Operate the Data Center

This layer captures a new category of AI infrastructure company: the powered campus developer. These are operators who acquire land, secure power rights, build or lease facilities, and deliver ready-to-use infrastructure to hyperscalers and AI companies. They sit between the construction layers (L16-L19) and the end customers (L26), solving the most acute problem in the AI buildout: the 5-10 year timeline to get new powered capacity online.

Nucleus Energy (NUAI) exemplifies the model. They develop NNN-leased powered sites — delivering land plus power to hyperscalers under long-term net leases where the tenant pays all operating costs. The business model monetizes the scarcity of grid-connected, permitted land rather than the data center building itself.

The most structurally interesting companies in this layer are the former Bitcoin miners pivoting to AI infrastructure. IREN, Hut 8, WULF (TeraWulf), and CORZ (Core Scientific) built GPU-dense facilities with behind-the-meter power for Bitcoin mining. When AI inference demand exploded, they discovered their stranded power assets — often in rural locations with cheap electricity — were exactly what hyperscalers needed. Core Scientific's pivot is the most dramatic: they emerged from bankruptcy in 2024 and signed a 200MW deal with CoreWeave, transforming from a distressed mining company into an AI infrastructure landlord overnight.

Applied Digital (APLD) occupies a hybrid position — building powered campuses specifically designed for AI compute, with pre-secured power and cooling infrastructure. Their model is purpose-built from inception for AI workloads, unlike the mining pivots which are retrofitting existing facilities.

The structural insight: these operators are arbitraging the power scarcity premium. The hyperscalers need powered capacity faster than they can build it themselves. Operators who already control power — through mining, utility relationships, or early land acquisition — can charge infrastructure premiums that reflect the true scarcity of grid-connected AI-ready capacity.