State of the
Data Center Market

The state of the industry in one chart

The headline: 2026 is the year data center demand decisively outran transmission capacity. Across the DCPI-scored markets, the average time-to-power is now 24-60 months. Markets that were "ready to build" in 2023 — Northern Virginia, Phoenix, Silicon Valley — are now "AVOID" or "CAUTION." Meanwhile, secondary markets (Cheyenne WY, Columbus OH, Dallas TX) have become the new gravity centers.

Top 10 BUILD markets of 2026

DCPI's verdict comes from two scores: BUILD needs excess power ≥ 65 and constraint ≤ 50. Time-to-power does not change a verdict; it carries 10% of the composite score that orders the markets below.

5 markets to watch (AVOID)

These markets earn the AVOID verdict in 2026 — high constraint scores, multi-year queue waits, or reserve margins below safe operating thresholds. None will recover quickly; if your build needs power before 2029, look elsewhere.

AI infrastructure: the new gravity

AI training clusters changed the physics of data center siting. A frontier-model training run pushes 50-500 MW for 6-18 months continuously. Three structural shifts followed in 2026:

1. Power location displaced fiber location. Pre-2024, hyperscalers picked markets by latency-to-customer. Post-2024, the constraint is "where is there 200+ MW of power available within 36 months?" Latency budget has become a secondary filter, not a primary one.
2. Behind-the-meter generation went mainstream. Microgrids, on-site solar+storage, and dedicated PPAs are no longer fringe — they're table stakes for any 100+ MW campus. The interconnection queue is too slow to wait on.
3. Secondary markets ate the trade. Cheyenne, Columbus, Mt Pleasant TN, San Antonio — these are the new Tier-1 candidates. DCPI tracks the migration: aggregate excess power score for "secondary" markets (rank 30-100) rose 18% in 2026 while Tier-1 markets fell 11%.

Grid stress: where the queue is breaking

Across the 11 tracked ISOs, three are now operating at or beyond safe planning margins. The pattern in the data is unmistakable:

SPP is the contrarian opportunity. Headroom is real, queue is short, and the wind generation backbone gives PPA buyers leverage. Expect 2026-2027 to be the SPP land-grab year.

Fiber: the second crisis

3,282 long-haul fiber routes is enough on paper. In practice, IPv6 routing, lit-vs-dark availability, and cross-border peering create local famines. AI inference workloads (low-latency, north-south traffic) and training workloads (high-bandwidth east-west) compete for the same capacity. 2026 watch: which secondary markets get dark-fiber consortia announcing first.

M&A: 1,800+ deals tracked

2024-2026 saw the largest single-asset-class M&A run in private infrastructure history. The pattern: KKR, Brookfield, GIP-style buyers competing for the same 50-200 MW campuses at $12-18M/MW. The DC Hub transaction database tracks 2,222 announcements; selected highlights:

"What used to be a $5M/MW asset class in 2019 trades at $15M/MW in 2026 because power-secured capacity has become harder to underwrite than the building. Buyers are paying for the queue position, not the concrete."

2026 outlook + the methodology

Three predictions for the next 12 months:

1. Behind-the-meter generation will become a default RFP requirement for every >50 MW build.
2. Two secondary markets (likely from: Cheyenne, Columbus, San Antonio, Mt Pleasant) will absorb their entire 2027 inventory before construction starts.
3. One Tier-1 ISO will declare a moratorium on new large-load interconnection requests, similar to ERCOT's 2024 freeze. PJM is the leading candidate.

How DCPI scoring works

Two scores set the verdict. Modeled time to power adds 10% to the ranking.

• DCPI Excess Power score (0-100): Available headroom MW vs market demand growth. Higher = more headroom.
• Grid Constraint score (0-100): Grid congestion + transmission limit factor. Higher = worse.
• Modeled time to power (months): Modeled months until power for a large new load. It adds 10% to the composite ranking and does not change the verdict.

BUILD = excess ≥ 65 and constraint ≤ 50. CAUTION = excess ≥ 50 and constraint ≤ 70. AVOID = everything else. Time-to-power is not a verdict input; it carries 10% of the composite rank. Full methodology: dchub.cloud/dcpi/methodology

The 2026-05-15 edition published 232 scores. DCPI is recomputed daily; the live series, 332 markets as of 2026-10-27, is at The State of Data Center Power.