DR: what the number means

How PJM wholesale cost rose

The press release breaks the year-over-year jump into components. Energy costs per MWh rose about 47.7%. Capacity costs per MWh rose about 207.1%. Transmission rose about 5.6%. Together they pushed total wholesale power cost from $76.16/MWh in H1 2025 to $114.50/MWh in H1 2026 in the headline comparison — a 50.3% increase.

Capacity is the slice most tied to forecast load growth. Monitoring Analytics has said that actual and forecast data center load was the primary driver of tight supply-demand balance and high prices in the 2025/2026 through 2027/2028 capacity auctions. AI labs and cloud builders do not pay this invoice directly; utilities, retail suppliers, and colocation customers eventually absorb it through rates and PPAs.

Data center power grid link between transformer and server cabinet illustrating AI load on wholesale electricity

Why 9.7% is a floor, not the full story

The monitor is explicit: the $11.11/MWh increment does not include data centers' effects on energy prices or transmission costs. Analysts who summarized the report for energy press note the true total pass-through is higher than 9.7%, even though nobody has published one combined percentage.

That matters when social posts round the figure to $11.10/MWh or call it "the cost of AI on your bill." The defensible claim is narrower: capacity-market treatment of data center load raised monitored wholesale cost by $11.11/MWh in H1 2026. Anything beyond that needs separate modeling.

Connection to AI infrastructure fights

PJM is where US AI buildouts meet ratepayer politics. The monitor has recommended that forecast data center demand not be included in capacity markets the way it has been, and that data centers be required to bring their own new generation before imposing costs on other customers.

That recommendation sits in the same policy conversation as:

• OpenAI's Grid Protectors flexible demand program in other regions.
• European moratoria and protests over dirty power for data centers.
• Congressional scrutiny of who pays for transmission upgrades, which we covered in Senate investigation themes on hyperscaler disclosures.

For model training economics, power is indirect but real: if your cloud region sits on PJM, colo quotes and PPA prices already embed capacity auction outcomes. Teams sizing always-on GPU clusters should treat grid headroom as a schedule risk, not only a real-estate question.

What competitors are doing on the same grid

Hyperscalers respond with flexibility products, not only new gas plants. Google's Voltus VPP expansion is an example of buying dispatchable demand response inside PJM instead of assuming unlimited interconnect. Nuclear uprate deals and behind-the-meter generation elsewhere on the US grid follow the same logic: bring capacity or buy flexibility.

Smaller AI labs rarely negotiate at that scale. They still face outcomes: longer interconnect queues, moratoria in hostile counties, and state bills like those covered in Baaja Senate permitting and transmission cost proposals.

Capacity markets in plain language

PJM runs two big markets that matter for this story. The energy market clears power for near-real-time delivery — prices spike when weather, outages, or fuel costs move. The capacity market pays resources to be available in future delivery years, based on forecast peak load. Data centers inflate the forecast because they are large, flat loads that utilities must plan around even when actual energy use is smoothed by batch scheduling.

Monitoring Analytics judged the energy market competitive in H1 2026 while calling recent capacity auctions not competitive, citing data center demand among the drivers. That split is why the $11.11/MWh attribution lives in the capacity column, not in a generic "AI tax on electrons" headline.

If you run training jobs only at night, you still affect planning peaks when a developer signs a 200 MW interconnect agreement. Grid accountants see the contract, not your cron schedule.

A worked example for finance teams

Suppose a mid-size lab reserves 5 MW of GPU colocation in Pennsylvania for two years. The colo passes through capacity charges as a fixed adder per MWh on top of energy. When PJM capacity costs per MWh rise 207% year over year in the monitor's H1 comparison, your all-in $/MWh for GPUs moves even if spot energy moderates.

The $11.11/MWh figure is not "multiply by your MW and you're done." It is an average attribution across the whole PJM footprint. Your contract might be above or below that increment depending on tariff class, hedging, and whether your host signed capacity before or after the latest auction.

Still, the direction is clear: AI load is now a variable in wholesale cost accounting, and monitors are publishing coefficients policymakers quote in hearings.

Retail bills vs wholesale statistics

Residential customers in PJM states do not see a line item labeled "data centers." Instead, utilities recover capacity obligations through general rate cases. The monitor's numbers are the evidence pack for those cases — the same way Google's nuclear uprate partnerships become talking points for "who funds new megawatts."

Community fights over dirty backup power at a new hall, like the Amsterdam protests we documented, often start from local air permits, but the economic argument increasingly cites regional wholesale curves monitors publish.

Comparison anchors outside PJM

PJM is not the only grid wrestling with AI load, but it publishes unusually detailed market monitor reports. ERCOT and CAISO face similar interconnect surges with different market rules. When you read a national "AI power cost" headline, check which ISO, which year, and whether the figure is energy, capacity, or both.

Nuclear cooling tower feeding server blocks as metaphor for low-carbon power feeding AI data centers

What builders should do with this report

1. Tag cloud region to grid operator. PJM is not MISO or ERCOT; cost stories do not transfer blindly.
2. Ask colo providers how capacity auction results flow into your MWh rate — some contracts lag auctions by a delivery year.
3. Model curtailment and flex revenue if you can participate in demand response; it is part of the new AI power stack.
4. Track state politics separately from this wholesale statistic; retail rate cases can move on different timelines.

How to read the headline numbers without overreaching

Three separate figures circulate in coverage of this report, and they answer different questions. The $11.11/MWh value is an attribution: it estimates how much of the monitored wholesale cost comes from including data center load in the capacity market. The 9.7% value expresses that same attribution as a share of the total. The 50.3% value is a year-over-year change in total wholesale cost, and it has many drivers besides data centers, including fuel costs and weather-driven energy prices.

Mixing these up produces bad claims. Data centers did not cause the whole 50.3% rise. The monitor's own framing is that capacity is the component where data center forecasts matter most, and that the energy market itself remained competitive. A careful reader should quote the number with its scope: capacity market, H1 2026, PJM footprint, monitor's methodology.

For journalists, analysts and policy staff, the practical test is simple. Ask which market the figure covers, which period it covers, and who pays it. If a source cannot answer all three, treat the claim as unverified until you check the primary report.

What to watch in the next capacity auction

Monitoring Analytics warned that price impacts are large and not reversible when forecast data center megawatts enter auctions before matching generation is online. PJM scheduled its next Base Residual Auction for June 2026 in the monitor's earlier Q1 language — after this post's H1 window but still within the planning horizon for 2027/2028 delivery years. If states or FERC act on bring-your-own-capacity rules before that auction, the $11.11/MWh attribution could shift in future reports even while absolute load grows.

None of that changes today's arithmetic: H1 2026 capacity accounting already shows a double-digit dollars-per-MWh data center adder on a monitored wholesale average above $114/MWh.

Related on explainx.ai

• Google and Voltus 1 GW VPP on PJM
• FERC rejects PJM curtail-first data center plan
• OpenAI Grid Protectors flexible demand
• Amsterdam data center moratorium protests
• Senate scrutiny of hyperscaler cost disclosures
• TSMC and memory supply chain stress

Primary sources: Monitoring Analytics PJM State of the Market (2026 Q1 report PDF) and PR Newswire press summary, accessed November 7, 2026.

Wholesale market statistics reflect the monitor's published H1 2026 analysis; retail bill impacts vary by utility and contract.