Summary

  • Exelon’s September 2026 PJM submission lists about 37 GW of large-load requests, with about 12 GW of high-probability capacity including in-service projects; its demand forecast is about 6 GW in 2031 and 8.7 GW in 2040.
  • PJM will publish its 2027 long-term load forecast in January. The number to watch is the demand that remains after project status, ramp schedules and utilization assumptions have done their work.

One filing, three different numbers

Exelon’s September 29 presentation to PJM gives the data-centre boom a useful set of labels. Its total request list is about 37 GW of capacity. Roughly 12 GW, including load already in service, meets Exelon’s high-probability screen; about 25 GW of requests sits outside the forecast. The high-probability starting point fell from 19 GW in the prior-year submission. The filing’s demand forecast is about 6 GW in 2031 and 8.7 GW in 2040, including 7.5 GW in ComEd territory and 1.2 GW in PECO territory.

Capacity is what a project asks to draw. Demand is what Exelon expects to appear on the system after timing and use are applied. For projects with a signed Transmission Security Agreement, or a comparable binding agreement, Exelon requires a submitted letter of credit and uses an approximately four-year ramp with 80% utilization. A project without a TSA must have entered Phase 1 engineering, with a customer deposit and approved design costs; the default assumption is about eight years and 70% utilization. The spreadsheet’s biggest number does not get to skip the calendar.

That calendar starts with a meter. Exelon’s October 30 earnings presentation puts ComEd’s in-service data-centre load at about 800 MW through September 2026 and reports a 28% compound annual growth rate from the comparable year-to-date period in 2022. That is billed load in northern Illinois. It is a different measure and territory from a request list containing future ComEd and PECO projects.

The company’s October 30 investor deck also calls about 11 GW “high probability load”: roughly 9 GW in ComEd and 2 GW in the Mid-Atlantic. It says about 45% of ComEd’s high-probability load has TSAs backed by roughly $1.2 billion of collateral. PJM’s September filing covers ComEd and PECO data centres and reports 12 GW of high-probability capacity. Exelon publishes no project-by-project crosswalk between the two totals. That is the fact missing from the headline arithmetic: which projects, territories and dates connect the two disclosures.

PJM’s guideline asks for a customer-level record: facility owner, location, site control and permitting, project stage, and financial responsibility. Missing customer data can cause an adjustment to be rejected or discounted. At that point, a load forecast is an engineering and credit file, not just a slide of capacity.

Forecasts have a construction bill

PJM says it has no load interconnection queue. Utilities and other load-serving entities submit large-load adjustments to PJM’s long-term forecast, with contract status used to identify firm commitments. PJM describes near-term Electric Service Obligations or Construction Commitments as examples of firm commitments; longer-dated projects carry non-firm status and more uncertainty. Its annual process sets a November/December review point and a January publication date for the 2027 forecast.

Exelon’s own presentation identifies more than $1 billion of new-business transmission opportunity to connect committed high-density load. It also lists $12–17 billion of transmission opportunities beyond its plan, an estimate made as of Q1 2026 and largely expected from 2030 onward. Those figures give the forecast a capital consequence; they are company-identified opportunities, not a project award or approved spending schedule.

Investors should read PJM’s next forecast as a dated construction map. If near-term demand holds while later years move, the ramp assumptions are doing the work. If the 2031 figure falls after the status review, the request list was a poor proxy for buildable load. The actual meter, the submitted capacity and the PJM demand curve belong on separate lines. The market gets the story wrong when it lets one label impersonate all three.

Sources