01
The gap
AI inference needs to run close to users, in builds from 1 to 50 MW, in metros where power is available. Hyperscale operators build 100 MW campuses in a few places. Small, power-ready inference halls are not being built systematically where the demand sits.
02
The pipeline
The pipeline has two sources. The first is construction-ready sites with commercial power that we identify and screen ourselves: several hundred so far, with 7 GW of power identified across them. The second is the retrofit of existing telecom central offices, where the building, the power and the fiber are already in place. We are in discussions with a top 10 telecom provider on a partnership to retrofit several thousand of their sites. Details are shared under NDA.
03
The model
We build to suit for one tenant at a time, where that tenant needs capacity. Tenants who need high availability take a mirrored second site 20 to 50 miles away.