For years, large infrastructure projects treated power capacity as a simple quantity question: does a site have access to 100 MW, 500 MW, or a gigawatt. AI infrastructure is changing that calculus. A gigawatt available in three years is of little use to a campus whose compute, customers, and capital are ready now. The metric that matters most has shifted from planned megawatts to megawatts that can actually be delivered on schedule.
Power Planning Moves Earlier
That shift is pulling power planning much earlier into the development process. Lawrence Berkeley National Laboratory has identified more than 40 approaches to speeding up large-load grid connections, spanning load forecasting, interconnection, resource procurement, market operations, and cost allocation. The sheer number of approaches reflects how many separate steps, from grid studies to transmission upgrades to equipment procurement and permitting, can each delay when power is actually usable.
Sites Are Moving to Chase Power
Developers are already adjusting where they build. Reuters reported that European AI data centers planned to come online between 2026 and 2028 sit an average of 175 kilometers from major cities, compared with 46 kilometers for projects delivered between 2022 and 2025. JLL attributed the shift to cheaper land, available energy, and faster grid connections, a sign that power availability is increasingly dictating site selection rather than proximity to population centers.
New Procurement Models Emerge
Procurement approaches are evolving as well. Within PJM Interconnection, more than 50 GW of generation projects already hold grid connection agreements, yet permitting, equipment, and workforce constraints continue to slow development. PJM is now facilitating bilateral agreements between generators and large loads outside the traditional procurement process, while the Federal Energy Regulatory Commission has directed grid operators to expand options like co-location, behind-the-meter generation, and more flexible transmission service. Each option shortens some delays but shifts more of the execution burden onto the developer, whether that means securing turbines and fuel, finding the right co-location site, or betting on a generator being built on time.
Equipment Supply Chains Set the Pace
SpaceX offers a stark example of how a single bottleneck can define a project timeline. Elon Musk said the company will begin casting gas-turbine blades and vanes in-house because turbine generator demand is straining suppliers. Specialized turbine blades reportedly take 60 to 90 weeks to manufacture, and Musk said in-house production could cut delivery time by as much as 18 months. The takeaway for data center developers is that speed-to-power depends on the full supply chain behind every megawatt, not just interconnection queue position.
A Phased, Flexible Approach
A more resilient power strategy spreads risk across utility service, bilateral contracts, onsite generation, storage, and other distributed resources, each on its own timeline. That allows developers to bring an initial power block online while larger grid or generation projects proceed in parallel, pairing modular power and cooling systems with phased compute buildout rather than waiting for a complete campus energy system on day one.
Filtering Real Demand from Speculation
Credible demand forecasting matters too. Reuters reported that large-load requests, driven mostly by data centers, have outpaced actual US data center demand in several major power markets. States including Texas are now requiring deposits and development milestones to separate financed, shovel-ready projects from speculative capacity reservations that make it harder for utilities to plan generation and transmission efficiently.
The strongest AI infrastructure projects combine speed with certainty: developers who control land, financing, and permits, understand their load ramp, and can show utilities and generation partners a realistic, staged plan for bringing power online.