# How it works > Canonical HTML: https://initkoa.org/infrastructures/kristal-farms/how-it-works > Markdown mirror: https://initkoa.org/infrastructures/kristal-farms/how-it-works/index.html.md > Route: /infrastructures/kristal-farms/how-it-works > Source: app/infrastructures/kristal-farms/how-it-works/page.mdx > Generated: 2026-06-13T02:25:42.872Z [Open the HTML page](https://initkoa.org/infrastructures/kristal-farms/how-it-works) "A plain-language walkthrough of how Kristal Farms convert clean power into useful compute and useful heat, with a tenancy model designed for privacy, safety, and reversibility.", # How it works Kristal Farms is a **modular compute site designed around heat reuse**. Instead of treating server heat as waste, the infrastructure is built so that **compute and heat are co-products**: - tenants run compute workloads, - the site captures and reuses waste heat for **district heating** and **greenhouse production**, - and the whole system is operated with **clear, governable priorities**. ## The system in one picture This is not “a data center near a dam.” It is **compute modules near heat users**, with a network link to the outside world. ## The three loops ### 1) Power loop (clean electricity → predictable capacity) **What it does:** provides stable electricity to modular pads. - A local substation distributes power to multiple pads. - Pads can be added or removed without redesigning the entire site. - Operations can sequence pad start/stop to protect stability. - Power & grid → ### 2) Thermal loop (waste heat → useful heat) **What it does:** turns server heat into community value. - Pads run servers using liquid cooling. - Heat is transferred through **heat exchangers** into a separate “community heat loop” (no mixing). - Heat is delivered to: - public buildings (phase 1), - greenhouses sized to absorb seasonal surplus (phase 2+). - Thermal storage smooths mismatches between “compute heat supply” and “heat demand.” **Heat-first rule:** the operating logic prioritizes **useful heat delivery** over maximum compute throughput when there is a conflict. - Heat-first design → - Cooling & water → ### 3) Network loop (local pads → global compute) **What it does:** exports compute safely and reliably. - Each pad has redundant fiber connectivity to the site network. - The site uplinks to a regional hub for external traffic. - The network is operated as critical infrastructure: uptime, monitoring, and incident response. - Fiber & network → ## The tenancy model (privacy by architecture) Kristal Farms is designed for “black-box” tenancy: - **The host provides:** the pad, power handoff, cooling/thermal handoff, and fiber handoff. - **The tenant controls:** everything inside the module (hardware, software, keys, workloads). - **The host monitors:** physical and infrastructure metrics (power draw, temperatures, flow, alarms), not tenant data. This is a deliberate boundary: it keeps the infrastructure useful without becoming a surveillance platform. - Tenancy model → ## Governance: who decides what A system that blends compute + heat + environment needs explicit governance. Kristal Farms is designed so decisions about: - heat priorities, - environmental limits, - capacity allocation, - and community benefits are made through named committees and a defined operating charter—not ad hoc by operators. - Governance → - Metrics & dashboard → ## Resilience, safety, and reversibility Kristal Farms is treated as infrastructure that must be: - **safe under failure** (degrade gracefully), - **transparent in operation** (measured, observable), - **reversible** (modules removable; site restoration planned). - Environment & safety → - Reversibility → ## Phasing: how it scales without breaking The project is intended to scale by adding modules and extending heat distribution: 1. **Phase 1:** prove the loop (pads + heat delivery + basic governance + monitoring) 2. **Phase 2:** expand heat network and greenhouse capacity; add redundancy 3. **Phase 3:** replicate the pattern as a stable operating model - Phasing → - Go / No-Go gates → ## Why this structure matters Most compute infrastructure optimizes for compute alone. Kristal Farms optimizes for **civic outcomes**: - heat becomes a local public good, - compute becomes a tenant product, - and governance makes tradeoffs explicit. - Why this exists →