The Interconnection Gate
Everything downstream of the fence is engineering. The gate is politics, physics, and a queue. A rack cannot draw a watt the grid will not deliver, and the grid delivers on its own schedule — measured in years, not quarters. So the first design decision in any compute buildout is not the chip, the cooling, or the topology. It is the point of interconnection: how many megawatts, at what voltage, energized on what date. Master this and the rest of the stack falls into place as a series of consequences. Miss it and the most elegant data hall in the world sits dark, waiting on a transformer and a study it cannot control.
Knowledge check
Why is the point of interconnection, not the IT load, the true starting constraint for a large compute site?
Show the answer
Because the grid caps how much power can physically and contractually reach the site, and it sets that cap on a multi-year timeline the developer does not control. Megawatts available, voltage class, and the energization date are fixed at the gate; every downstream design choice — transformer count, distribution voltage, redundancy tier — is a consequence of what the interconnection will actually deliver and when.