PJM Moves the Meter Inside the Data Center

A data center can wait up to seven years for a grid connection, then be asked to help stabilize that same grid. PJM’s proposed choice—on-site generation or curtailed demand—moves the meter inside the facility while giving the regional operator a stake in what happens behind it.

A seven-year queue moves the power plant inside

Data-center developers once bought electricity as an external service. They chose a site, secured a utility connection and designed the facility behind the meter. Power distribution and conditioning mattered, but the project schedule assumed utility capacity would arrive before the servers did.

Developers could no longer make that assumption once US grid-access waits reached up to seven years. The same reporting described projects turning to aeroderivative turbines, diesel generators and on-site power plants to gain more control over when electricity became available.

When developers bring generation behind the fence, they turn a megawatt procurement problem into a facility-design problem. They also assume decisions about fuel, redundancy, switching and operating conditions that they previously purchased as a service.

Once a developer generates, distributes and conditions electricity on site, the power system becomes part of the computer. Fuel supply, breakers and switching logic determine whether the processors can operate at all.

Private power creates a public reliability obligation

In January 2026, PJM Interconnection unveiled a plan that would require large data centers to provide on-site generation or curtail electricity use. PJM would make each facility’s operating behavior part of its effort to prevent a large-scale outage.

When a data center switches from utility power to local generators, it changes conditions on the grid. Unannounced disconnections may contribute to cascading outages. A facility operator may see a successful failover because the servers stayed online; the grid operator sees a large block of demand disappear while generation and transmission remain arranged to serve it.

An operator focused solely on facility uptime can rationally choose an abrupt disconnection that destabilizes the regional network. As more facilities make the same choice, backup power becomes an operating condition of the grid rather than passive insurance.

The reporting on those disconnections counted more than 200 data centers across 30 square miles in Data Center Alley, consuming roughly as much electricity as Boston. With that many facilities concentrated in one region, their operators’ switching decisions carry public consequences.

Developers add private generation to escape delayed access. PJM’s plan would require them to expose some of that flexibility through curtailment or self-supply rules. The cited January report establishes the proposal, not a final rule or a measured reliability result.

Sungrow’s 130% rally outruns deployment evidence

The demand backdrop is substantial. In April 2025, the International Energy Agency projected that global data-center electricity use could more than double to about 945 terawatt-hours by 2030 and account for nearly half of US electricity-demand growth. That forecast does not establish demand for any particular supplier’s equipment.

A Q4 2025 market report linked Sungrow’s roughly 130% share-price gain in 2025 to exports and AI-infrastructure sentiment, even as US tariffs weighed on Chinese infrastructure companies. Investors were widening the AI supply-chain map beyond processors, memory and networking.

That report does not identify a Sungrow deployment in a data-center microgrid, grid-forming inverter system or battery-storage control system. The omission does not prove that no deployment exists; it means the share-price gain cannot serve as evidence of one.

The same report supports a claim that Sungrow faces tariff exposure. It does not cite an FCC action against the company and therefore cannot support a claim that such an action occurred. Sungrow’s rise documents investor enthusiasm without documenting purchases by data-center operators.

Generators, batteries, the grid and compute must move together

Once a facility combines local generation, utility supply, storage and flexible computing loads, its operators must decide which resource acts, at what moment and under whose signal. They must start generators, discharge batteries, manage grid connections and reduce computing loads without turning a facility correction into a regional disturbance.

Google signed agreements with five US utilities to reduce data-center consumption by as much as 1 gigawatt during peak periods. Google made that capacity available by making computing load controllable and exposing that control to the wider system.

PJM’s generation-or-curtailment plan would recognize two ways a facility can reduce its draw during grid stress. A facility’s controller determines whether local generation, storage, grid supply and flexible load can work together under both facility and grid constraints.

Developers evaluating on-site generation must therefore ask more than how many megawatts a turbine can produce. They must also determine whether the facility can start it, coordinate storage, manage the grid connection and curtail compute when an operator calls for relief.

Data-center autonomy remains conditional

Under PJM’s January 2026 plan, facility operators could add generation, switch to backup power or reduce consumption. PJM would still need visibility into when and how those moves occurred. Facility operators gain more choices only when the regional operator can coordinate them.

The seven-year queue moved the meter inside while leaving PJM’s reach intact. Behind it now sit a generator, a breaker and a curtailment contract; across the line, PJM still needs to know when the building disappears.