The solid lines show 2024 (the best year in our dataset). The dashed lines show the worst weather year across 2015–2024. The gap between them is the risk that single-year analysis hides — and the reason Monte Carlo matters for grid planning.
At 70% and 90% renewables, the grid already exceeds reliability standards with zero data centers in bad weather years. The problem isn't just data centers — it's the combination of high renewables, variable weather, and insufficient dispatchable backup.
Model: hourly merit-order dispatch, 10 weather years (2015–2024), demand-normalized to 2024, per-2 GW resolution. Solid = 2024, dashed = worst year.
With forced outages, the tipping point shifts to zero. The deterministic model put the tipping point at 4–6 GW. Run it stochastically (200 draws, 7–10% forced outage rates, ±3% demand noise) and the 70% RE grid is already unreliable with no data centers at all — 55 mean blackout hours, P90 of 115. At that point you're no longer asking how many data centers the grid can handle. You're asking whether there's enough dispatchable gas to keep the lights on, period.
Same grid, same weather — different answer. The deterministic model greenlit 4–6 GW of data center load. The stochastic model says the 70% RE grid fails at 0 GW. The only variable that changed is whether you include forced outages. That single modeling choice is the difference between approving billions in interconnection and pulling the emergency brake.
What Drives the Answer?
Across eight investigations, we varied dozens of parameters. Not all of them matter equally. This table summarizes which inputs drive the key findings and which ones don't — the core output of Analysis Driven Modeling.
| Parameter | Range Tested | Effect on Breaking Point | Decision Impact |
|---|---|---|---|
| Weather year | 2015–2024 (10 years) | 2–3 GW (worst year) to 4–6 GW (best) | Determines if grid survives |
| Forced outage rate | 7–10% (stochastic, 200 draws) | Shifts breaking point to 0 GW at 70% RE; adds ~15 GW to gas requirement | Makes or breaks reliability |
| RE penetration | Current / 70% / 90% | Higher RE = more volatile; 70% RE grid unreliable at 0 GW DC in worst year | Determines investment path |
| DC load flexibility | 0–50% flexible; 3–10% DR | 5% DR doubles grid capacity (4 GW → 8 GW); 30% flex cuts blackouts by ⅓ | Reframes DC as asset, not just load |
| Transmission limits | Single-node / 2-zone / 6-zone | 67h → 528h blackout for same 10 GW scenario (8× increase) | Spatial fidelity is critical |
| Gas-electric coupling | 0.5%–5% coupling strength | At 3%+ coupling, any shock >20% cascades to total collapse | Changes decision from “weatherize” to “redesign” |
| Dispatch intelligence | Heuristic / 6h forecast / perfect foresight | 6h forecast captures 97% of optimal value; minor capacity effect | Operational improvement, doesn't change capacity needs |
Sensitivity ranges reflect actual parameter sweeps from Q1–Q8. See individual investigation pages for detailed results.