PJM 모델링 오류로 요금납부자 120억 달러 낭비

$12B of US ratepayers' money wasted on a modeling mistake and PJM wants to do it again American Grid design needs an overhaul, Why it is good to be full of cold air. Robert Boswall , Reyk Knuhtsen , Jeremie Eliahou Ontiveros , and 2 others Aug 16, 2026 ∙ Paid 69 5 Share Earlier this year, we explained why residents of the PJM area, America’s largest electricity market, have seen their power bills rise by ~20%. We argued that the main culprit was PJM’s auction design choices and how PJM models demand and supply. To better understand the extent of the problem, our Energy Model team spent the last 6 months reverse-engineering PJM’s main system model, the ‘Reserve Requirement Study’, which to date has been a black box. This study is how PJM decides what type of and how many power plants to buy to make sure electricity is reliable, using an annual auction, and spending billions each year. Armed with a reconstructed model we argue that the problem is worse than we thought. PJM’s model includes errors that we estimate have cost all of its 66 million residents a total of $12B between 2025 and 2027 alone. We share our method in the annex of this newsletter for our subscribers; as well as the results of the model in our PJM Model dashboard . Our live rebuild of the Reserve Requirement Model is a tab available exclusively to our Energy Model clients; which also includes a quarter-by-quarter forecast of the whole US grid tracking >40,000 grid-connected power plants, and every single behind-the-meter datacenter power order. PJM’s model is structurally anti-growth with a poorly designed capacity market that is globally unique and a governance system that is too big to function. These failings magnify the negative impact of bad system modeling, which is the focus of this report: PJM underestimates by ~4 gigawatts the existing power plants it already has; owing to a methodology which doesn’t account for the higher efficiency of power plants in winter and improved power plant resilience since Storm Elliott. PJM has wasted ~$12 billion of ratepayers’ money from 2025 to 2027 due to this weak methodology, which dramatically overstated the supply/demand shortfall it faced. Household electricity bills would have risen much less if PJM’s model was accurate. PJM’s emergency auction is putting ratepayers at risk by signing contracts for too much power without committed counter-parties. Using PJM’s demand and supply curves we calculate that better modeling would have resulted in $6.7B of savings with only 0.014GW (yes, 14MW) less power procured for 2025/26; then $4.9B and 0.8GW for 2026/27. More power meaning less in savings might be counter-intuitive but we have the supply and demand curves to show how these disproportionate impacts occur. PJM has forced itself into operating at the limit, so inaccurately modeling power plants’ capacities has a massive impact on auction costs. PJM has constrained its own supply of new power by having one-year contracts that start too soon after they are signed, as well as being slow to connect these new plants to the grid. This means fulfilling any demand growth requires paying a big premium for new power plants to be built at unrealistic speeds. PJM also runs the only capacity market in the world that does not distinguish between new and existing power plants. So that premium paid for new power plants is also paid to existing power plants for doing nothing. The same growth in a market that splits new and existing power plant auctions avoids the volume effect of repricing the entire fleet; which means ratepayers are better protected from price spikes. Despite four record-breaking auctions costing $63B, PJM will be short of the amount of generation it needs to run reliably, and plans to run an “emergency auction” from September 30th to October 21st with results by December 2nd. SemiAnalysis concludes there is an additional 3.8GW of reliable power on PJM’s system by taking into account increased turbine efficiency from cold air, and reduced risk of winter failure after federally mandated asset winterization investments. This is the equivalent of eight large gas power plants which would cost around $10 billion to build today . This 3.8GW would negate 56% of the 6.8GW that PJM plans to procure in its emergency auction. PJM’s emergency auction presents serious risks, it will sign contracts running to 2043 that are supposed to be paid for by new large loads. These are not sales of electricity; the plants get paid just for existing, whether or not the forecast demand ever arrives. But PJM is doing this with no committed counter-parties . PJM does not have its own money, everything it spends is from ratepayers. Every PJM state will have to pass (but none have done so yet) their own cost-allocation policy to pass on costs to participating new large loads. But those large loads can opt out if they contract their own additional power plants directly. If they participate there is little clarity on how much of the auction’s capacity they are expected to pay for. Participation brings no other benefits, like accelerated interconnection timelines. Many datacenter developers have written off PJM regardless, going elsewhere or planning behind-the-meter power configurations. If no other counter-party emerges, then once again those left holding the bag will be the residential ratepayers. In this report, we dive into PJM’s capacity market and introduce our proprietary reverse-engineered model. We believe that PJM fails to accurately model two critical supply characteristics, despite key stakeholders recognizing them to be both real and of importance: PJM does not credit how cold, dense air makes gas plants stronger electricity generators by up to 25% in winter; and PJM’s risk calculations have not been updated to recognize that as long ago as 2024, 400 of PJM’s ~700 gas plants had invested in winter reliability measures in response to the outages from Storm Elliott. There are deeper problems at PJM than just modeling. Its current auctions were designed two decades ago, and it has not experienced demand growth since. Today PJM is structurally anti-growth. PJM is failing to adapt to this new era of load growth because of a governance deadlock , a vetocracy of vested interests. If it cannot achieve market reform perhaps it can stem the ratepayer bleeding by improving how capacity is modeled, especially when winter is coming. This report will dig deep into PJM’s capacity market. After providing some introductory context, we adopt the following structure: We start with the fundamentals and explain how PJM models supply, demand, and reliability risks. We cover their major modeling methodology change in 2024 from EFORd to Reserve Requirement Study and the implications of that change. We introduce our reverse-engineered model and cover in greater depth how we calculate the role of various resources, outages, reliability, and fleet size. We get to the core issue: PJM’s lack of consideration of the increased throughput of gas power plants in the winter, and their improved reliability through Asset Winterization. We then dive into the pricing impact. PJM’s capacity auctions are highly complex and we explain why and how PJM’s 66M residents’ electricity bills were $12B higher than they needed to be. Lastly, we provide some thoughts on the future of PJM and its upcoming Reliability Backstop Auction. This is an emergency mechanism hoping for datacenters to pay for their own capacity. We believe that ratepayers could, again, be at risk. Let’s go. This report is a collaboration between SemiAnalysis and Nathan Iyer in his personal capacity. Context If you have not been following along, let us catch you up: Power bills in PJM states jumped 17-24% in June 2025, and the state regulators labeled the capacity auction as the main driver. Read our PJM vs ERCOT article to understand PJM’s capacity auction, and how singular it is relative to other energy markets. PJM constrains its own