Data center power capacity has moved from a facilities line item to a board-level infrastructure risk. Grid interconnection queues that stretch five years, transformer lead times of up to four years, and a projected near-tripling of US data center electricity demand by 2030 mean that CIOs planning AI compute expansions - on-premises or in colocation - face hard constraints that no amount of capital spending can fix on a short timeline.
451 Research, Datacenter Services and Infrastructure Market Monitor and Forecast, September 2025; reported by S&P Global and Data Center Dynamics.
| Value (GW) | Grid Power Demand (GW) |
|---|---|
| 2024 | 50.5 GW |
| 2025 | 61.8 GW |
| 2026 | 75.8 GW |
| 2028 | 108 GW |
| 2030 | 134.4 GW |
The demand surge behind the power wall
US data centers consumed roughly 50.5 GW of grid power in 2024. By 2025 that figure is forecast to reach 61.8 GW - a 22% year-on-year jump - and 451 Research (S&P Global) projects demand will climb to 134.4 GW by 2030, nearly triple current levels. That forecast, published in the Datacenter Services and Infrastructure Market Monitor in September 2025, covers colocation and hyperscale facilities and does not fully account for enterprise-owned infrastructure outside the major cloud providers.
The driver is not incremental server refresh. It is AI. Training and inference workloads are power-dense by design: a modern GPU rack can draw 60 to 100 kW, compared with 10 to 20 kW for a standard compute rack. As enterprises move from piloting AI to running it at scale, aggregate facility power requirements are rising faster than most site plans anticipated. Gartner placed energy-efficient computing on its six-trend I&O watchlist for 2026 at its Infrastructure and Operations Conference in December 2025, explicitly linking the pressure to AI adoption and energy constraints.
Grid interconnection wait times now average five years. A CIO who identifies a data center site today cannot realistically expect utility power above 20 MW inside that window without an existing queued position.
Why the grid cannot keep up
The supply side of this equation is structurally slow. Research from Lawrence Berkeley National Laboratory shows that grid interconnection wait times have more than doubled over the past 15 years. New projects now spend an average of five years in the queue before reaching commercial operation. A CIO who identifies a data center site today and needs utility power above 20 MW cannot realistically expect grid connection inside that window without an existing queued position or a pre-negotiated utility agreement.
High-voltage transformers are the hard stop inside that queue. Lead times have stretched to two to four years, according to infrastructure advisory firm EnkiAI (February 2026) and Hanwha Data Centers (February 2026). Transformers are not a commodity item - they are custom-engineered to site specifications, and global manufacturing capacity has not scaled to match demand. A shortage at this single component level can delay an entire facility buildout regardless of how quickly civil construction proceeds.
The regulatory environment is also shifting around these constraints. The US Department of Energy proposed federal jurisdiction over new grid loads above 20 MW in October 2025, a signal that large data center power draws are entering the oversight tier previously reserved for generation plants. State legislatures introduced 238 data center bills in 2025, with roughly half focused on energy concerns, according to government relations firm MultiState (Canary Media, December 2025). CIOs with large footprints should expect regulatory engagement to become part of infrastructure planning, not just permitting.
What Deloitte's executive survey reveals
In April 2025, Deloitte surveyed 120 US power company and data center executives and found that grid stress ranked as the single leading challenge for data center infrastructure development. That result is notable because it combines the perspective of utility operators - who see demand requests they cannot fulfill quickly - with data center operators who are experiencing the delays firsthand. Both sides of the transaction are flagging the same bottleneck.
For enterprise IT leaders, the practical implication is that colocation providers and build-to-suit developers are themselves constrained. Signing a colocation contract does not transfer the power risk entirely. If a provider's own interconnection position is weak or their transformer delivery is delayed, committed capacity may slip. CIOs need to add power-readiness questions to the RFP and due-diligence process for any colocation expansion.
A power-aware infrastructure strategy for IT leaders
The organizations best positioned for this environment are those treating power capacity as a site-selection criterion rather than a post-selection problem. That means evaluating utility interconnection queue position, existing substation capacity, and on-site generation options - including diesel backup and, increasingly, gas turbines or fuel cells - before committing to a facility. It also means engaging with colocation partners on their specific grid agreements and transformer delivery schedules.
Load flexibility is a related lever. Demand response agreements with utilities can improve grid access in constrained markets; some utilities will fast-track interconnection for loads that agree to curtailment during peak periods. For AI workloads, batch inference and non-time-sensitive training jobs are candidates for flexible scheduling. CIOs should work with infrastructure architects to identify which workloads can tolerate controlled power variance and build that flexibility into capacity contracts.
Cooling efficiency compounds the power equation. Liquid cooling and direct-to-chip systems reduce the power usage effectiveness ratio and effectively extend usable capacity from a fixed power envelope. Organizations that have not yet evaluated liquid cooling for GPU infrastructure should treat that assessment as urgent: every percentage point of PUE improvement represents real megawatts freed for compute. Gartner's 2026 I&O imperative makes clear that boards will begin asking about energy consumption in the same breath as AI capability.
Energy-efficient infrastructure planning is no longer optional for CIOs with AI ambitions. Power procurement belongs on the critical path alongside compute procurement, and the planning horizon is measured in years, not quarters.
