PJM’s capacity auction just put AI’s grid bill in public view

The most important climate-technology development today was not a new battery chemistry, a solar efficiency record, or another promise that artificial intelligence will optimize the grid someday. It was a bill.

PJM Interconnection, the grid operator serving 13 eastern states and the District of Columbia, released capacity-auction results showing that its 2028/2029 capacity price cleared at $325 per megawatt-day across every listed pricing point. PJM’s results spreadsheet shows 138,317.8 megawatts of annual-equivalent resources cleared in the base residual auction. Multiplying those two numbers by 365 days gives roughly $16.4 billion in annual capacity payments attached to this auction cycle before retail allocation, utility-specific pass-throughs, and other charges. Reputable reporting by The New York Times, republished by The Philadelphia Inquirer, says the latest auction will add about $6.3 billion in costs to consumers and businesses over the next three years, an estimate attributed to Monitoring Analytics, PJM’s independent market monitor.

That is the change: the AI-and-data-center electricity story moved from projections into a price signal that households, small businesses, factories, schools, and local governments will eventually recognize on utility bills. Capacity markets are not the whole electricity bill. They are the grid’s insurance premium: payments to resources that promise to be available when demand peaks. But when that premium jumps or stays pinned near a cap, it is a warning light. The system is saying that dependable supply, transmission deliverability, and demand flexibility are not keeping pace with load growth.

PJM framed the result plainly. In the Inquirer’s account, PJM President and CEO David Mills said the auction results show electricity demand continuing to grow faster than electricity supply. That is the sentence to watch. It is not a software metaphor. It is steel, turbines, transformers, substations, interconnection queues, gas availability, battery duration, transmission corridors, permitting, and customers big enough to change a regional load forecast.

What changed

PJM’s public auction files show three numbers that matter for readers.

First, the clearing price: $325 per megawatt-day for the RTO and all listed locational deliverability areas in the 2028/2029 base residual auction. PJM’s planning-parameter report says that $325/MW-day is the FERC-approved price cap for this auction. In other words, this was not a free-floating market clearing at some lower equilibrium; it hit the ceiling established for the auction.

Second, the cleared amount: 138,317.8 MW of annual-equivalent capacity cleared in the base auction. A megawatt of capacity is not the same as a megawatt-hour of energy. Capacity is a promise to be available; energy is actual electricity generated or saved. But the capacity number is still a reliability signal because it reflects what PJM was able to procure to meet future peak conditions.

Third, the demand backdrop: PJM’s planning report listed a 2028/2029 forecast peak load of 165,953.5 MW, up 1,374.5 MW, or 0.8%, from the prior auction’s forecast peak load. It also listed a PJM RTO reliability requirement of 156,012.9 MW before adjustment for Fixed Resource Requirement obligations, up 3,612.9 MW, or 2.4%. The public spreadsheet reports a separate 145,149.1 MW RTO reliability requirement line in the load-pricing sheet and 138,317.8 MW of base UCAP obligation. The exact retail effect will vary by utility zone and customer class, but the direction is not ambiguous: PJM is pricing scarcity.

One quick translation: the reported $6.3 billion in additional costs, spread across PJM’s roughly 67 million people over three years, is about $94 per person before any business-vs-residential allocation. That is not a forecast of an individual household bill, because utilities recover capacity costs differently and commercial loads carry a large share. It is a scale check. This is not pocket lint. It is a regional infrastructure bill arriving through electricity markets.

Why it matters for climate tech

The clean-tech lesson is uncomfortable but useful: electrification does not decarbonize by magic. Electrification works when clean supply, flexible demand, wires, storage, and market rules expand together. If one part races ahead, costs and emissions can leak out somewhere else.

Data centers are the obvious accelerant. Northern Virginia is already the world’s largest data-center cluster, and PJM covers that territory. AI training and inference loads are not the only reason PJM is tight. Retiring fossil plants, delayed interconnections, transmission constraints, winter reliability concerns, and rising peak demand all matter. But data centers are unusually consequential because they can add large, concentrated, around-the-clock loads faster than the grid can build new deliverable resources.

This is where greenwash enters wearing a hard hat. A hyperscale company can buy renewable energy certificates, announce a power purchase agreement, or fund a solar farm and still increase local reliability costs if its facility needs firm power at a constrained node before transmission, storage, and dispatchable capacity arrive. A megawatt-hour certificate is not the same as a megawatt of dependable capacity at 6 p.m. during a heat wave or a winter cold snap. Matching annual clean-energy purchases does not necessarily relieve a local substation, a transmission import limit, or a capacity shortfall.

