Case Study · Data Center Siting

Where would you build a 100 MW AI data center?

A worked example of how to use grid data to shortlist a site. We take one realistic brief and run it across five candidate grids on the four metrics that actually decide a build, system-level room, power price, carbon intensity and reliability, then show how to reproduce it live for your own parameters.

The brief

Hypothetical: a 100 MW AI training cluster, online within ~3 years

  • Power: 100 MW continuous, 24/7, roughly a small city's worth of load.
  • Cost priority: wholesale electricity is the dominant operating cost; every $10/MWh ≈ ~$8–9M/year at this scale.
  • Carbon priority: a net-zero parent and low-carbon-compute customers, Scope 2 emissions matter.
  • Timeline: needs a grid that can actually connect new large load on schedule, not one with a multi-year queue.

No single grid wins on all four. The exercise is about trade-offs, and seeing them side by side is exactly what a grid-intelligence view is for.

Five candidates, four metrics

Indicative profiles for comparison. Open the live app for current, sortable figures, values move hourly.

GridPower costCarbon (gCO₂/kWh)HeadroomConnection outlookFits brief?
Texas · ERCOTLow~320HighFast by US standards; energy-only marketCost + speed
Virginia · PJMMid~290TightWorld's largest hub, but queue & capacity now constrainedEcosystem, not capacity
QuébecVery low~5ModerateHydro-rich, cool climate; allocation-dependentCost + carbon
Sweden (SE)Low~45ModerateHydro + nuclear + wind; northern zones favourableCarbon + cost
Pacific NW · USALow~80ModerateHydro-backed, cool; established DC corridorBalanced

Carbon figures are matched to national/official sources; system-level room and connection outlook are directional. The point isn't the exact number, it's that the ranking flips depending on which metric you weight.

Reading the trade-offs

IF COST + SPEED

Texas (ERCOT)

Cheapest fast-to-connect option, but ~320 gCO₂/kWh works against a net-zero story unless paired with renewable PPAs (which ERCOT has plenty of).

IF CARBON IS KING

Québec / Sweden

Near-zero (~5) and low (~45) carbon with low cost. Trade-off is allocation and distance from major demand centres.

IF ECOSYSTEM MATTERS

Virginia (PJM)

Unmatched connectivity and operators, but system-level room and the interconnection queue are now the binding constraint, not power itself.

IF BALANCED

Pacific NW

Low cost, low carbon (~80), cool climate and an established corridor, a strong all-rounder when no single metric dominates.

Run this yourself, live

This case study is a snapshot; the real value is doing it with current data for your own brief. In PowerGridIQ you can:

① Open the live map and switch the layer to Carbon, Price or Utilization to see the world re-colour by your priority. ② Use the access read to rank regions on system-level room, which is a screen rather than confirmation that a specific site can take 100 MW. ③ Open the Data Hub, sort the all-regions table by any column, filter to your shortlist, and export it to CSV for your own model. ④ Open any region for its fuel mix, reserve margin and reliability detail.

Open the live comparison →

? FAQ

What's the single most important grid metric for siting a data center?

There isn't one, it's the combination. Power cost dominates operating expense, carbon intensity drives location-based Scope 2 emissions, system-level room and the interconnection queue decide whether you can connect on schedule, and reserve margin signals reliability risk. The right site is the best weighted trade-off for your priorities.

Why does the interconnection queue matter as much as price?

Generation can be built far faster than the transmission and interconnection that connect new load. In the busiest US markets, queues stretch years and only a fraction of projects ever connect, so a cheap grid you can't join on time may be worse than a pricier one you can.

How current is the data behind this?

Live grids update sub-hourly from system operators (EIA, ENTSO-E, IESO, AEMO and others); carbon and fuel-mix figures are matched to national/official sources. Regions without a live feed are clearly labelled as estimates.