For most of the last decade, the power industry moved at a predictable pace — steady load growth, long infrastructure cycles, capital plans stretched across 20-year horizons. That world is gone. While the rest of the economy fixated on GPU prices and server specs, the buildings housing those servers started pulling electricity at a scale that is forcing the entire power sector to rethink its assumptions from scratch.
This is not a story about tech giants building bigger campuses. It is a story about what happens when a single industry grows fast enough to bend the trajectory of the entire electrical grid.
The Numbers That Changed Everything
In 2022, global data centers consumed roughly 460 terawatt-hours (TWh) of electricity. Significant, but manageable. Utilities had it in their forecasts.
Then AI arrived at scale.
By 2025, data center electricity demand surged 17% in a single year, according to the International Energy Agency — a rate that dwarfed overall global electricity demand growth of 3%. The IEA now projects consumption will exceed 1,000 TWh by 2026 and roughly double by 2030. Gartner estimates AI-optimized servers alone will account for nearly half that load by decade's end.
What makes these numbers genuinely disruptive is not just their size — it is their concentration. Data centers cluster in specific places: Northern Virginia, central Texas, the Pacific Northwest, the outskirts of Dublin. In the PJM transmission region around Virginia, data centers account for over 90% of projected new power demand. Ireland hit 22% of its entire national electricity consumption from data centers in 2024. You cannot add that much load, that fast, to a grid built for a completely different demand profile without hitting serious friction. That friction has arrived.
What It Looks Like on the Ground
Utilities are revising their long-range load forecasts upward for the third consecutive year. American Electric Power (AEP), one of the largest U.S. transmission companies, has customer commitments for 24 gigawatts of new demand by 2030, with 18 GW coming from data centers alone. That volume would be five times AEP's current system size. These are not projections — these are binding service agreements on the books today.
Grid interconnection queues — the regulated process by which new loads get approved to connect to the transmission network — have become a hard bottleneck. Power constraints are extending data center construction timelines by 24 to 72 months in the most affected U.S. markets. Of the 12 GW of U.S. data center capacity announced for 2026, only 5 GW was actually under construction, with most of the rest stalled over power access.
The response has been striking. By early 2026, roughly one-third of planned new U.S. data center capacity was designed to operate wholly or partly off the public grid — a figure that was effectively zero in early 2025. On-site gas turbines, battery storage, and nuclear power purchase agreements are no longer edge cases. They are becoming standard practice.
The Transformer Bottleneck Nobody Saw Coming
Of all the knock-on effects, the most underreported is the strain on transformer supply chains. A transformer is the device that steps voltage up or down at various points in the electrical system — at generation plants, substations, and where transmission power connects into a facility. Without transformers, nothing else in this conversation works.
Before 2020, lead times ran 24 to 30 months. Today, many markets report timelines of two to four years, with some high-voltage units pushing five. Wood Mackenzie modeled a 30% supply shortfall for power transformers across the U.S. fleet in 2025. Demand for generator step-up units has grown 274% since 2019, and substation transformers are up 116%. Over half of U.S. distribution transformers — roughly 40 million units — are already past their expected service life, adding replacement pressure on top of new-build demand.
Data centers are a primary driver of this crunch. Facilities running dense GPU clusters operate near full electrical load continuously, requiring specialized transformer configurations that most of the manufacturing sector was not built to produce at this volume. Procurement teams at hyperscalers and colocation providers — companies that lease out data center space — are now placing capacity reservations with a voltage transformer manufacturer 18 to 24 months before a project breaks ground. Electrical equipment, which represents under 10% of total data center cost, has become 100% of the bottleneck.
Who Needs to Be Paying Attention
This shift reaches further than most people assume. The affected parties include:
- Power utilities and grid operators, recalibrating planning cycles and rate structures in real time as load growth projections keep climbing
- Electrical infrastructure manufacturers — transformer makers, switchgear suppliers, substation builders — navigating a demand surge with limited near-term capacity to expand
- Data center developers and real estate firms, for whom power availability now ranks above land cost and location in site selection
- Enterprise technology buyers relying on colocation capacity, who are absorbing cost increases and availability constraints flowing directly from the power crunch
The common thread is that decisions made in the next 12 to 24 months — about procurement, siting, grid engagement, and equipment sourcing — will determine project outcomes for years after.
What to Do About It
The organizations navigating this environment well share a few consistent behaviors. First, they treat power procurement as the opening move, not the closing one. Securing a grid connection or locking in on-site generation is now the milestone everything else schedules around. Second, they engage transformer suppliers and utilities far earlier than traditional project timelines would suggest — ideally before a project is fully designed. Third, they build redundancy into their power strategy deliberately, combining grid access with on-site generation and maintaining relationships across multiple equipment suppliers in different geographies.
The data center boom is not a temporary disruption to work around. It is the new shape of the energy landscape — and the organizations that internalize that early will be in a meaningfully better position than those still waiting for conditions to normalize.
