When Shoulder Season Disappears

The electricity system has traditionally depended on a predictable pause. Spring and autumn brought milder demand, giving generators and transmission owners time to take equipment offline for maintenance before the next winter or summer peak. That pause is becoming less dependable in the regions where large digital loads are growing fastest.

On 15 October 2026, PJM described more than 61,000 MW of generation outages during an unusually tight shoulder-season week. Only 66% of fleet capacity was available. About 15,000 MW of the outages were unplanned, while PJM had already recalled roughly 10,000 MW. The operator also said approximately 11,000 MW of data-center load is already connected to its system.

Those figures do not mean 11,000 MW of live IT capacity, nor do they establish how much individual campuses can use at any moment. They do show that the old planning assumption — lower seasonal demand creates room for maintenance — is under pressure from loads that operate year-round.

For data-center investors, developers and operators, this is more than a grid-operations story. It changes how delivery dates, redundancy, backup generation and expansion options should be underwritten.

Seasonal slack is now an infrastructure variable

A development model may include a utility commitment, a substation program and a phased campus plan. It often treats the surrounding system as a stable backdrop. The PJM evidence suggests that the system’s maintenance calendar itself now belongs in the risk register.

A project can have a credible interconnection path and still face a difficult energisation environment if transmission works, generator maintenance and large-load growth converge. This is why a grid offer should not be treated as deliverable capacity. Our earlier framework on when a grid connection becomes deliverable separates queue position, network works, energisation conditions and operational availability.

The same distinction matters after connection. A campus may be electrically connected without having all phases commissioned. It may have facility power without the cooling, controls and tenant fit-out required to support critical IT load. From grid MW to live IT capacity sets out that evidence ladder.

The practical conclusion is simple: seasonal system headroom can no longer be assumed. It needs to be evidenced.

Connected load is not the same as usable capacity

PJM’s approximately 11,000 MW figure is a measure of connected data-center load across its footprint. It is not a site inventory, a leasing metric or a claim about spare capacity. Analysts should resist converting it into any of those categories.

PowerlandMap separates at least five concepts:

1. generation capacity available to the grid; 2. utility or transmission connection capacity; 3. facility electrical capacity behind the meter; 4. critical IT load designed or commissioned inside the facility; 5. actual operating demand at a given time.

The market routinely collapses these categories into a single megawatt headline. That creates false comparisons between projects and can overstate both supply and demand. A power deal is not IT capacity explains why contracted generation should never be copied into a data-center supply total.

The same discipline applies to announcements. On 15 October, Mitsubishi Heavy Industries and Ingeteam announced a U.S. partnership covering power, cooling, energy management, power electronics and controls for AI data centers. It is a relevant supplier signal, but it is still a memorandum of understanding. It does not prove a named deployment, an order, a construction start or commissioned megawatts.

Backup generation is becoming a system instrument

PJM said a U.S. Department of Energy order could have allowed the operator to direct backup generation at data centers and other large loads during emergency conditions. PJM did not use that authority in the cited event.

That distinction matters. The existence of emergency authority is evidence of system stress and an operational option. It is not evidence that backup generation was dispatched, that a campus can operate indefinitely off grid, or that local environmental permits can be ignored.

For operators, backup assets now sit at the intersection of four obligations:

• facility resilience;
• grid emergency procedures;
• fuel and emissions limits;
• customer service commitments.

A serious diligence process should test the operating duration, permitted regime, maintenance state, fuel logistics, switching architecture and contractual control of those assets. Our analysis of grid-interactive data centers explains how flexible load and on-site resources can support capacity strategy without pretending that contingency assets are normal supply.

For investors, this affects valuation. A project with clear grid-operating rules, validated backup capability and credible phased energisation deserves different treatment from one that simply lists generators in a technical schedule. Power-backed AI infrastructure finance shows why secured and phased power is increasingly part of the collateral story.

Power constraints do not replace other constraints

As grid scrutiny rises, it is easy to let electricity dominate the entire decision. That is a mistake.

On 15 October, the Texas attorney general said letters had been sent to 156 data-center developments seeking accurate water-use information. The announcement does not identify the individual projects or establish that any operator violated the law. It does show that resource reporting can become a material development issue at portfolio scale.

Water withdrawals, consumption, cooling design and heat rejection should therefore be checked alongside power evidence. Liquid cooling and data-center water use provides a framework for avoiding another common shortcut: assuming that a cooling technology label determines site-level water impact.

Efficiency labels also need context. A design PUE, a rating certificate and an operating measurement answer different questions. Efficiency ratings do not replace asset diligence explains how those indicators should be combined with delivery, commissioning and commercial evidence.

What developers should document now

The disappearing shoulder season calls for a stronger evidence pack. At minimum, a project should document:

• the exact capacity basis for every megawatt figure;
• the utility and transmission milestones behind each delivery phase;
• seasonal outage and maintenance assumptions;
• the status of substations, feeders and network reinforcement;
• operating limits for backup generation and demand response;
• cooling and water requirements by phase;
• dependencies between energisation, commissioning and tenant readiness;
• the next dated evidence that can move the project to a higher-confidence status.

This is also a portfolio-management problem. Multi-site developers need a common vocabulary so that a grid offer in one market is not compared with commissioned IT load in another. Scale changes the operating model describes the interfaces required to keep capacity claims accountable across markets.

Recent expansion signals reinforce the need for that discipline. EdgeConneX announced an Australian market entry with Stockland and planned campuses in Sydney and Melbourne. The announcement supports a planned-market-entry classification. It does not, on its own, prove construction, grid connection or live capacity.

A better benchmark for the next cycle

The most useful question is no longer “How many megawatts are announced?” It is “Which megawatts can be delivered, operated and expanded through the actual constraints of the system?”

That benchmark requires dated evidence, a clear capacity basis and explicit unknowns. It should be refreshed whenever a utility milestone, regulatory action, commissioning event or operating constraint changes.

PowerlandMap’s approach is to preserve those distinctions rather than smooth them into one market number. Why power benchmarks now define data-center deals applies the method to transactions, while a market study should change a decision explains how the evidence becomes actionable for site selection, expansion and investment.

The disappearance of shoulder-season slack does not mean every project is blocked. It means the background conditions can no longer be treated as background. Grid operations have become part of data-center asset diligence.