PrintData center financing assumptions under scrutiny
BY SourceMedia | MUNICIPAL | 08:00 AM EDTRating agencies, analysts and stakeholders increasingly question the modeling assumptions underpinning data center financings.
In the Midwest ? where data centers are now proliferating, including in small, remote municipalities ? those questions raise concerns for issuers and investors.
Michigan, Ohio, Indiana, Illinois, Wisconsin and Iowa currently account for 147 of 926 planned data centers across the U.S. That's fewer than Texas' 149 and Virginia's 214 planned projects, but among the highest shares in the country, according to Cleanview.co, a power infrastructure market intelligence firm.
When a large load customer like a data center arrives, prompting a surge in electricity demand, utilities and municipalities make certain assumptions about projected load growth.
Those assumptions can impact transmission line upgrades, substation and interconnection facilities, land use planning, tax revenue projections, rate recovery and bond-financed infrastructure, among other things.
Often, in the modeling that supports those decisions, projected demand is treated as continuous. And some types of electrical load are modeled as fixed.
That means the financing treats the contractually signaled demand and the actual demand ? which in practice is decreasing as data center technology becomes more efficient and demand more intermittent ? as the same, even as the odds grow of a gap emerging over the life of the data center asset, said Neil Osnato, an infrastructure analyst and the founder of Persistence Analytics Group.
The risk for issuers and investors is not that these data center projects will fail immediately. It's that the gap between assumptions and actual demand will widen gradually, Osnato said.
What happens to the bonds tied to data centers when more efficient chips, artificial intelligence model efficiency or operational redesigns lead to reduced loads, impacting repayment assumptions that depend on projected load growth?
"In terms of AI training data centers, which are located in remote areas, we do think that those assets are exposed to overbuild and technological obsolescence risks because they lack alternate use cases," said Anubhav Arora, senior director of global infrastructure and project finance at Fitch Ratings, during a May 13 webinar on data centers.
"These overbuild and technological obsolescence risks could materialize in situations such as, let's say, if power efficiencies for AI training improve over time ? they're already improving ? or when differentiation between various AI models disappears, or in case some of the AI models for hyperscalers are not profitable and they're not able to monetize those," Arora said.
It varies by region whether states or municipalities control data center location decisions and resource management, said Dan Aschenbach, a utility financial and credit risk consultant.
"That's a risk for the city, if for some reason the model doesn't work out, and the load didn't come," Aschenbach said. "And that is happening now... My understanding is the efficiencies that some of these data centers are developing are going to help lower demand growth going forward."
He suggested measures like requiring power to be purchased directly before a data center is developed.
"These decisions at the local level, if they're not able to put in protective measures... (or if utilities build new generation and) growth doesn't appear, for whatever reason, they're stuck within a stranded cost and that gets shifted then to residential ratepayers," Aschenbach said.
Cost recovery mechanisms can help protect utility ratepayers, but they don't answer the question, "What exactly are we building around and for how long?" said Osnato.
He drew a parallel between data centers and the Brightline Florida situation, saying by email that in Brightline's case, there's a misalignment because "the financing structure was built around demand, revenue and capital assumptions that have not matured fast enough to support the debt profile."
The broader question, Osnato said, is whether public obligations are structured on assumptions that will hold up over time.
"There could be some mismatches that occur when the projections that are being used to justify borrowing money and the structured bond contracts do not align with how the project is actually going to perform," said Deanna Malatesta, a professor at Indiana University's O'Neill School of Public and Environmental Affairs and an expert on public-private contracts who co-authored the book "Contracting for Public Value."
Right now, data center contracts "are not anywhere near as transparent as they should be," she said. "It's difficult to actually compare and to determine what the real issues are, because they like to make sure there's, to be quite honest, a lack of transparency."
The industry is "more in control than anyone else right now," Malatesta said, "and the municipalities are kind of jumping at the opportunity and accepting on their face promises, for instance, of keeping a certain demand load, or actually increasing the number of people hired."
She said she's seen some utilities move away from cost-of-service models and implement fixed payment schedules instead. Utilities are also starting to require take-or-pay contracts.
"That seems to be one way they're approaching trying to protect themselves, but again, I'm not sure it really covers the risk in any kind of acceptable way," Malatesta said.
Some of these questions are being addressed on a state-by-state basis, especially in states that are seeing high load growth, said Tom Falcone, president of the Large Public Power Council, an association of 29 public utilities.
