WASHINGTON — The challenges of meeting soaring forecasts of data center load growth dominated the Organization of PJM States Inc. (OPSI) Annual Meeting on Oct. 6-7.

PJM CEO Manu Asthana said much of the discussion has centered around reliability and affordability, but what is at stake is national competitiveness over the next century as the U.S. races to keep pace with China in developing artificial intelligence technology. Electricity supply is proving to be a significant bottleneck, he said, as China brought 428 GW of new supply online last year compared to the 49 GW completed in the U.S.

Senior Director of Market Operations Tim Horger laid out PJM’s latest proposal in the Critical Issue Fast Path (CIFP) process focused on large load growth: expediting interconnection studies for large generators, tinkering with voluntary load flexibility through price-responsive demand (PRD) and demand response, and creating more of a role for state utility commissions in reviewing the RTO’s load forecasts. He spoke on the first panel during the meeting, titled “Data Center Load Growth: Is further adaptation at the wholesale level needed?”

The expedited interconnection track (EIT) is designed to create a parallel study process for projects that carry a high certainty of reaching commercial service in a time frame that allows them to address the reliability gap, while minimizing the impact to the wider queue by limiting participation to 10 resources annually.

PJM also is considering requirements for large loads to provide financial commitments before they can be included in the load forecast, a proposition Horger said has been welcomed by data center developers. He said that builds on recent requirements that large loads obtain firm service agreements from their utilities three years in advance before their load can be included in the capacity market. Beyond those three years, he said the ability to have certainty that a particular service request will result in actual load growth becomes murkier.

Data Center Coalition Vice President of Energy Aaron Tinjum said the industry is supportive of expanding commercial readiness verification, such as requirements for electricity supply agreements; permitting reform for construction of new supply; standardization of submitting utility forecasts; and construction milestones for large loads. He said forecasting is foundational to the conversation, as it allows projects to proceed more quickly and with more confidence.

Independent Market Monitor Joe Bowring said he is amazed PJM has not attempted to exercise more authority over requests to adjust the load forecast it publishes, arguing that its stance abdicates the role of maintaining reliability to instead managing unreliability. He said PJM should implement a load interconnection queue that prevents large loads from coming online until they can be served reliably, with an expedited pathway for those bringing their own generation — a concept the Monitor is to present at the Oct. 14 CIFP meeting. He questioned whether it makes sense for PJM to allow large loads to sign up to receive service the RTO cannot provide.

While Bowring said improving the forecast is an important step in understanding the scale of the problem, he cautioned against spending too much time focusing on solutions that do not move the needle on ensuring new load is matched by capacity. He said Monitoring Analytics has been working to improve its own load forecasting, which has long relied on a bottom-up look at the next three years; that has been supplemented with a longer-term, top-down layer looking at the amount of large load that is reasonably expected to come to fruition across the U.S.

Looking at the availability of the chips used by the most power-hungry data centers and the amount of capital expenditure available to the industry, Bowring said about 60 GW of data center growth is expected across the country by 2030. That can be further divided across regions with sensitivities that assume that the share of large load growth will continue along existing projects or following trends in construction or announced projects, which creates a range of 22 to 26 GW of growth within PJM.

Horger said it’s not PJM’s place to call “balls and strikes” on which large load facilities are likely to be built and incorporated into the load forecast. Expanding the RTO’s role in developing the forecast would be complicated by the disparate requirements that states and utilities have on when large loads can be included in the forecasts submitted to PJM, with some requiring contracts and financial commitments.

Arnie Quinn, Vistra senior vice president of regulatory policy, said the Monitor’s proposal would effectively prevent new load growth until the 2030s and faulted PJM’s EIT proposal for requiring new resources to be sponsored by state utility commissions to qualify, which he said could put regulators in a precarious position. He said more focus should be put on who is bearing the risk associated with load growth, suggesting that more of it should be placed on load-serving entities signing up large loads by requiring them to procure capacity or pay a penalty.

He said ensuring the forecast is accurate would guide what forms of load flexibility PJM should pursue, arguing it would be a very different prospect for a consumer to enroll in a program when curtailments are to be expected every few years or much more regularly.

The Needs of AI vs. Cloud Computing

Aroon Vijaykar, Emerald AI senior vice president of strategy and commercial, said data centers appear as inelastic demand while investment in processing power remains high, but that is likely to moderate down the road and create more of an incentive for flexibility as power prices remain high. The large complexes expected to come online also have more of an incentive to explore the range of flexibility options available to them when compared to small consumers. Emerald develops software to allow AI load to be shifted across data centers to manage power consumption based on signals from LSEs and electric distribution companies.