The auction also complicates the usual “data centers will pay for themselves” pitch. They do create tax revenue and construction jobs. Some can anchor new clean-energy procurement. But if regional capacity costs rise by billions and those costs flow through to captive ratepayers, the public is effectively underwriting part of the reliability envelope. That does not make data centers inherently bad. It means the bargain needs to be explicit.

Who is affected

PJM serves all or parts of Delaware, Illinois, Indiana, Kentucky, Maryland, Michigan, New Jersey, North Carolina, Ohio, Pennsylvania, Tennessee, Virginia, West Virginia, and the District of Columbia. The people most exposed are not just AI users or data-center neighbors. They include low-income households on fixed incomes, small businesses with thin margins, manufacturers with high electricity intensity, school districts, hospitals, and water utilities.

The geographic effects will not be even. PJM’s results show the same $325/MW-day clearing price across listed zones, but the load obligations and local grid constraints differ. The planning report flagged import constraints and transmission-upgrade delays in parts of the system, including Dominion-related upgrades in the eastern PJM footprint. That matters because the data-center buildout is heavily concentrated in Virginia and nearby markets. When load grows in a constrained place, the consequences can spread through the market even if the servers are not in your county.

Communities near proposed data centers are also affected before the bill arrives. They face land-use changes, backup-generator emissions, water demands where evaporative cooling is used, noise, tax-abatement debates, and the risk that promised clean power is measured on an annual accounting sheet rather than at the local grid node. Communities are infrastructure stakeholders, not scenery around a server farm.

Generators are affected too. A $325/MW-day capacity price is a strong revenue signal for existing plants and new resources. That can help keep reliability resources available and may improve the economics of batteries, demand response, uprates, and new generation. But it can also keep fossil capacity profitable if cleaner firm alternatives and transmission do not arrive quickly. A high capacity price is a symptom; it is not automatically a clean-energy policy.

What readers should do

First, treat “100% renewable” data-center claims as incomplete unless the company discloses where and when the power is matched, whether it is hourly or annual, what capacity or flexibility it brings, and whether local grid upgrades are paid by the project or socialized across ratepayers. Ask for megawatts, megawatt-hours, location, timing, and deliverability.

Second, watch state utility commissions and PJM stakeholder proceedings, not just company press releases. The real decisions will be in interconnection rules, transmission cost allocation, large-load tariffs, standby-service requirements, demand-response participation, and whether new data centers must bring clean firm capacity or flexible load commitments with them.

Third, separate clean energy from grid adequacy. Solar and wind are essential and usually cheap, but PJM’s scarcity signal is about dependable capacity at peak risk hours. Batteries help, especially for daily peaks, but duration, charging windows, winter performance, and interconnection location matter. Demand response from data centers should be on the table: if a facility can shift nonurgent computing, use onsite storage, or throttle workloads during grid emergencies, that has real system value. If it cannot, it should not market itself as a flexible climate asset.

Fourth, look for who pays. A fair large-load policy should make new, unusually large customers cover the grid upgrades and reliability costs they impose, while still allowing genuinely beneficial projects to connect. The worst outcome is a quiet transfer: private AI revenue, public grid risk.

The bottom line is not that AI data centers should be banned everywhere or that capacity markets are villainous. The bottom line is that climate-tech credibility has to survive contact with the utility bill. Today’s PJM result says the grid is becoming the limiting reagent for the AI buildout. If developers want to claim they are powering the future, they need to show not only clean megawatt-hours on paper, but deliverable, reliable, locally accountable power in the places where the load actually lands.

Sources

• PJM Capacity Market (RPM) page and 2028/2029 Base Residual Auction files
• PJM 2028/2029 Base Residual Auction results spreadsheet
• PJM 2028/2029 Base Residual Auction planning-parameter report
• The Philadelphia Inquirer/New York Times: Data centers to add billions in power costs in 13 states, including Pennsylvania and New Jersey
• Reuters: PJM power grid auction hits price limit, falls short of reliability goal

---

Shadowfetch is a technology publication. Explore Shadowfetch Linux — our own Linux build — and the Shadowfetch apps on the App Store.