The growing demand for AI and data centers is pushing the U.S. power grid to its limits. The latest capacity auction by PJM, the largest grid operator in the United States, shows that capacity prices for 2027/28 have hit the upper limit of $333.4, while simulated "shadow prices" suggest that the true costs could be 60% higher. Goldman Sachs has warned that if future auctions remove the price cap, electricity bills could double, forcing the United States to face an extreme test of choosing between advancing AI development and maintaining basic power reliability.
The latest capacity auction results from PJM Interconnection, the largest grid operator in the United States, reveal a stark reality: the surge in artificial intelligence and data centers is pushing the American power system to its limits.
In the 2027/2028 Base Residual Auction that concluded on December 17, the price of generation capacity in the PJM market climbed to $333.40 per megawatt-day (MW-day). This price not only broke the record set in July but also reached the cap approved by the Federal Energy Regulatory Commission (FERC). By comparison, the average price between 2017 and 2024 was just $108 per MW-day, highlighting the intense competition for power resources.
However, the most shocking data in this auction report is hidden in the appendix. Detailed information shows that if the price cap were removed, the simulated market-clearing price would have been as high as $529.80 per MW-day. This means that in a fully unregulated market environment, the enormous demand brought by data centers would have driven electricity prices up by nearly 60% above the current capped level.
In its analysis of this auction, Goldman Sachs noted that while the prices were in line with expectations, the procured supply volume once again failed to meet reliability requirements, exacerbating concerns about resource adequacy. With declining reserve margins and a lack of new capacity, the U.S. power market is facing a difficult trade-off between 'AI development' and 'grid stability,' posing significant upward risks to future electricity bills.
Surge in Data Center Demand and Supply Bottlenecks
The rapid expansion of the U.S. data center industry is tightening the power market significantly. Data indicates that in November, U.S. data center power demand capacity increased by 1.6 GW in a single month (a 4% rise month-over-month), far exceeding the average monthly growth of 0.26 GW from 2017 to 2024, bringing the total U.S. data center capacity to 44.6 GW.
The PJM market covers key regions, including Virginia’s 'Data Center Alley,' which is at the epicenter of this surge in demand. Among all regional power markets, PJM recorded the largest increase in capacity, reaching 0.45 GW, of which Virginia alone contributed 0.27 GW. This rapid growth in demand directly led to the spike in real-time electricity prices last summer and drove up capacity prices.
Despite strong demand, the response from the supply side has been unusually sluggish. Goldman Sachs pointed out that although Trump promised to revive nuclear energy, actual new capacity has been slow to come online. In this auction, only approximately 350.7 MW of new generation capacity and about 423.6 MW of upgraded capacity were procured, with a total cleared capacity of around 134,747 MW. This reflects that even with high prices, there has been no significant stimulation of new capacity entering the auction process.
The Hidden True Costs
A core risk revealed by this PJM auction is the 'shadow price.' Simulated data in the report appendix shows that without regulatory-imposed price caps and floors, all prices for 2027/2028 (except for the DOM LDA region) would clear at $529.80 per MW-day.
This figure is not only 60% higher than the actual clearing price but also significantly exceeds the $388.57/MWh-day in the simulated auction for 2026/2027. This implies that, under the same conditions, the expansion of data centers could potentially drive up electricity costs by 40% within a year.
More concerning is that even under the assumption of complete deregulation with prices surging to $529.80, the additional cleared capacity attracted by the grid would only be about 800 MW more than the current level. This indicates that the largest power grid system in the United States faces hard physical constraints—no matter how high the price, it cannot 'generate' tens or even hundreds of gigawatts of new electricity needed to meet the cyclical demands of AI and data centers in the short term. It is estimated that if not constrained by price controls, related wholesale electricity costs would increase by $9.9 billion.
Deterioration in Grid Reliability Metrics
The auction results further confirm the deterioration in grid reliability metrics. In this auction, the reserve margin dropped from 18.9% in the previous instance to 14.8%, falling approximately five percentage points below the target reserve margin of 20%. The decline in this key indicator directly signals supply tightness.
Goldman Sachs noted that although the increase in Effective Load Carrying Capability (ELCC) values (from 69% to 92%) drove demand response to 1,847 MW in this auction, factors such as increased battery resources failed to fundamentally reverse the supply shortage situation. Regarding resource types, the generation resources cleared in this auction were largely consistent with those in the previous one, indicating structural stagnation.
Goldman Sachs warned that the next auction, scheduled for June 2026 for the delivery years 2028/2029, faces significant uncertainties since no mandatory price caps or floors have been set by FERC for this round.
If projected based on the current simulated uncapped price growth rate, PJM's price in the next round could fall within the low-to-mid $600/MWh-day range, effectively doubling the current level. Goldman Sachs believes that unless an upper limit is set for the next auction and market conditions ease, utility bills will face extremely negative impacts. As market observers have pointed out, unless structural changes occur before next year’s auction, the United States will face the dilemma of having to choose between 'AI development' and 'air conditioning (AC) electricity usage' by 2026.
Editor/jayden