The headline numbers from the AI infrastructure buildout are staggering. Hundreds of billions in announced capital. Gigawatt-scale campuses breaking ground. A global race to deploy compute infrastructure faster than competitors.
The reality, tracked carefully by Sightline Climate across 140 US projects, is considerably more sobering.
WHAT THE DATA ACTUALLY SHOWS
Under construction: 5 GW (42%)
Announced, no physical progress: 7 GW (58%)
No power strategy disclosed: 3 GW (25% of total)
Source: Sightline Climate, 2025
The construction timelines tell the rest of the story. A data center takes 12 to 18 months to build once ground is broken and power infrastructure is in place. For a 2026 commissioning date, physical construction should have started in late 2024 at the latest. Projects with no physical progress as of mid-2025 are not going to hit 2026 targets.
THE 25% WITH NO POWER STRATEGY
The most striking figure in the Sightline analysis is that 25% of announced projects have not disclosed any power sourcing strategy. These are not projects in early planning — they are projects with announced commissioning dates and, in many cases, announced capital commitments.
A data center without a power strategy is not a data center. It is a press release. The capital behind these announcements is real. The compute is real. The power infrastructure does not yet exist.
WHY THE BOTTLENECK IS STRUCTURAL
The power equipment shortage driving this stall is not a temporary supply chain disruption of the kind seen during COVID. It is a structural imbalance between demand and manufacturing capacity that will take years to resolve.
- Transformer factories are running at or beyond capacity worldwide
- New factory construction takes 3–5 years and requires specialised equipment and labour
- Utility interconnection queues in the US have grown from months to years in AI hotspots
- Switchgear and cable face similar constraints driven by the same demand surge
GEOGRAPHY MATTERS
The stall is not uniform across US markets. Some regions — particularly those served by utilities with proactive grid investment and available substation capacity — are processing connections faster. Texas ERCOT, certain PJM markets and parts of the Pacific Northwest have more available capacity than the headline figures suggest.
Conversely, Northern Virginia — the world's largest data center market — is effectively full for new large-scale power connections. New projects in Loudoun County face interconnection timelines of 5–7 years from application to energisation.
THE IMPLICATION FOR EUROPEAN MARKETS
The stall in US data center deployment is driving serious interest in European alternatives. France, Ireland, the Netherlands and the Nordic countries all offer something the US markets currently cannot: available grid capacity on shorter timelines.
France is particularly interesting because its nuclear-heavy grid offers price stability and low carbon intensity alongside available industrial land in regions with legacy HV infrastructure. Several US hyperscalers are actively evaluating French locations that would have been considered secondary choices in a less constrained environment.
WHAT HAPPENS TO THE STALLED PROJECTS
The 7 GW of stalled US projects will sort into three categories over the next 24 months:
- Projects that secure power — through equipment orders finally delivering, or by acquiring sites with existing capacity
- Projects that relocate — to markets with available grid capacity, including international markets
- Projects that are cancelled — the capital is redeployed to viable alternatives
For investors and developers following this space, the ability to identify which projects fall into which category — and where the viable alternatives are located — is a significant competitive advantage.