Serving U.S. electricity demand growth needs new solutions by Stephen V. Arbogast

Like a creature arising from some lagoon, a new challenge has burst upon the electric power business.  AI data centers, hurrying to capture the burgeoning market for artificial intelligence, are scouring the country for sites.  When they find one, they look to local providers for large blocks of highly reliable power – 150 MW, 500 MW even 1 GW per site.  Cloud computing data centers and a renaissance in domestic manufacturing are also adding to load growth.  Little of this demand was anticipated as recently as a year ago.  Consequently, state & utility Integrated Resource Plans (IRPs) suddenly became outdated.  Those plans, which foresee massive new investments to replace carbon-fired facilities, already looked challenging.  With the new ‘load’ outlooks, their successful execution may have transitioned from challenging to improbable.

Several dimensions characterize this execution challenge.  There are always risks associated with engineering, contracting, and constructing a large number of generating facilities.  Today, before those issues can even be addressed, a series of ‘upstream’ challenges must be faced:

1. How will the additional AI-serving facilities be financed, given everything else utility balance sheets will be funding to achieve Net Zero 2050?
2. Will this unanticipated demand growth complicate plans to de-carbonize electricity by adding to the huge capital investment programs in existing Net Zero 2050 plans?
3. What will happen to the price of power as this mass of new facilities comes online?
4. Finally, will retail customers be subsidizing data center electricity prices? Presently, large industrial customers pay prices as much as 50% below the cost of retail power.  Will that practice be continued for AI’s new ‘load?’

These questions are especially acute for states such as Virginia, North Carolina, South Carolina, and Georgia.  These states offer attractive locations for data center sites.  All can provide low- cost land and labor.  Several offer rich incentives for businesses to locate there plus low taxes.  Some also have committed to Net Zero 2050 plans; these appeal to AI firms eager to burnish their low carbon credentials.

Finally, all four states offer ‘regulated power markets’ where utilities can ‘rate base’ new facilities.  New generation plants are thus included in a required revenue calculation, which is then used to set power prices targeting an equity return for the utility.  Historically, these regulatory structures have effectively incentivized the funding and construction of new facilities.  It thus is no coincidence that AI data center site searches have gravitated to the Southeastern states and their regulated power markets.

Upon closer inspection, the arrival of AI/data center/manufacturing load is raising a raft of issues.  Georgia Power is dealing with the massive cost overruns at its Vogtle II & III nuclear plants.  These have not only strained the company’s balance sheet but required a 12%, $1.8 billion power price increase in 2022.  Georgia’s rate payers and the state’s utilities commission will not welcome demands for additional price hikes to fund unexpected AI load growth.  South Carolina is feeling similar effects given the failed construction of the Summer nuclear plant and its utility’s (SCANA) subsequent bankruptcy.  SCANA’s new owner, Dominion, has its own problems.  Dominion is constructing a large offshore wind project that is 25% over budget.  That project, and a somewhat less supportive posture by its regulator, have led to a ‘forced’ divestment of non-core assets and an almost 50% decline in its stock price.  North Carolina’s Duke Energy is in the best shape financially.  However, it is now required by state law to pursue Net Zero de-carbonization by 2050.  Duke’s filed IRP’s foresee record levels of capital spending to accomplish this, over $30 billion for North Carolina alone between now and 2035.  In sum, all area utilities faced significant financial challenges prior to the new AI-related load appearing.

The new EPA proposed emission standards for coal and natural gas power plants are another complicating factor.  Under these rules, utilities will have to adopt expensive carbon capture and/or hydrogen co-firing for fossil fuel plants by 2032-35 or retire the coal plants and operate the gas units below 50% capacity.  None of this would be cheap.  All of it will exacerbate the strains on balance sheets, power prices and tensions between classes of customers.

Area utilities and regulators are just coming to grips with these new challenges, and new solutions are being considered.  This was driven home by a recent conversation with members of the NC Utilities Commission Public Staff, where this question was broached – “Is there any way that the balance sheets of these AI companies could be accessed to fund new facilities?”  It was the right question.  Indeed, it opens a pathway to answers that could address not only the funding issue but the power price and customer fairness issues as well.

Potential solutions can indeed be based on the financial strength of many AI firms.  These customers are giving priority to securing firm, reliable, power.  They certainly have financial heft that can be deployed to secure it.  If one looks at the balance sheets of Microsoft, Meta, or Google, one sees large cash balances, relatively little debt and strong cash flow.  Indeed, these firms often have negative net debt when cash and marketable securities are deducted from outstanding debt.  Consequently, their ability to put financial strength behind the construction of new power generation is unquestionably there.

