WASHINGTON — Uncertainty in load forecasts presents competitive firms and regulated utilities with polar opposite incentives, NRG Vice President of Regulatory Affairs Travis Kavulla said at S&P Global’s Nodal Trader Conference on Oct. 24.

“Data centers and the consequences of unbelievable load forecasts are really not good,” Kavulla said. “On the power generation side, it could actually, paradoxically, lead to underinvestment in power generation that relies on market forwards. Investors and companies like mine might just throw up their hands and say, ‘This is so unbelievable. We’re standing pat. We don’t want to overbuild.’”

While companies that operate purely in the markets might wait to see how much demand actually shows up — which could prove overly cautious — utilities have different incentives, he said.

“It might induce overspending and overbuild because of the moral hazards and financial incentives that are present on that side of the industry,” Kavulla said. For “people who, unlike merchant generation, don’t have to wear the risk of the spending that they include in rate base and have a captured base of customers, there will always be someone there to pay for it, even if data center growth doesn’t materialize.”

The situation also can induce a crisis for politicians, which risks laws being passed that compound the problems facing the industry, he added.

“What PJM is really asking the community for generation to do is to add more than the amount of generation — about 40,000 MW — that was added during the entire turnover of the shale revolution, of coal to gas, in a shorter period of time than that generation got online when supply chains were less challenged,” Kavulla said.

Other markets like ERCOT also have huge projections for future demand, but forward prices remain low. Forward prices in PJM shot up right after the RTO released its first load forecasts that included booming demand from data centers and other large loads, but they fell shortly after.

Capacity prices have shot up in PJM over the past two auctions, but the price impact of new load coming onto the system is absent in energy forwards, Kavulla said.

“We also have ERCOT, which shows similar activities,” Kavulla said. “No capacity market here to remove some of the pressure of trading off load growth. Instead, you would expect, certainly, given what we saw in the previous projection of demand growth well exceeding supply additions, for the market to be significantly impacted, and that’s just not the case.”

The situation has many explanations, but three of them stand out for Kavulla, including that load growth is fake. But even a fraction of the projections becoming a reality should push forward prices up given how tight many markets are and the well documented issues getting new supply on the grid.

Another option is expecting politics to intervene rather than letting prices rise enough to incent a market response to add more supplies, he said.

“There will be some deus ex machina of policy that could either be ‘bring your own generation’ that causes entry to happen outside of the energy price fundamentals,” Kavulla said. “There could just be a price cap on energy. There could be government funding or rate-regulated entry of generation. Something will happen that causes the energy markets not to be doing the lift associated with the load growth projections.”

The other issue is that the markets themselves do not lead to deals that far in the future with competitive retail contracts lasting one to five years.

“There might be a lot of people wanting to sell you something on a longer daily basis, but not a lot of people willing to buy something on a longer daily basis,” Kavulla said. “All things being equal, that ends up working to the counter effect of incremental demand outpacing incremental supply.”

All three can be true at the same time, and Kavulla argued the situation can be fixed with more of a top-down approach to large loads.

Data center developers and others currently bring plans to utilities, which collect them and are aggregated across ISO/RTO markets. But Kavulla said it would work better for markets to publicly announce available headroom and then request large loads apply to use it. The Alberta Electric System Operator has put that idea into practice.

“We are going to say we have X megawatts of capacity available in the Alberta market today, and we are going to tender that out and have data centers who are interested in developing quickly in this province subscribe it,” Kavulla said. “In a series of literally three stakeholder meetings in the Alberta regulatory process, they kind of rolled out this design.”

The approach is similar to how the natural gas industry gets customers to sign up for new pipelines, which shows FERC in its regulatory process that there is a need for new capacity, he added.

Capacity markets worked reasonably well at incenting new generation in the past, but they need some reforms to deal with the current paradigm, Concentric Energy Advisors CEO Danielle Powers said during a panel Oct. 23.

“When we deregulated, we lacked the planning function — period,” Powers said. “It used to be that the utilities plan, and they planned for the right mix at the right time, and that was all left to market forces, largely. And so, I think you can have both.”

Regulators and ISO/RTOs instead have focused on “nibbling around the edges.” While resource accreditation is important, it is vital that the grid get the right mix of resources online so it can operate reliably, she said.

The markets have been adding supply but not enough generation that also can provide key grid balancing services, said NERC Senior Vice President Mark Lauby.

“We have solar panels, but they don’t always provide the kind of meat and potatoes of frequency and mass that we expect,” Lauby said. “So how are we going to supplement that?”

NERC has been studying large loads for 18 months, and their proliferation has major implications for how the grid will be operated in the future, Lauby said. The grid always has been operated so it can absorb losing a large generator and, in some cases, large loads going offline instantly.

“Now we’re talking about losing the city of San Francisco at one time,” Lauby said. “So how do we manage that? What are some of the adjustments we need to make in the system? What are the interconnection requirements so we can make sure that the system remains reliable?”

NERC is considering new standards to ensure the system can be operated reliably, and in its stakeholder-driven mandatory standards process, that could involve the large loads helping to develop new rules.

Part of the issue is misaligned timing: If a regulated utility had gone to a state commission and asked to build a couple of gigawatts on speculation just two and a half years ago, they would have been laughed out of the room, said Abram Klein, managing partner at Appian Way.

“We’re at a little intermediate period where the markets have gotten a little bit tighter,” Klein said. “But there’s a lot of positive sides. I mean, the solar plus wind plus batteries is working much better than expected.”

One area that could stand improvement is the demand side, with studies from Duke University’s Tyler Norris and Goldman Sachs saying that could help the grid meet rising demand from large loads. But that needs to be priced into the market, as opposed to the DR in PJM that suppresses energy prices whenever it is called on, Klein said.

Traditional DR will not work with data centers because of the “rebound event,” where if 4 GW of data center load is asked to go offline peak, it will just create a new net peak whenever it is allowed back on the grid, said Arcus Power’s Wish Bakshi, lead data and AI consultant.

Data centers can offer flexibility, but it would work better to “underclock” chips, which Bakshi compared to operating a V12 like a much less powerful V4 engine.

“You turn it down, so it keeps running and training, and at 8 p.m. or 9 p.m., they basically kick back up, so you don’t have that crazy spike, and you’re basically ramping up very slowly,” Bakshi said. “That’s kind of how you do it with these AI data centers.”

The other option is just to pay high power prices to keep the data center running, which can make economic sense given that its owners invested tens of billions of dollars just in the high-end Nvidia chips that make it work, Bakshi said.

Another option is shifting compute load to another data center, but that comes with its own limits.

“There’s only so much fiber optic cable that go across the country that can actually do that,” Bakshi said. “Is it possible? Yes. But only the big shops can do that. Google’s already working on it.”