Why AI capacity is moving to secondary markets

For a generation, the data center map clustered around a handful of primary corridors. AI demand is now pulling capacity outward, not for novelty, but because the primary corridors have run into the one wall money cannot quickly move: available power.

The primary-market wall

The established corridors are increasingly defined by long interconnection queues, constrained transmission, and rising local resistance to new large loads. Even with capital and chips in hand, a buyer can find that the binding constraint is a grid that simply cannot add their load on a useful timeline.

What makes a secondary market serious

"Secondary" does not mean second-rate. A secondary market becomes a credible AI site when it offers:

• Real, available power. Capacity that can be contracted and delivered, not just theoretically present.
• Competitive industrial economics. A power cost structure that holds up across a multi-year, power-dominated workload.
• Carrier-diverse fiber. High-bandwidth, low-latency paths to major cloud regions already in place.
• Speed to power. The ability to deploy without joining a multi-year queue.

The TVA footprint across the Southeast is a clear example of all four appearing together, which is why it has moved from overlooked to actively sought for compute.

Demand is structural, not a spike

The broader backdrop is not a temporary surge. AI data center demand is projected to grow for the better part of a decade, and power has been identified as the binding constraint on that growth. When demand is structural and the primary markets are constrained, capacity flows to where power is, and secondary markets stop being secondary.

The map is not being redrawn by preference. It is being redrawn by megawatts.