Questions 1–4 used a single-node model — the whole grid as one point. But ERCOT isn't uniform. West Texas has 60% of the wind and 40% of the solar, but limited gas and a congested 20 GW export corridor to the load centers. Does siting data centers in West Texas — where they can absorb curtailed renewables — change the answer?
To answer this, we split the model into two zones connected by a transfer limit: Zone W (West Texas) and Zone E (rest of ERCOT). Same dispatch algorithm, but now topology-aware.
67 hours vs. 528 — same grid, same year. The single-node model says 10 GW of data centers cause 67 blackout hours. The 2-zone model says 528. Nothing changed except acknowledging that transmission lines have capacity limits. A model without topology can't even represent the question "where should we build?"
Model: 2-zone hourly dispatch with 20 GW transfer limit. Same algorithm as Q1–Q4, applied per-zone.
Going Deeper: 6 Zones
Two zones is a start, but ERCOT isn't binary. We expanded to six zones — West TX, Panhandle, North TX (DFW), Houston, South TX, and Coastal — connected by eight capacity-limited transmission corridors. Now we can see which specific zone constrains growth.
More zones, more nuance. At 10 GW DC, everything in Houston: 639 blackout hours. WTX/NTX split: 394 hours. Spread evenly: 469. Two zones couldn't distinguish these — the optimal placement depends on which specific corridors you're willing to congest, and you need enough zones to see those corridors individually.
Model: 6-zone hourly dispatch with 8 transfer corridors (WTX→NTX 12 GW, PAN→NTX 8 GW, NTX→HOU 15 GW, NTX→STX 8 GW, HOU→STX 6 GW, STX→COS 4 GW, WTX→STX 5 GW, COS→HOU 5 GW). Greedy surplus→deficit matching with post-dispatch gas-electricity transfers. Zone allocations are illustrative estimates based on public ERCOT data, not validated against ERCOT’s internal network model.
Fidelity has a floor, not just a ceiling. Single-node: 67 hours. Two zones: 528. Six zones: Houston is worst, WTX/NTX split is best. The single-node model didn't give the wrong answer to "where should we build?" — it couldn't represent the question at all. Some decisions have a minimum model complexity below which the analysis is meaningless.