The artificial intelligence revolution has a power problem — and it’s getting more expensive by the month.
Across the United States, a staggering number of data center construction projects have been delayed or outright canceled, not because of financing shortfalls or waning demand, but because of a far more fundamental constraint: the electric grid simply cannot deliver enough power fast enough. Industry researchers tracking the sector estimate that more than half of planned U.S. data center projects are now stalled, caught in interconnection queues that stretch years into the future.
For capital markets, the ripple effects are substantial — touching utility bonds, infrastructure financing, data center REITs, and the valuations of every company racing to dominate AI computing.
The Scale of AI’s Electricity Appetite
To understand why the grid is buckling, consider the numbers. A single large-scale AI training cluster — the kind used to develop and run frontier language models — can consume 50 to 100 megawatts of continuous power. A hyperscale data center campus housing multiple such clusters can demand 500 MW or more. For context, 500 MW is roughly equivalent to the power consumption of a mid-sized American city.
Google, Amazon Web Services, and Microsoft collectively announced more than $300 billion in AI infrastructure capital expenditure commitments for 2025 and 2026. Meta Platforms has outlined its own multi-year buildout, and smaller players — from Applied Digital to CoreWeave — are competing for capacity they can lease to AI labs and enterprise clients.
The result is an unprecedented surge in electricity demand at precisely the moment the U.S. grid is grappling with the retirement of aging coal plants, the integration of variable renewable energy, and decades of underinvestment in transmission infrastructure.
Interconnection Queues and the Financing Bottleneck
The most acute bottleneck is grid interconnection. Before a data center can draw power from the grid, it must apply for a connection study, pass environmental review, and often fund costly transmission upgrades. In high-demand regions — Northern Virginia (the world’s largest data center market), Phoenix, Silicon Valley, and the Dallas-Fort Worth Metroplex — interconnection queues now stretch three to seven years.
For project developers and their lenders, this creates serious capital markets friction. Construction financing is typically structured around a project timeline and a revenue-start date. A three-year interconnection delay does not merely postpone income; it can trigger covenant violations in construction loan agreements, force expensive refinancings, or push projects into default if anchor tenants lose patience and walk away.
Real estate investment trusts (REITs) specializing in data centers — including Digital Realty and Equinix — have flagged interconnection delays as a material risk in recent filings. The concern is reflected in how capital markets are pricing new project debt: spreads on data center construction loans have widened as lenders factor in elongated timelines and execution risk.
Transformer Shortages Compound the Problem
Power grid delays are not the only culprit. A persistent shortage of large power transformers — the enormous pieces of equipment that step voltage up or down for transmission — is adding 18 to 36 months to grid upgrade timelines nationwide. U.S. manufacturers cannot meet demand, and import lead times from Korea and Germany have extended sharply.
The shortage traces back to a confluence of factors: surging data center demand layered on top of electric vehicle infrastructure buildout, renewable energy interconnections, and post-pandemic manufacturing disruptions. The result is that even projects with approved interconnection agreements cannot always proceed on schedule.
Who Is Winning the Power Scramble
The capacity crunch is proving to be a windfall for companies that already hold operating data center capacity or that have secured long-term power purchase agreements. Applied Digital, which operates AI-focused data centers on secured power contracts, surged more than 14% in trading on April 14 as investors recognized the competitive advantage of locked-in capacity in an undersupplied market.
Similarly, the Oracle-Bloom Energy fuel cell agreement — a deal to deploy 2.8 gigawatts of on-site fuel cell generation at Oracle data centers — underscores a broader capital markets trend: hyperscalers and co-location providers are increasingly willing to fund off-grid or behind-the-meter power solutions rather than wait years for utility connections.
Small modular nuclear reactors (SMRs) have attracted commitments from Amazon and Microsoft for exactly this reason. By generating power on-site, these technologies sidestep the interconnection queue entirely — at the cost of high upfront capital and long permitting timelines of their own.
Utility Stocks and Infrastructure Bonds
For fixed-income investors, the power crunch is reshaping the credit story for regulated utilities. Companies with large service territories overlapping major data center clusters — including Dominion Energy in Virginia, APS in Arizona, and Oncor in Texas — are presenting regulators with requests for accelerated grid investment programs worth tens of billions of dollars.
These capital expenditure plans, if approved, translate into rate base growth that supports utility bond ratings. Several analysts have upgraded regulated utility bonds in high-demand corridors, citing the predictable revenue stream from data center load growth as a credit positive. The flip side: utilities that fail to secure regulatory approval for their investment programs face pressure from large industrial customers seeking alternative supply arrangements.
Infrastructure debt funds — a growing corner of private credit markets — have meanwhile moved aggressively into data center project financing, offering construction lending that traditional banks have pulled back from. The premium they command reflects genuine execution risk, but also the scarcity of capital willing to navigate complex interconnection timelines.
The Capital Allocation Reckoning
The AI infrastructure boom was always going to collide with physical constraints at some point. The power grid bottleneck is that collision, and it is forcing a recalibration of capital allocation across the sector.
Projects in power-abundant markets — the Midwest, rural Texas, the Pacific Northwest near hydropower — are drawing a disproportionate share of new announcements. The 1.2-gigawatt campus proposed for Irwin County, Georgia, and a 2-gigawatt project floated for New Mexico reflect a geographic diversification away from traditional coastal markets where grid capacity is exhausted.
The companies and investors best positioned in this environment are those that secured power early, built relationships with regulators before demand spiked, and structured project financing with realistic timelines. Those that assumed grid connections would follow demand on a short fuse are now renegotiating — with lenders, with tenants, and with utilities — to keep projects viable.
For the broader capital markets, the message is clear: in the race to build the infrastructure layer of the AI economy, electrons are the new bottleneck. And whoever controls the power wins.
Disclosure: This article was produced with AI assistance and reviewed before publication. It is for informational purposes only and is not investment advice.