PJM Load Growth: What Data Center Developers Need to Underwrite

PJM's 2026 forecast shows data center load moving from site factor to market constraint.

Data center developers in PJM territory are no longer underwriting a site and then checking power. Power is the underwriting case.

PJM's 2026 Long-Term Load Forecast says summer peak load across the RTO is projected to grow 3.6% per year over the next 10 years, reaching 222,106 MW in 2036. Net energy load is projected to grow faster, at 5.3% per year, reaching 1,437,629 GWh in 2036. PJM specifically identifies growth in data center load as a forecast adjustment across zones including AEP, ATSI, APS, BGE, ComEd, Dominion, PECO, PEPCO, PPL and others.

For developers, that matters because the same megawatt request now means different things by zone, substation, queue position and timing. Build works with institutional development teams where that distinction changes the go or no-go decision before a land deal reaches investment committee.

Why PJM load growth is a development constraint

PJM is not a single power market from a site selection perspective. It is a 13-state regional transmission organization with zones, locational deliverability areas, utility territories and local transmission constraints. A 100 MW requirement in Northern Virginia is not the same underwriting problem as 100 MW in Ohio, Pennsylvania or New Jersey.

The 2026 forecast makes that plain. PJM lists data center load as a reason for forecast adjustments in many zones, but the practical effect is uneven. Some zones have substation capacity, transmission headroom and clearer utility processes. Others have demand growth arriving faster than transmission upgrades, generation additions or interconnection studies can absorb.

That creates three development risks.

First, the site can be physically viable but electrically stranded. The land, zoning and fiber can work while the power delivery schedule kills the project.

Second, a utility indication can be directionally positive without being financeable. A preliminary capacity conversation is not the same as a committed delivery date, upgrade cost estimate or service agreement.

Third, the market can change while the developer controls the land. A site that looked competitive at LOI can lose its edge if competing large-load requests arrive in the same node, utility queue or transmission pocket.

The numbers developers should read first

The headline load growth number is useful, but it is too broad for underwriting. Developers need to pull the forecast apart.

Start with the PJM zone. PJM says individual zones show 10-year annualized summer peak growth ranging from negative 0.2% to 6.4%, with a median of 1.6%. That spread is the market. A zone near the high end is not automatically bad, but it needs a stronger explanation of where capacity comes from.

Then look at winter peak. PJM forecasts winter peak load growth of 4.0% per year across the RTO over the next 10 years. AI infrastructure is often framed around cooling and summer peak, but winter reliability can matter for capacity planning, gas-electric coordination and firm load treatment.

Then look at net energy. PJM's 5.3% annual net energy growth projection is a warning that load is not just peaking higher, it is consuming more around the clock. That is closer to the data center reality. Training clusters, inference workloads and high-utilization colo deployments do not behave like ordinary commercial demand.

Finally, read the load adjustment process. PJM says near-term forecast years need firm commitments, including Electric Service Obligations or construction commitments, while longer-term projects are treated as non-firm and derated. Developers should mirror that discipline internally. A power assumption should be weighted by evidence, not enthusiasm.

How AI changes the underwriting workflow

AI does not create grid capacity. It does make the capacity question harder to miss.

A practical workflow starts with automated site triage. The system scores parcels against utility territory, substation proximity, transmission voltage, queue congestion, known large-load activity, permitting overlays, fiber routes, water risk and land control status. Weak sites are removed before the team pays for deeper diligence.

The next step is evidence extraction. AI can parse PJM reports, utility tariffs, interconnection filings, integrated resource plans, county approvals and state siting documents. The output is not a single answer. It is a structured evidence file that shows what supports the power case and what contradicts it.

Then the development team runs timing scenarios. The base case should include requested load, staged ramp, backup generation assumptions, potential behind-the-meter resources, interconnection milestones, likely network upgrades and utility decision dates. The downside case should assume delays in study completion, transformer procurement, transmission upgrades and regulatory approval.

Human judgment still matters most at three points: interpreting utility conversations, negotiating service structure and deciding whether to control the site before the power case is firm. AI can surface the pattern. It cannot replace the commercial call.

What this means for capital allocation

The old underwriting sequence treated power as a diligence item. In PJM, that is now too late.

For institutional capital, the question is not whether data center demand is strong. The question is whether the specific site can deliver capacity on the tenant's timeline at a cost the project can absorb. That turns PJM load data into an investment input.

A credible investment committee memo should show the zone forecast, utility position, queue exposure, power delivery milestones, upgrade cost range and fallback strategy. It should also separate what is confirmed from what is assumed. That separation matters because the strongest data center markets now attract the most speculative load requests.

The implication is direct. Land basis is not enough. Fiber is not enough. Entitlement path is not enough. In PJM, the developer with the clearest power evidence has the better site, even if the dirt costs more.