Methodology

How the Data Center Index estimates grid and gas impact

From announced megawatts to expected load, energy, gas burn and a 1-to-4 bolt impact rating. Every figure marked “expected” or “estimated” is a Discovery Energy model output, not a reported number.

Scope

  • Coverage: data center campuses of 100 MW or more in the US and Canada that entered service since 2021, or are under construction, approved, announced or proposed. Crypto-mining sites converting to AI/HPC are included and tagged by segment.
  • Threshold: 100 MW. Below roughly 20 MW the grid effect is minor.
  • Data: public sources only: news reports, regulatory and utility filings, grid operator publications, government sites, company releases and public trackers. No paid datasets. Every source is listed in the source log.
  • Geography of impact: bill effects follow the grid region and utility a campus connects to. Campuses with on-site gas generation shift the impact from electric bills to the gas system.

Method, step by step

  1. Normalize size

    Published MW can be IT load, facility load, utility service, on-site generation nameplate or campus potential. IT load × 1.2 (PUE); generation nameplate × 0.8 (plants are built with spare units); others as published.

  2. Phase to 2030

    Horizon factor by facility MW: up to 1,000 MW counts 100%; 1,000 to 3,000 MW counts 60%; above 3,000 MW counts 30%.

  3. Weight by build stage

    Expected MW = facility MW × horizon × status probability (table below).

  4. Convert to energy

    Average MW = expected MW × 75% load factor. Annual GWh = average MW × 8.76.

  5. Split grid from on-site

    On-site gas and fuel-cell campuses count 100% on-site; hybrids are split by on-site nameplate × 0.8 ÷ facility MW.

  6. Estimate gas burn

    MMcf/d = average MW × 24 × heat rate ÷ (1,036 Btu/cf × 1,000). On-site uses the on-site heat rate; grid share uses the region's gas share of new supply × 7,500 Btu/kWh. Design-day Dth/d uses a 95% load factor. Check: 1 GW at 6,000 Btu/kWh ≈ 139 MMcf/d, matching Enverus's 140.

  7. Electric bill effect

    Where a market monitor quantifies the data center share of capacity cost (PJM): capacity price × 365 × data center share per MW of peak load contribution. For 2028/29: $325 × 365 × 38% = $45,078 per MW-year, or $7.35/MWh at a 70% load factor. Other markets are calibrated from resource plans, rate cases and tariffs in later passes.

  8. Rate the impact

    Each campus gets a grid score and a gas score; the rating is the higher of the two (scale below).

Impact rating: 1 to 4 bolts

The rating measures strain on the system, not who pays. Bolts are solid for operating or under-construction campuses and outlined for approved, announced or proposed ones. The label shows whether the grid score, the gas score or both drive the rating.

RatingGrid-served load by 2030Local design-day gas drawTypical meaning

Gigawatt-class
800 MW+120,000+ Dth/dReshapes regional planning and capacity prices; pipeline expansion territory

Regional
400–800 MW60,000–120,000 Dth/dAbout a large power plant's output; new transmission; may need a new gas lateral

Utility-scale
150–400 MW25,000–60,000 Dth/dDedicated substation and high-voltage lines

Local
Under 150 MWUnder 25,000 Dth/dServed by expanding an existing substation
  • Local gas draw = on-site design-day gas + dedicated-plant gas + 50% of regional grid gas.
  • Metro rating (sum of a metro's campuses): 2 bolts from 500 MW / 50,000 Dth/d, 3 from 1,500 MW / 150,000 Dth/d, 4 from 3,000 MW / 300,000 Dth/d.
  • Current distribution (87 campuses): 4 bolts 23 · 3 bolts 29 · 2 bolts 32 · 1 bolt 3. Drivers: grid 71 · gas 10 · grid + gas 6.

Assumptions

ParameterBaseLowHighBasis
PUE (IT → facility)1.201.101.40Discovery Energy assumption for new hyperscale/AI builds
On-site nameplate → served load0.800.700.90Redundancy (N+1/N+2)
Average load factor75%60%85%Near-flat 24/7 load; Dominion GS-5 defined around 75%
Design-day load factor95%90%100%Peak-day pipeline sizing
Heat rate, grid gas serving new load7,500 Btu/kWh6,9609,000EIA Table 8.2; new CCGTs average 6,960
Heat rate, on-site turbines/engines9,500 Btu/kWh8,90011,000EIA: gas turbines ~11,000, engines ~8,900
Heat rate, on-site fuel cells7,000 Btu/kWh6,5008,000Discovery Energy assumption
Gas heat content1,036 Btu/cf1,0301,040EIA average
Example large-user load factor70%50%85%For bill examples

Build-stage probabilities

StatusCounts toward 2030Meaning
Operating100%In service, no further phases announced
Operating, expanding90%Partly energized; later phases under construction
Under construction80%Site work underway
Approved65%Utility agreement or power contract, grid allocation, or zoning approval
Announced35%Named developer and site, no construction
Proposed15%Developer only, no named tenant, power plan unresolved
Cancelled / withdrawn0%Kept for history

Gas share of new supply by region

RegionGas share
PJM60%
ERCOT55%
SPP50%
MISO60%
Southeast utilities75%
West: Mountain/Northwest50%
West: Desert Southwest60%
Alberta (AESO)90%
Alaska90%
NYISO60%
ISO-NE70%
CAISO35%
Ontario (IESO)40%
Québec (Hydro-Québec)0%

Version 0 judgment, to be calibrated against utility resource plans. Campuses served by plants built for them use a 100% gas share.

Tariff protection

Share of data-center-driven grid cost assumed to reach other customers: Strong 30%, Partial 60%, None 100%, Under review 100%.

Confidence grades

GradeMeaning
AVerified against a primary source
BSeveral consistent sources
COne secondary source
DConflicting or estimated figures

Timeline model

The timeline spreads each campus's expected MW across years by build stage until real dates replace it: operating, all by end of 2025; operating-expanding, 50% by 2025 and 25% in each of 2026 and 2027; under construction, up to 1 GW announced half in 2026 and half in 2027, 1 to 3 GW 40/30/30 over 2027–2029, over 3 GW 30/35/35 over 2028–2030; approved, half in 2027 and half in 2028; announced, 30/35/35 over 2028–2030; proposed, half in 2029 and half in 2030.

Calibration

MeasureIndexBenchmarkSource
Expected data center load by 203054.9 GW62 GW new, 2026–2030 (Lower 48)Enverus, July 2026
PJM expected load by 203016.4 GW35.1 GW accepted, 2026–2031PJM 2026 load forecast
Added gas burn by 20305.1 Bcf/d3–10 Bcf/dS&P Global, PwC and others

The index sits inside these ranges while still missing much of the 100 MW to 1 GW tier, so totals will rise. Status probabilities are tightened if the index runs hot.

Limits

  • Published capacity figures often conflict across trackers; the index records the basis used and flags conflicts in each campus note.
  • The bill-effect calculation is quantified only for PJM so far.
  • Metro gas headroom (needed for the gas stress ratio and the ±1 local adjustment) is still being compiled.