Moreover, such firms may be willing to do so.  They are discovering that power grids across the country are already stressed and that their demands for large blocks of additional power are not always welcome.  By putting their balance sheets at the disposal of their power providers, the AI firms can secure better supplier response.  The key structure for making this happen is a long-term power purchase agreement (PPA) with prices sufficient to support structured/project financing.  Current market data indicates that these data center firms are sufficiently hungry for committed firm power that long-term PPAs are already being offered to power providers.  Sources indicate that Meta alone is doing PPAs in Georgia for 1 GW.

Such PPAs open the door to move the debt required to finance data center load off the utilities’ balance sheets.  This is well-developed financial territory.  Much of the wind and solar industries was constructed on this basis.  Duke, Southern Company, Dominion, and others could have their de-regulated subsidiaries form partnership structures.  These could bring in private equity partners like Brookfield and Orion, and/or the utilities could employ Master Limited Partnership (MLP) structures to tap public market equity.  In all cases the utility would be the General Partner responsible for building and operating the assets.  However, the partnerships and their debt would be structurally separate from the utilities, off their balance sheets and out of the ratios used by the rating agencies when assigning debt ratings.  The General Partner utilities could provide project completion guarantees should they be required.  These plus the PPAs from such strong power customers would provide the assurances project lenders would need.  Such structures might even merit investment grade credit ratings, enabling the partnerships to tap the project bond market for long term fixed rate debt.

What then of the electricity price and customer class issues?  The structured finance approach offers opportunities to address these as well.  AI data center customers should be charged a ‘new capacity’ price in their PPAs.  Historically industrial customers have enjoyed large discounts justified by the large load that they bring utilities.  Such prices could be justified on the basis that utilities had more than adequate capacity, largely depreciated generation plants and thus welcomed the load.  This however took place when two other conditions prevailed: 1) overall load was growing slowly or hardly at all; and 2) the customers in question were usually manufacturing firms bringing numerous jobs to the local economy.

Neither of these conditions now applies to AI or data centers.  Electrification trends are already driving load growth above historic trends, and AI/data center demand is pushing projections much higher.  AI/data centers also are not large employers.  Regulators will thus be justified in asking: “why should retail customers be ‘subsidizing’ these firms?  Far from making the utilities and power market stronger, they are straining the grid and posing impediments to executing our Net Zero plans.”  Charging the AI firms a ‘new capacity price’ over prevailing industrial customer rates should ease such concerns.  Indeed, the prices should be high enough to not only support the partnership debt load but to pay a hefty equity return to its investors.  Doing so squares fully with the economic principal that the customer of a new dedicated facility should pay prices sufficient to remunerate both the debt and equity capital funding the project.

There is one other obstacle to consider.  Regulated utilities may resist the idea of carving assets out of their rate base.  Indeed, the story utilities are telling Wall Street is that the Energy Transition will replenish their rate bases, turning them into growth stories.  A structured approach to financing AI data centers does not contradict this story; indeed, it should enhance it.  Right now, the Southeast U.S. utilities are looking at all the capital investment they can handle before taking on AI-related load.  Using a structured finance approach for some projects is thus a case of assuring that the utilities’ future budgets are financially manageable.

By placing some projects into structures lodged in their unregulated subsidiaries, utilities may also achieve two other goals.  The structured entities could achieve costs of capital equal to or even lower than the ‘mother-utility.’  Such an outcome would reflect higher leverage in the structure, the strong customer PPAs, and the fact that the parent utility may be facing higher debt costs due to the large base budgets being funded.  Second, the utilities could see a boost to their equity returns.   Lodging some structures in unregulated subsidiaries will free their projects from the regulatory caps on equity returns.

Some utilities may argue that structured approaches lodged in unregulated subsidiaries also bring higher risks.  This perspective must be considered in light of the fact that their own risk profiles will be changing as they execute their massive capital programs.  Moreover, any structured deals should benefit from significant de-risking elements. Demand and price risk will be mitigated by firm PPAs from premier credits.  Completion and operating risk will still reside with the utility as General Partner. With the AI customers paying higher PPA prices and the structure being more leveraged, the utility can capture an attractive result: protection of its stand-alone debt rating plus a boost to its consolidated equity return.

Structure finance thus has a lot to offer the power industry as it confronts unexpected AI load demand.  Moreover, these approaches could provide answers to two other issues: 1) how to employ tax incentives offered by legislation like the Inflation Reduction Act (IRA) and 2) how best to share the risk of First of a Kind (FOAK) generation like small modular nuclear.  Built out supply chains will be needed to bring down costs of next gen-generation.  Right now, the risks associated with FOAK first plants are deterring utilities from deploying these needed technologies. Structured finance’s applications to this issue, and to the related topic of tax incentive best use, will be the topic of our next paper.