PJM's Grid Is 6.8 Gigawatts Short — and Data Centers Are Driving the Crisis

On July 14, the PJM Interconnection — the largest power grid in the United States, serving 67 million people across 13 states and Washington, D.C. — finished its latest capacity auction. The result was a third consecutive reliability shortfall and a $16.4 billion payout that hit the federal price cap. Capacity prices stayed at $325 per megawatt-day for the 2028/2029 delivery year. And the grid still came up 6,831 megawatts short of its own reliability target.

That is not a market. It is a system running on emergency measures, with data center demand as the primary driver.

This story is infrastructure, not politics. The numbers are what they are. And they paint a picture of a grid being overwhelmed by AI compute demand faster than anyone — regulators, utilities, or the hyperscalers themselves — was willing to admit.

The Numbers: Third Consecutive Shortfall

PJM's Base Residual Auction for the 2028/2029 delivery year procured 138,318 MW of unforced capacity, plus 10,864 MW from fixed-resource requirement regions, for a total of 149,182 MW of supply. That is 6.8 GW below the level required to meet the one-event-in-10-year reliability standard. In plain English: PJM cannot guarantee the grid can handle a bad summer without rolling blackouts.

The auction cleared at the FERC-approved price ceiling of $325/MW-day, a 2.5% decrease from the prior auction's cap of $333.44/MW-day. Without the price collar — imposed after governors of all 13 PJM states petitioned FERC — clearing prices would have reached $554.72/MW-day across most of the footprint and $776.69/MW-day in the ComEd zone in northern Illinois. The total auction cost would have been $29.7 billion instead of $16.4 billion.

The timeline tells the story of how fast this escalated. In the 2024/2025 delivery year, capacity cleared at $28.92/MW-day. Four years later, the same capacity costs $325/MW-day — a 1,024% increase. And that is with the price collar suppressing the real market signal.

What Drove the Shortfall

PJM officials attributed roughly 2 GW of the increased demand forecast directly to data center development. But the independent market monitor's numbers tell a more complete story. According to Monitoring Analytics, data centers drove $6.3 billion of this auction's $16.4 billion total cost — 38%. Across the last four auctions, the cumulative total attributed to data centers is $29.4 billion, or 46% of $63.6 billion in total capacity costs.

The market monitor's director, Joseph Bowring, put it bluntly: "PJM is continuing to act like it's business as usual. You have to open your eyes and recognize that it is really a paradigm shift, and failing to do that imposes costs on other customers."

Demand response — resources that can be called on to reduce load during peak times — actually fell by 277 MW to 7,365 MW. When even the price cap doesn't stimulate demand response, something fundamental is broken in the auction design.

Meanwhile, PJM officials have warned that without urgent action, the grid could face a potential 60-gigawatt shortfall within a decade. Sixty gigawatts is roughly the entire electricity generation capacity of a country like Argentina. And it is being absorbed by one region's data center buildout.

The Cost to Ratepayers

The numbers matter, but the translation to real people's bills matters more. In the 2024/2025 delivery year, a typical household using 850 kWh per month paid roughly $2.30 in capacity charges. For the current pricing cycle, that figure is $17.10 to $17.40 per month. The NRDC projects that without a price collar, it could reach $70 per month by 2028.

For industrial customers, the impact is starker. A 10-MW industrial facility saw its monthly capacity charge jump from roughly $6,000 in 2024 to an expected $70,000 in 2028, according to Peter Cavan, head of strategy for Unison Energy. "Do not expect this to materially change in 2029/30," he noted.

The PJM states' governors asked for the price collar to protect ratepayers. It worked in the short term. But Julia Hoos, head of USA East at Aurora Energy Research, warned of a trap: "Lowering prices was definitely politically attractive in the short term, but now we're well on our way to facing an intervention doom loop." The collar suppresses the price signal needed to attract new generation investment, which means the shortfall persists, which means the collar stays in place. A loop that benefits nobody except the generators who get guaranteed above-market prices without having to compete.

What This Means: The System Is Overwhelmed

This is not a failure of the capacity market design. It is a failure to update that design for a world where a single data center campus draws as much power as a small city. The current PJM construct was built for gradual load growth — the kind that comes from population increase and economic expansion. It was not built for 2 GW of new demand in a single auction cycle driven by AI training clusters.

The market monitor's recommended fix is a separate capacity auction for large loads — data centers above a certain threshold would buy their own capacity instead of being subsidized by the broader market. Bowring called for it directly: "There's only one way to do what hyperscalers agree is the right thing to do, and that is to run a separate auction. That's good for the hyperscalers because it allows them to get capacity and be served reliably, and it's good for other customers because it separates out the impact from the data center."

The NRDC went further, calling for PJM to "weed out some data centers" from the capacity auction entirely and force them to pay for their own power plants and transmission infrastructure. Claire Lang-Ree, an NRDC advocate, stated plainly that PJM can solve the near-term problem "by removing large loads — like data centers — that have not brought their own new supply from the capacity auction."

This is the real debate that the auction results have forced into the open. The question is no longer whether data centers are straining the grid. The question is whether ratepayers will keep subsidizing the hyperscalers' AI ambitions.

What Comes Next

The next concrete milestones are clear. PJM is expected to file a backstop auction proposal with FERC in September 2026. The December 2026 base residual auction for the 2029/2030 delivery year will show whether any of the current pressure has eased. And FERC's August 2026 deadline for six grid operators to justify their large-load interconnection rules will force a regulatory reckoning.

The White House Ratepayer Protection Pledge, signed in March 2026 by Google, Meta, Microsoft, Amazon, OpenAI, Oracle, and xAI, was a non-binding promise. The real test is whether any of those companies will voluntarily enter a separate capacity construct — or whether PJM will have to impose one.

In the meantime, PJM is planning a backstop procurement to plug the 6.8 GW gap. It is an emergency measure for an emergency situation, but emergency measures have a way of becoming permanent when the underlying conditions don't change. And the underlying conditions — AI data center demand growing at a pace that outstrips every forecast — are not changing.

For the 67 million people served by PJM, the bottom line is this: your electric bill is going up because AI needs power, and the grid operator's market design was not built for this era. The question is not whether costs will rise. It is whether the people who are causing the increase will be the ones who pay for it.

— Allan Ali, Sylt.ing

===SUMMARY=== PJM's 2028/2029 capacity auction cleared at $325/MW-day price cap but fell 6.8 GW short of reliability targets — third consecutive year. Data centers drove $6.3B of the $16.4B total cost. Ratepayers face 7x capacity charge increases while the grid operator struggles to keep pace with AI demand.