REPORT2026-01-01·Bloom Energy (survey of hyperscalers, colos, utilities/IPPs, equipment suppliers)

Bloom Energy — 2026 Data Center Power Report: When Power Defines Growth

Bloom Energy
COMPILED NOTES

US IT load capacity projected ~80 GW (2025) to ~150 GW (2028); developers expect power up to 2 years earlier than utilities believe deliverable; onsite-power expectations rose 22% in 6 months to ~33% of sites by 2030 (fuel cells leading onsite tech at 47%)

Bloom Energy — 2026 Data Center Power Report

Abstract

Bloom Energy's fourth installment of its Data Center Power Report (following the inaugural report, a Mid-Year Pulse, and prior waves back to April 2024) surveys hyperscalers, colocation providers, utilities/IPPs, and equipment suppliers (n=92 core respondents in the November 2025 wave; longitudinal series April 2024 n=30, November 2024 n=64, April 2025 n=100, November 2025 n=152 total across respondent types, 84% US-based) to track how power availability is reshaping the data center industry. Findings are corroborated by follow-on interviews with industry leaders.

Key Contributions

  1. Geographic reallocation of data center growth. Total US IT load capacity is projected to roughly double from ~80 GW in 2025 to ~150 GW in 2028 — more than twice the level projected in 2024 forecasts. Texas is projected to reach >40 GW by 2028 (~30% of total US demand, a 142% increase in relative market share); Georgia's share rises 75%. Legacy markets decline sharply: California -50%, Iowa -60%, Oregon -67%, Nebraska -75% in relative market share. "Rest of US" collectively gains 21% share as developers extend siting into power-advantaged secondary markets (Idaho, Mississippi, Maryland, Kansas, Wisconsin, North Dakota, Louisiana, Kentucky, South Carolina, Indiana, South Dakota).
  2. Gigawatt-scale campuses become normal. Share of new campuses expected to exceed 1 GW rises from roughly 1 in 5 by 2030 to nearly 1 in 3 by 2035 (20% → 32% of portfolio in the 1-2 GW + >2 GW bands). A single 1 GW campus consumes roughly 20% of New York City's entire electricity load. ERCOT raised its 2030 data-center-growth estimate from 29 GW to 77 GW between 2024 and 2025; PJM revised 2030 peak-demand estimates up 10%.
  3. Time-to-power misalignment between developers and utilities is widening. Utility/IPP respondents say time-to-power will take ~1.5-2 years longer on average than hyperscalers and colocation providers expect. The gap widened over the prior six months specifically in Northern Virginia, the Bay Area, and Atlanta. More than half of developers report securing power got more difficult over the past twelve months.
  4. Onsite (behind-the-meter) generation is becoming structurally entrenched, not a bridge. Average expectation among hyperscalers/colos is that roughly one-third of US data centers will run on 100% onsite power by 2030 — up 22% from the prior (April 2025) survey wave — and 44% by 2035. 73% of respondents are actively evaluating or already deploying onsite power (up from 27% not evaluating). Among onsite technologies under evaluation/deployment: fuel cells lead at 47%, reciprocating engines 38%, mobile turbines 33%. Permanent onsite generation is displacing "temporary bridge-to-grid" as the top-ranked long-term strategy for minimizing both development timelines and cost — a pattern the report compares to cement, refining, and steel, other power-intensive industries that self-generate.
  5. Next-generation electrical architectures are arriving within two years, driven by rising rack density: by year-end 2028, 60% of respondents expect to adopt high-voltage central busways and 45% expect DC (direct current) distribution architectures — both chosen partly because they integrate more cleanly with onsite generation (fewer power-conversion stages between generation and IT load).

Methodology

Double-blind surveys commissioned between Bloom Energy and respondents (hyperscalers, colocation providers, utilities/IPPs, equipment suppliers), fielded April 2024 (n=30), November 2024 (n=64), April 2025 (n=100), and November 2025 (n=152, the wave underlying this report; core analysis subsample n=92-113 depending on question). Supplemented by structured interviews with industry leaders to pressure-test survey findings against real-world implementation. Geographic market-share figures sourced from DC Byte analysis as of October 2025 (50 states + DC).

Results

See Key Contributions above — the report is organized around five numbered findings, each with its own data visualization (state-level market-share map, campus-size distribution charts by year, time-to-power expectation-gap chart by metro, onsite-generation-share donut charts, next-gen-architecture adoption bar charts).

Limitations

  • Bloom Energy is itself a leading fuel-cell (stationary onsite generation) vendor to data centers — the survey is commissioned by an interested party, and the finding that fuel cells lead onsite-technology evaluation (47%) should be read with that in mind, though the underlying respondent pool (hyperscalers, colos, utilities, equipment suppliers) is third-party.
  • Figures are respondent expectations/intentions (survey sentiment), not realized deployment — e.g., the "33% onsite by 2030" figure is what developers say they expect, not a confirmed capacity buildout.
  • US-weighted sample (84% US-based respondents); limited visibility into non-US markets.
  • n=92-113 for the core analytical questions is a moderate sample for claims segmented down to individual metro markets (e.g., the 15-market time-to-power chart).

Full Content

This is the single richest primary-adjacent industry data source ingested this cycle for the AI-datacenter-power lens. Three findings matter most for this KB's thesis that power is now a binding constraint on the AI buildout alongside memory and packaging: (1) the 80→150 GW three-year doubling of US IT load capacity, which quantifies the scale of the problem; (2) the systematic 1.5-2 year gap between what developers expect and what utilities say they can deliver, which is the demand-side mirror of the FERC-filing-based interconnection-queue analysis in Power Grid Infrastructure for AI Data Centers (arXiv:2606.00941, ingested same day) — two independently-sourced datasets (developer survey vs. regulatory filings) now agree the grid cannot deliver power on the AI industry's preferred timeline; and (3) the entrenchment of onsite/behind-the-meter generation as a permanent strategy rather than a stopgap, with fuel cells the leading technology under evaluation — directly relevant to this KB's on-site-gas/nuclear/SMR coverage gap that this cycle's discovery pass targets (see also SMR nuclear datacenter deal tracker, ingested same day, for the specific nuclear-SMR subset of onsite generation).


Source: Bloom Energy — 2026 Data Center Power Report: When Power Defines Growth. Published January 2026; survey fielded November 2025.

RELATED · IN THE BASE
Bloom Energy — 2026 Data Center Power Report: When Power Defines Growth | Knowledge Base | MenFem