Reading the Facility as an Intelligence Surface
Faraday's read: the whole course reduces to one question you can ask about any data center — how well does whoever runs it read the second machine? The compute story is public and crowded. The facility story is measured physics that most models never open: power margin, cooling headroom, redundancy that is real or nominal, plant drifting toward failure, and how far forward the operator can see. Track it separately and you see constraints and value the compute surface structurally cannot show. That gap is the edge, and it closes slowly.
Knowledge check
Why should both operators and investors track the facility layer separately from the compute workload?
Show the answer
For the same structural reason from two positions. The operator's actual product is power, cooling, space, and uptime, so all of their real constraints — margins, redundancy, stranding, failure risk — live in the facility layer and are invisible in compute metrics like utilization or revenue-per-rack. The investor is buying two machines with different lifespans; the facility is the long-lived asset where durable value and the largest risks (whole-hall failures, obsolescence, capacity limits) sit, while the compute turns over every few years. Both would otherwise price or run only half the asset and be surprised by the machine they stopped watching.