The Data Center Power Crunch: Is the Market Overestimating AI’s Energy Appetite?

For the past few years, the narrative surrounding the AI revolution has been simple: data centers are power-hungry, and the grid isn't ready. This has turned once-sleepy utility stocks into high-growth plays, with investors piling into the sector to capitalize on the massive infrastructure build-out required to power the next generation of compute.

But according to Andy DeVries, Head of Investment Grade Credit and Utilities at CreditSights, the market might be missing a critical piece of the puzzle: the supply side.

In a recent episode of Odd Lots, DeVries argued that while the demand for AI compute is real, the current utility infrastructure pipeline suggests we are barreling toward a significant oversupply by 2030.

The Math of the "Power Crunch"

To understand the risk, you have to look at the raw numbers. Current data center power consumption sits at approximately 45 gigawatts. Industry forecasts suggest that number will climb to 90–95 gigawatts by 2030.

However, when you look at what utilities are actually building, the numbers tell a different story. Utilities are currently working on connecting roughly 110 gigawatts of near-term supply. Even after adjusting for Power Usage Effectiveness (PUE)—the energy required for cooling and lighting—the capacity currently in the pipeline is already approaching the demand projections for 2035, not just 2030.

"The utilities are working on connecting almost as much as you need by 2035," DeVries noted. "There is a lot of supply of data centers coming, and it’s very unclear if there’s going to be demand for this."

Why the Market Isn't Buying the Hype

If the demand for power is truly as aggressive as the tech giants suggest, you would expect to see that reflected in energy markets. Yet, the forward power curves tell a different story.

In Texas, for example, the market is projecting a massive 30-gigawatt increase in demand by 2030. Despite this, the forward power curve remains relatively flat. Furthermore, natural gas forward curves are currently inverted, suggesting that the market expects lower gas prices by the end of the decade—a strange signal if we were truly facing a permanent, massive surge in power demand.

The "Circular" Financing Risk

Beyond the potential for physical oversupply, investors should be wary of how these projects are being financed. We are seeing a rise in "circular" financing, where tech giants (like NVIDIA or OpenAI) invest in the equity of data center providers to secure compute capacity.

DeVries compares this to the vendor financing risks seen during the dot-com era with companies like Nortel. For fixed-income investors, the concern is twofold:

1. Off-Balance-Sheet Risk: Tech companies are increasingly moving data center debt off their balance sheets. While rating agencies may impute these liabilities, the structures often include "walk-away" clauses that could shift the financial burden to rate-payers or bondholders if the demand for these data centers fails to materialize.
2. Covenant Weakness: As competition for these deals heats up, documentation protections are beginning to erode. Investors should scrutinize credit docs for explicit guarantees that protect rate-payers from being left with "stranded assets" if a data center shuts down.

Investment Implications: Proceed with Caution

The "AI-powered utility" trade has been a massive winner, but the contrarian view suggests that the easy money may have already been made.

• For Equity Investors: Be cautious of utility earnings growth projections. If the projected demand does not materialize, utilities may face regulatory pushback on rate hikes, potentially leading to a valuation reset.
• For Fixed Income Investors: Scrutinize the credit documentation of off-balance-sheet vehicles. If you are buying debt, ensure there are ironclad guarantees that prevent the risk from being offloaded onto the public.
• The "Short" Opportunity: DeVries suggests that there may be tactical opportunities in power markets by being short power in the near term—due to the influx of solar supply—while remaining long on the back end of the decade.

Key Takeaways

• Oversupply Risk: Utilities are building capacity (110GW) that may outpace the actual demand growth (50GW) projected by 2030.
• Market Dissonance: Forward power and gas curves do not currently reflect the aggressive demand growth forecasts being touted by AI bulls.
• Credit Vigilance: Watch for "circular" financing deals and off-balance-sheet structures that lack clear protections for bondholders and rate-payers.
• Efficiency Gains: The rapid pace of AI efficiency gains could mean that compute power requirements grow slower than the current, linear projections assume.

As with any transformative technology, the long-term potential of AI is immense, but the infrastructure build-out is subject to the same boom-and-bust cycles that have defined every major industrial shift in history. Investors should look past the headlines and focus on the simple, boring math of supply and demand.