Deutsch
English
Español
Français
Nederlands
Português
The exponential growth of artificial intelligence data centers is creating a hidden systemic risk in power market finance that regulators and investors are only beginning to price. With global AI data center electricity consumption projected to reach 1,000 terawatt-hours (TWh) by 2026, utility companies are signing long-term power purchase agreements (PPAs) for intermittent renewables while simultaneously contracting for 24/7 nuclear baseload, exposing balance sheets to stranded-asset risk if AI demand growth falters or regulatory frameworks shift.
The Scale of the Risk
Data center electricity consumption is forecast to increase from 447 TWh in 2025 to 565 TWh in 2026, according to Gartner, with some estimates reaching 1,000 TWh when including broader AI infrastructure. The Big Five hyperscalers—Amazon, Microsoft, Google, Meta, and Oracle—will spend over $600 billion on infrastructure in 2026, a 36% increase from 2025, with approximately 75% targeting AI. To finance this buildout, technology companies issued a record $428 billion in bonds in 2025, with projections suggesting up to $1.5 trillion in cumulative debt by 2028.
Meanwhile, America's investor-owned utilities have unveiled a staggering $1.4 trillion capital spending plan through 2030, driven primarily by AI data center power demands—a 27% surge from last year's $1.1 trillion projection. This utility capital expenditure boom is financed through debt, inflating leverage ratios and attracting scrutiny from credit rating agencies.
The Two-Way Commitment Trap
Utilities are caught in a structural dilemma. To meet AI data centers' demand for 24/7 carbon-free energy, they are signing PPAs with renewable projects that generate intermittently, while also contracting for nuclear baseload power. Microsoft's landmark 20-year, 837 MW PPA with Constellation Energy to restart Three Mile Island Unit 1 exemplifies this trend, with the tech giant spending $1.6 billion to refurbish the plant. Amazon has expanded its nuclear offtake with Talen Energy to 1,920 MW through 2042, and Meta has committed to 6 GW of new nuclear power.
However, these long-term contracts create a two-way commitment that exposes utility balance sheets. If AI demand growth slows—due to efficiency gains, model optimization, or a macroeconomic downturn—utilities could be left with expensive PPAs for power they no longer need. The stranded asset risk in energy infrastructure is particularly acute for nuclear plants, which have high fixed costs and cannot easily ramp down.
Credit Rating Implications
Moody's and Fitch have begun scrutinizing utility leverage levels, noting that data center-driven capital expenditure is increasing debt loads. Morningstar DBRS recently took credit rating actions on eight data center transactions. The BIS Annual Economic Report 2026 warns that AI-driven investment is increasingly reliant on debt and complex funding structures, with private credit loans to AI companies surging from $3 billion in 2010 to over $40 billion in 2025.
Market Distortions and Price Spikes
The impact on power markets is already visible. PJM Interconnection's capacity auction prices surged from $28.92/MW-day in 2024-2025 to $329.17/MW-day in 2026-2027—a nearly tenfold increase. Data centers drove 63% of the increase, adding $9.3 billion in capacity costs that all ratepayers must absorb. Cumulative costs through 2033 could reach $100-$163 billion, with the average family facing a ~$70/month increase by 2028.
Goldman Sachs warns that data center demand could add 0.1% to core inflation. Electricity costs have risen 42% since 2019, with utilities requesting $31 billion in rate hikes in 2025 alone. The PJM capacity market price crisis highlights how concentrated demand from a single sector can distort wholesale power markets.
Financial Stability Concerns
The Bank for International Settlements (BIS) flagged four key pressure points in its June 2026 Annual Economic Report: renewed inflation, uncertainty over the durability of AI-related investment surges, financial vulnerabilities from elevated asset valuations, and record-high public debt. The BIS specifically warned that AI-driven investment could lead to boom-and-bust cycles, with circular financing—interlocking equity, debt, and supplier contracts where assets may be pledged multiple times—mirroring 2008-style rehypothecation risks.
If AI returns disappoint, leveraged hedge funds—now dominant sovereign bond buyers—face fire-sale pressure, creating a feedback loop from tech bust to sovereign debt crisis. The BIS warning on AI circular financing underscores how power market debt could propagate through institutional portfolios.
Regulatory and Political Responses
States have introduced over 238 data center-related bills. Oregon created the first dedicated data center rate class, Virginia's SB 253 shifts costs from households to data centers, and at least six states have introduced construction moratoriums while seven have restricted tax incentives. Ohio requires minimum billing, and West Virginia enacted certified microgrid districts to bypass interconnection queues entirely.
Grid interconnection bottlenecks remain the primary constraint: half of the 16 GW of new data center capacity targeted for 2026 faces delays pushing completion to 2027 or later. Power transformer lead times average 128 weeks, meaning hardware alone takes nearly three years to procure.
Expert Perspectives
"The convergence of AI infrastructure buildout with energy transition financing creates a novel intersection of technology, energy, and financial stability risk that markets are only beginning to price," said James O'Connor, a financial analyst covering energy infrastructure. "Utilities are making billion-dollar commitments based on demand forecasts that have never been tested in a downturn."
The BIS report echoes this caution: "More frequent supply disruptions could entrench higher inflation expectations, while AI-driven investment is increasingly reliant on debt and complex funding structures."
FAQ
What is the AI power-load risk?
The AI power-load risk refers to the financial stability threat posed by utility companies signing long-term power purchase agreements for both intermittent renewables and nuclear baseload to meet AI data center demand, creating stranded-asset risk if AI growth falters.
How much electricity will AI data centers consume in 2026?
Projections range from 565 TWh (Gartner) to 1,000 TWh (broader estimates), up from 447 TWh in 2025.
Why are utilities at risk?
Utilities are committing to $1.4 trillion in capital spending through 2030, financed by debt, while signing long-term PPAs that could become uneconomic if AI demand growth slows or regulatory frameworks shift.
What are regulators doing?
The BIS, IMF, and central banks are flagging risks. States have introduced over 238 data center-related bills, including rate class changes, cost-shifting measures, and construction moratoriums.
Could this cause a financial crisis?
The BIS warns that circular financing and leveraged positions in AI-related debt could create a feedback loop from tech bust to sovereign debt crisis, though the probability remains debated.
Conclusion
The AI power-load risk represents a novel intersection of technology, energy, and financial stability. With central banks and regulators increasingly focused on the potential for a synchronized correction in power-market debt, market participants must carefully assess the two-way commitment trap facing utilities. The coming months will test whether the industry can manage this convergence without triggering cascading defaults in deregulated power markets.
Follow Discussion