Not every company building AI infrastructure needs to compete for GW- scale campuses. There’s an increasingly attractive opportunity further down the market that gets overlooked @DarkHorseDRK CEO Joel Block made some interesting points at the Maxim Growth Summit today about the segment of AI infrastructure DarkHorse is positioning itself to serve. The company’s strategy centers on smaller, roughly 10–50 MW facilities, particularly for distributed AI inference workloads, using existing power infrastructure across the TVA. One of the biggest advantages ex-BTC miners have transitioning toward AI infrastructure isn’t necessarily the buildings or equipment they already own. Much of that hardware, cooling, networking, and electrical infrastructure must be replaced or substantially upgraded to accommodate high-density GPU workloads. The real advantage is access to power. Existing utility connections, energized sites, substations, and established relationships with local utilities can potentially remove years of development uncertainty compared with starting from scratch. DarkHorse currently has 50 MW of operating power capacity across 4 locations in Tennessee and Kentucky. They recently secured a conditional agreement for another 50 MW in Hopkinsville, bringing its contracted regional power footprint to 100 MW. Converting those MWs into revenue-generating compute infrastructure is the next challenge. But what’s interesting is while much of the market focuses on hyperscalers securing hundreds of MWs for large AI campuses, there’s a separate opportunity emerging around smaller MW deployments serving inference and other distributed compute requirements. DarkHorse intends to use prefab, modular data center designs that can be manufactured offsite and deployed in stages, potentially reducing construction complexity and improving deployment timelines. Constrained power availability is increasingly bringing larger operators and hyperscalers into conversations about smaller sites. Emerging neoclouds may have strong equity backing but limited operating histories or IG-grade credit profiles. Larger, creditworthy counterparties can materially change how lenders evaluate the same underlying infrastructure. My takeaway is that AI infrastructure won’t be built exclusively through massive centralized campuses. There is room for a more distributed network of smaller facilities where power availability, deployment speed, and workload requirements align. DarkHorse is attempting to position itself in that segment rather than compete head-on with the largest developers.