The $6.5 Billion Load That Doesn’t Exist Yet

Data centers that haven’t been built yet just cost ratepayers $6.2 billion.

That’s not a typo. PJM Interconnection’s December capacity auction — the mechanism that ensures enough power plants exist to keep the lights on across 13 states — priced at a record $329.17 per megawatt-day. Data centers accounted for 40% of the $16.4 billion total. And the majority of that cost came from facilities that are still blueprints, projected to materialize by 2027 or 2028.

The grid is now pricing in phantom load.

The ghost in the auction. Capacity markets are supposed to ensure resource adequacy — enough generation to meet peak demand three years out. They’ve always incorporated load forecasts. But PJM’s latest auction marks a structural break: speculative demand from a single industry sector has become the dominant cost driver in the largest wholesale electricity market in the world.

The math is straightforward. PJM covers 65 million customers across a territory stretching from New Jersey to Illinois. When capacity prices jump from $28.92/MW-day (the 2025/2026 result) to $329.17/MW-day, every ratepayer absorbs the increase. Commercial buildings in Philadelphia, manufacturing plants in Ohio, hospitals in Virginia — all paying more because hyperscalers filed interconnection requests for facilities that may or may not get built.

This isn’t an argument against data centers. It’s an observation about what happens when capacity markets collide with exponential demand growth from a sector that moves faster than grid infrastructure.

The interconnection paradox. Here’s where it gets interesting. These data centers need power before the grid can deliver it. Average interconnection timelines in PJM now exceed four years. Transmission upgrades take longer. The capacity auction says the power must be available by 2026/2027, but the physical infrastructure to serve these loads won’t exist on schedule.

This creates an arbitrage that battery storage fills by default.

Aligned Data Centers proved the model last month — a 31 MW/62 MWh battery system alongside a Pacific Northwest facility, operational in 2026. The battery doesn’t replace grid power. It bridges the gap between when the data center needs to operate and when the grid can fully serve it. It shaves demand charges. It provides backup during the inevitable transmission constraints.

Industry analysts now project a 20 GW behind-the-meter battery opportunity from hyperscalers alone through 2035, with 9 GW materializing by 2030. At least 25—33% of incremental data center demand is expected to be met by on-site storage. Not because batteries are cheap. Because waiting for the grid is more expensive.

The ratepayer revolt nobody sees coming. Consumer advocates are already sounding alarms. When a single capacity auction increases costs by an order of magnitude, regulators face pressure. The political dynamics are predictable: residential customers and small businesses will resist subsidizing data center buildouts through their electricity bills.

This pressure creates a structural incentive for behind-the-meter solutions. Every megawatt of load that a data center can serve on-site — through batteries, on-site generation, or demand flexibility — is a megawatt that doesn’t inflate the capacity auction. Regulators in PJM states are already examining whether data centers should face different cost allocation mechanisms.

But the implications extend far beyond data centers. Commercial and industrial customers across PJM territory are staring at capacity cost increases that will flow through to retail rates within 18 months. Buildings that can reduce their grid dependence during peak periods will see material savings. Buildings that can’t will absorb costs driven by an industry they have no connection to.

The ERCOT warning. Texas offers a preview of what happens when storage deployment accelerates too fast in response to market signals. ERCOT battery interconnection applications dropped 50% in the second half of 2025. Only 85% of projects with signed agreements are likely to reach completion. Queue timelines now stretch beyond four years.

The ERCOT slowdown isn’t a failure of storage economics. It’s a market finding its clearing price after a gold rush. PJM is entering its own gold rush phase now, with capacity prices 11 times higher than two years ago. Storage developers will flood the zone. Some will overbuild. The market will correct.

But the underlying driver — data center load growth — isn’t cyclical. Microsoft, Google, Amazon, and Meta have committed hundreds of billions to AI infrastructure. The load is coming. The question is whether it arrives on the grid’s timeline or the hyperscalers’ timeline.

The new grid topology. What’s emerging is a power system where the largest loads are partially self-served. Data centers with on-site batteries and generation. Commercial buildings with storage systems that arbitrage between peak and off-peak. Industrial facilities that can island during grid stress events.

This isn’t the distributed energy future that rooftop solar advocates envisioned. It’s messier and more pragmatic. It’s driven not by environmental mandates but by the simple fact that centralized grid expansion can’t keep pace with concentrated load growth.

PJM’s capacity auction made the economics undeniable. When phantom load from unbuilt facilities can drive $6.5 billion in costs across a 13-state grid, the value proposition for on-site storage stops being theoretical.

The grid was designed for an era when demand grew at 1—2% annually and new power plants took a decade to build. Data centers are growing demand at 15—20% annually and need power in 18 months. Something has to give.

It already did. It cost $329.17 per megawatt-day.