AI Data Center Site & Power Needs

---

Artificial intelligence is accelerating the buildout of specialized data center infrastructure, and power availability has become the main constraint on where new facilities can be built. Recent forecasts show U.S. data center power demand rising from 31 GW in 2025 to 41 GW in 2026 and 66 GW in 2027, more than doubling in two years. As a result, site selection is shifting toward locations with faster access to power, stronger grid capacity, and greater flexibility to add onsite generation. 

AI data centers are no longer defined only by compute needs; they are increasingly shaped by electricity, land, and interconnection timelines. Developers are now prioritizing regions that can deliver power quickly, support large campuses, and reduce permitting delays. The result is a more selective market in which power-rich geographies are gaining share while legacy markets face tighter constraints. 

Successful AI data center deployment depends on a narrower but more demanding set of site-selection criteria than traditional cloud infrastructure. The most important factors are power availability, speed to power, land for expansion, fiber connectivity, and resilience against climate and regulatory risk. In practice, the best sites are now those that can support large-scale AI loads without forcing operators to wait years for utility delivery.

The biggest challenge for AI expansion is securing enough reliable electricity. Goldman Sachs Research now expects U.S. data center power demand to climb from 31 GW in 2025 to 41 GW in 2026 and 66 GW in 2027, while data centers’ share of total U.S. peak summer power demand rises from 4.1% to 8.5% over the same period. Only about 50% to 60% of the capacity scheduled for the next one to two years is expected to come online on time, reflecting delays, cancellations, supply chain issues, and labor constraints. 

High-speed connectivity remains essential, especially for AI workloads that move large amounts of data across training, inference, and storage systems. Sites need strong fiber access, multiple carriers, and route diversity to reduce latency and limit single-point failure risk. Connectivity is still a core requirement, but it is now secondary to power in many site-selection decisions. 

The strongest U.S. markets are increasingly those with available land, natural gas access, faster permitting, and room for utility expansion. Bloomberg Energy’s 2026 report argues that Texas and the Southeast are gaining share because they can offer land and power more readily than legacy hubs. Goldman Sachs also notes that reliability risk is rising in the Mid-Atlantic, Mid-Continent, and Northwest, while Texas and Georgia face less severe tightening because of planned generation additions. 

The major established markets still matter, but the center of gravity is shifting. Texas and Georgia are positioned to absorb more growth, while markets such as Virginia, California, Oregon, Iowa, and Nebraska are expected to lose share by 2028. That shift reflects a simple reality: AI development is following power, not just proximity to end users.

Building AI data centers remains expensive, and costs are rising as power density increases. Higher-capacity sites require more robust electrical systems, larger land footprints, and more complex cooling and backup power designs. The economics increasingly favor developers that can secure power early, because delays can quickly erode returns and market timing.

Onsite power is becoming a permanent part of data center planning rather than just a backup option. Bloom Energy’s 2026 survey found that many operators expect to rely more heavily on onsite generation, and it reports that one-third of hyperscalers and colocation providers expect to run data centers entirely on onsite power by 2030. The same report says data centers are moving toward high-voltage busways and direct current architectures to support higher power density and faster deployment.