"It starts with a good load forecast, and the challenge of load forecasting when it's large data centers is? they come to town and they use huge amounts of energy, so rather than 20 megawatts, they may ask for 400. ? So you're going to be building infrastructure for that demand," he said. "And then the second related challenge is that all the information is on the other party's side of the table."
Utilities have started asking for long-term contracts and seeking large load tariffs to ensure the parties driving projected load growth will pay for the infrastructure buildouts involved, Falcone said.
The contracts with hyperscalers involve rate classes, and "we're going to charge you the appropriate price for that rate class as we go through rate processes," he said. "And if you come to town and you request 400 megawatts, you're on the hook to pay for maybe 80% of those (costs), whether you use it or not."
A better way to address risk would be to build in formal relational contract mechanisms, Malatesta said. Formal relational contracts prioritize transparency and sharing risks and rewards so that all parts of the contract are enforceable, she said.
Those frameworks draw on ideas from behavioral economics about relational norms and what really prompts contract compliance, Malatesta said. They replace vague good-faith commitments with explicit risk-sharing formulas.
"Perhaps a pre-agreed mathematical formula that automatically adjusts rates and capacity fees, or tax payments based on deviations from, say, some baseline assumption," Malatesta said. The idea is "sharing both gains and pains," and hedging risks to protect taxpayers.
"At the end of the day, that's probably what's going to happen, where this is lopsided," she said. "It's going to end up being taxpayers who end up paying, or ratepayers."
Due to the size of these data centers, Aschenbach said, technological efficiencies and other changes, like projects canceled due to public outcry, also raise credit risks.
"If you've planned now for new generation to meet that growth, and then it doesn't come, that type of concentration risk... that's a big credit risk right now," he said.
There are different levels of risk underwriting that take place on data center deals, said Charles Renner, a partner based in the Kansas City office of Husch Blackwell, who has worked on both the developer and municipal sides of data center projects.
"On the utility side, we are seeing a consistent, steady increase in the underwriting that is taking place for each, almost like per data center ? based on location, existing capacity, anticipated growth patterns in that area and the presumed load demand that the data center developers are providing," he said. "There's a fair amount of information exchange."
He said it's not yet universal to have a floor as well as a ceiling on rate payments from a data center, but he has seen an evolution over the last two years.
On water and wastewater, he said, the same trend applies. Sometimes tax incentives are part of the picture.
The "generation of property tax revenue off a data center site, a portion of that is targeted to absorb the long-term costs of the water infrastructure," he said.
The expectation is that when a data center opens, it generates a certain amount of tax revenue, and "maybe 20% of (that) is focused on the infrastructure," Renner said. "If it performs at the level that's anticipated, there's economic growth everywhere, but you're talking about what happens when the usage is lower, and those issues are there, and I certainly see that."
Data centers support "a very fast-moving industry," he said, and he "wouldn't be surprised to see some version" emerge of what Malatesta proposed, particularly around infrastructure financing.
In the Fitch webinar, Arora raised power supply risk and construction risk for related infrastructure, implying utilities are currently struggling to meet projected demand.
"We need interconnection, which we know that there are massive queues in the U.S. right now for interconnection," he said. "And then there is substation construction and construction of other power-related infrastructure. We're already seeing some delays in terms of substation construction where utilities are backlogged."
Falcone said utilities are taking a gamble when they build out infrastructure for data centers.
"How are we going to project the 20-year trend on (efficiency), or better cooling technology, or more efficient chips?" he said. "But if we're making 30-year investments, 40-year investments, they can't be off of short-term or no contracts."
He pointed to his recent testimony before the U.S. House of Representatives' Committee on Energy and Commerce, in which he said LPPC members plan to invest $166 billion in new infrastructure and build 59 gigawatts of new generation to meet rising demand over the next decade.
"Forecasts are inherently uncertain when a few very large customers drive the projected growth. Utilities and regulators need to distinguish committed load from probable or speculative load," Falcone said in the testimony.
A footnote to the testimony notes that in August 2025, "Google
Malatesta said data center projects involving public financing should require a high degree of transparency.
"In any kind of democratic government, we certainly need a situation where people know what they're voting for or approving," she said. "The longer the term of the contract, the more uncertainty there is."
At a municipal level, "risk managers need to be right front and center helping to make these decisions of what a city will do about providing energy to a data center," Aschenbach said. "At some point things could fall apart."
He added, "There will be somebody that's going to be in trouble."