The training phase of AI load tends to be less interruptible because of the risk of introducing errors to complex calculations, but reducing the response time on inference queries can deliver outsized reductions in load, Vijaykar said.

Bowring said the focus on load flexibility is an opportunity to rethink PJM’s market structures, arguing the PRD model isn’t well suited to the task, and the flexibility Vijaykar outlined could be the starting point for new market structures. He has often advocated for shifting DR to the demand side of the capacity market.

Tinjum said hyperscalers represent a small subset of data center usage compared to the shift to cloud computing, which often gets conflated with AI load growth. He said it can be difficult for data center operators who contract server capability out to smaller users, such as with cloud computing, to participate in load flexibility when their contracts require minimum uptimes. Backup generation can provide some curtailment capability, but diesel units can create a flood of noise and air quality complaints when operated for extended or regular periods in populated areas, such as Data Center Alley in Northern Virginia. The optionality and incentives for flexibility should reflect the diversity of data center users, he said.

For most data center developers and operators, the capacity and energy market revenues from DR participation are nice to have, but not a core focus, Tinjum said. If the program could be tied to interconnection timelines, that could provide much more value, he said.

During the OPSI Market Monitoring Advisory Committee meeting Oct. 7, Bowring said data centers should be required to bring their own generation and not be allowed to outbid regular consumers for capacity resources. He said it’s become a regular refrain to say the markets should be allowed to work to bring the generation needed to serve data centers, but some of the outcomes that could produce, such as blackouts or capacity being taken out of the market through bilateral contracts, are not functional solutions.

“The market can’t solve the problem of having 30,000 MW of capacity drop out of the sky,” he said.

Bowring said his statements should not be taken as advocacy for lower prices, but instead as an effort to find ways of applying cost-causation principles to the risks associated with data center load. He said those impacts already are being seen, with an analysis from the Monitor finding that the $175/MW-day price floor implemented in the 2026/27 Base Residual Auction (BRA) — which cleared at the $329/MW-day maximum — would have been relevant if data center load had been removed. (See PJM Capacity Prices Hit $329/MW-day Price Cap.)

David Mills, chair of the PJM Board of Managers, said the RTO is trapped in a “multidimensional Gordian knot” of trying to solve for price and hold reliability constant, while adding 20 to 30 GW of supply to serve data centers and controlling the associated emissions. That also is caught in a political challenge where some of the same voices advocating for lower prices are encouraging the economic development from data center development.

He suggested the impact to residential and commercial ratepayers could be controlled by states implementing bifurcated ratemaking systems.

New Jersey Board of Public Utilities Commissioner Zenon Christodoulou questioned if high energy prices and the rush for generation and interconnection equipment could be crowding out investment in the infrastructure required for electric vehicles, reshoring industry and electrification.

Mills and Bowring both said load growth outside data centers and some heavy industry remains fairly limited and unlikely to outpace the ability for the electric industry to respond without the added pressure from large loads.

“This is an outlier event in the sense that we’ve got all this new load coming in a short period of time, and your question is a valid one because it might eat up that surplus,” Mills said.

Future of the Capacity Market

The impact of large load growth also weighed on a pair of panels focused on speeding pace of new supply and the future of the capacity market.

Denise Foster Cronin, East Kentucky Power Cooperative vice president of federal and RTO regulatory affairs, compared the scale of data centers to adding a new zone to PJM, but without the requirement that a new entity seeking to join PJM demonstrate that it can procure the capacity it needs. She said LSEs should be active servers of their load and capacity auctions should return to their residual nature.

PJM Vice President of Market Design and Economics Adam Keech said the reliability backstop procedures may be worth revisiting. He said the trigger for the backstop — three consecutive BRAs that clear short of the reliability requirement — was designed at a time when the scale of load growth and reliability degradation was not envisioned. The backstop allows PJM to conduct a procurement process for transmission and generation owners to submit solutions to the reliability issue, including new generation to receive a multiyear commitment.

PJM Executive Vice President of Operations, Planning and Security Aftab Khan said the RTO is on pace to complete its transition to a cluster-based interconnection study process in April 2026, clearing tens of gigawatts worth of projects to proceed to development. There has been a slowdown in the pace of new generation coming online, with developers reporting issues around financing, permitting, siting and policy changes.

The majority of the resources in the queue and with interconnection service agreements that have not yet entered service are solar, which does not carry a high effective load-carrying capability rating. He said the RTO has major concerns if that is the only resource type coming online over the next few years.

“It’s important for PJM that we have the right generation portfolio mix,” he said.