Hydropower-sited modular compute for AI: A Canada-first economic–ecological pathway

AI compute is growing fast, but most new capacity is sited far from clean power and useful heat sinks, locking in high cooling overheads and grid stress. We outline a portable approach that colocates modular, encrypted “black-box” AI containers on standardized lease pads at cold-climate hydropower sites. The host delivers metered power, fiber, and a cooling interface; tenants operate within a strict SLA and retain full data privacy. A simple thermal rule—reuse → storage → reject—routes server heat first to nearby homes and public buildings, then to buffer tanks, and only lastly to air-side (or permitted non-contact water) rejection. Seasonal greenhouses act as a flexible sink in warmer months and downshift in deep winter so households take priority. When renewable headroom and idle compute align, a governance-led “Kristals” layer converts surplus cycles into structured, versioned knowledge units with provenance, granting preferential access to contributors while preserving tenant isolation. Co-governance (including FPIC), transparent metering, and phased deployment make the model replicable. The result is a leasing-first blueprint that turns AI capital into community wealth through pad rent, avoided heating fuel, dignified local jobs, and an open knowledge commons.