The Data Center Power Index scores data-center markets on two daily-refreshing numbers — Excess Power (buildable headroom) and Constraint (friction to new load) — and combines them into a BUILD / CAUTION / AVOID verdict plus a single 0–100 composite rank. This is the canonical, machine-readable reference behind The State of Data Center Power and every /dcpi/<market> page.
Verdict model · v3 · recomputed daily · last reviewed 2026-05-31
Higher means more buildable headroom — capacity that can absorb new data-center load. It blends ISO grid headroom, approved-but-unbuilt generation (queue velocity), signed-contract subscription ratio, local renewable surplus, and distance from saturated tier-1 concentrators (Northern Virginia, Phoenix, Silicon Valley).
Higher means more friction. It blends interconnection-queue wait time, transmission-congestion (LMP spread), substation distance to load, water stress, and regulatory friction (moratoria, special-use permitting).
The verdict is derived directly from the two scores. These thresholds are the live values in production (derive_verdict):
if excess >= 65 and constraint <= 50: verdict = BUILD elif excess >= 50 and constraint <= 70: verdict = CAUTION else: verdict = AVOID
| Excess Power | Constraint | Verdict |
|---|---|---|
| ≥ 65 | ≤ 50 | BUILD |
| ≥ 50 | ≤ 70 | CAUTION |
| anything else | AVOID | |
A single sortable rank derived from the two scores plus time-to-power (TTP), with a verdict-aware quality multiplier so a data-sparse market can't outrank a real BUILD market. This is the live formula (derive_composite_score):
raw = (excess * 0.60)
+ ((100 - constraint) * 0.30)
+ ((1 - min(ttp_months, 60) / 60) * 100 * 0.10)
composite = clamp(0, 100, raw * verdict_multiplier)
| Verdict | Multiplier | Rationale |
|---|---|---|
| BUILD | 1.00 | Trusted, actionable |
| CAUTION | 0.85 | Trusted but bordered |
| AVOID | 0.60 | Known grid issues |
| LOW_SIGNAL | 0.35 | Data integrity unknown — heavy penalty |
Time to Power (TTP) is the estimated months from application to energization, capped at 60 in the composite. It is published per market alongside the scores.
| Signal | Source | Cadence |
|---|---|---|
| Grid headroom · fuel mix · carbon intensity · demand | 7 live US ISOs (ERCOT, CAISO, NYISO, MISO, PJM, SPP, ISO-NE) via EIA-930 v2 + ISO feeds; intl: Hydro-Québec, AESO, Nord Pool | 15 min – hourly |
| Interconnection-queue depth + data-center share | Per-ISO interconnection queues (ERCOT MIS, PJM, CAISO, MISO, SPP, NYISO, ISO-NE) | daily |
| Transmission congestion (LMP spread) | ISO LMP node data, rolling window | 15 min |
| Substations | HIFLD substations (126,427) | quarterly |
| Water stress | EPA + USGS | monthly |
| Regulatory friction | County/state filings (curated) | quarterly |
The index recomputes daily. Underlying ISO source pulls run every 15 minutes to hourly. Stale data points are flagged in the JSON.
Ranked leaderboard (all markets, every score + verdict): /api/v1/dcpi/leaderboard. Per-market detail: /dcpi/<market-slug>. The rolled-up daily report: /api/v1/reports/state-of-power. No API key required. All CC-BY-4.0.
Note on versions. An earlier methodology page described a weighted four-component DCPI v2 model. The live index uses the verdict model documented here (Excess Power + Constraint → BUILD/CAUTION/AVOID); this page is the source of truth that matches production.