Northern Virginia already hosts more data centers than the next five largest U.S. markets combined. Now Dominion Energy's territory faces another 4,000 MW of data center load by 2028—nearly double the 2,100 MW its substations can currently handle. The $2.3 billion question: who covers the transmission upgrades?
PJM's interconnection queue shows over 30 GW of data center requests across its footprint, with hyperscalers like Microsoft, Google, and Amazon already breaking ground on facilities. The load is coming. The grid isn't ready. And a pending FERC ruling will determine whether data center operators pay directly for the infrastructure they need—or whether 65 million ratepayers split the bill.
The Arithmetic Doesn't Work
PJM's 2024 load forecast projected 1.2% annual growth through 2030. In data center-heavy zones, the actual pipeline implies growth rates three to four times higher. The grid operator has lost more than 5.6 GW net over the past decade as plant retirements outpaced new construction, according to filings with regulators. Now it faces a demand surge it never planned for.
The mismatch shows in the interconnection queue. About 38 GW of resources have cleared PJM's queue but remain unbuilt due to external constraints—permitting delays, supply chain bottlenecks, financing gaps. Meanwhile, BloombergNEF projects U.S. data center power demand will double by 2035, with PJM bearing a disproportionate share given Northern Virginia's dominance.
PJM's response: fast-track 11.8 GW of mainly gas-fired generation through its Reliability Risk Initiative, shaving an estimated 18 months off normal interconnection timelines. The RRI attracted 94 applications—a signal of how desperately developers want to accelerate. But gas plants collide with the clean energy commitments driving hyperscaler procurement.
The Capacity Gap Closing In
The Capacity Gap Closing In. Source: Dominion Energy filings. Northern Virginia needs nearly double its current infrastructure to meet incoming demand.
The Clean Firm Paradox
Microsoft, Google, and Amazon have each announced 24/7 carbon-free energy targets. Meeting those commitments in PJM requires resources that don't exist at scale—or compete with existing buyers.
Nuclear plants generate 8,000-9,000 MW of carbon-free baseload across PJM, but most operate under PPAs expiring between 2027 and 2035. Data centers are competing with load-serving entities for limited output. Offshore wind projects totaling 6,400 MW hold interconnection agreements, but construction timelines extend to 2030 or beyond, with permitting delays pushing several projects back 18-24 months.
Battery-backed solar can firm renewable output, but the math is punishing: overbuild ratios of 3:1 or higher to achieve 90%+ availability. Advanced nuclear—NuScale, X-energy—offers theoretical solutions, but NRC licensing adds 36-48 months to construction timelines. Commercial operation remains years away.
Some developers have proposed behind-the-meter gas generation. That approach conflicts with corporate sustainability commitments and faces air permitting complications in nonattainment areas. PPL's June 2025 announcement crystallized the bind: no firm service capacity available until 2029 without $800 million in upgrades. Amazon and Meta expect facilities operational by 2027.
PJM's Interconnection Bottleneck
PJM's Interconnection Bottleneck. Source: PJM filings. 38 GW sits in limbo while 30+ GW of new data center load waits behind it.
Other Markets Offer Templates
ERCOT faced analogous load growth from cryptocurrency mining between 2021 and 2023. Its Large Flexible Load program required participants to curtail during emergencies—and participation exceeded expectations. Miners reduced consumption by 1,400 MW during the December 2022 cold snap, demonstrating that large industrial loads can function as demand response resources.
CAISO took a different approach: new large loads must demonstrate resource adequacy coverage before interconnection approval, showing contracted capacity sufficient for peak demand plus a 15% planning reserve margin. China's provincial grid companies in Inner Mongolia and Guizhou offer data centers preferential access to curtailed wind and solar below benchmark tariffs—reducing curtailment while serving load growth.
PJM has no equivalent framework. Its capacity market operates on three-year forward timelines that don't match hyperscaler deployment schedules. And as S&P Global notes, capacity prices remain their own solution only if they're high enough to incent new construction—which faces its own 36-month-plus queue delays.
Three Dockets to Watch
FERC Docket RM22-14 addresses cost allocation for network upgrades driven by large load interconnections. A ruling favoring direct assignment shifts billions to data center developers. Socialized allocation spreads costs across ratepayers who may never use a single AI model.
Where the Carbon-Free Power Is
Where the Carbon-Free Power Is. Source: PJM data. Roughly 15 GW of clean firm resources—but nuclear PPAs are expiring and offshore wind won't arrive until 2030.
PJM's capacity market filing (Docket ER25-1234) proposes accreditation changes affecting resources serving large industrial loads, potentially reducing capacity credits for units with high forced outage rates during peak periods. The change would ripple through procurement economics.
Virginia's State Corporation Commission (Case PUR-2025-00043) is examining whether transmission investments serving data center load are prudent. Intervenors argue ratepayers shouldn't subsidize infrastructure serving out-of-state technology companies. The political valence is shifting—economic development enthusiasm meeting ratepayer protection backlash.
The Tightening Window
For developers, the window is closing. Interconnection queue positions now carry waiting periods exceeding 36 months. Transmission upgrade costs escalate as queue depth increases. FERC Order 2023 implementation affects study timelines through 2026, adding procedural uncertainty to physical constraints.
Clean energy procurement requires increasingly creative structuring. Virtual PPAs provide RECs but not physical delivery. Sleeved PPAs with storage backing offer firmer products at premium pricing that erodes data center economics.
The next 18 months will set the template. If FERC assigns upgrade costs to load, hyperscalers face billions in unplanned capex. If costs socialize, ratepayer backlash could trigger state-level restrictions. Either way, the assumption that load can simply show up and the grid will accommodate—that assumption is already obsolete.
About the Author
Dr. Sayonsom Chanda
Dr. Sayonsom Chanda is an electrical engineer and senior scientist with more than a decade of experience in developing AI, ML, and other advanced computing solutions for the electric utility industry in US and India. He is also an energy policy thinker and a published author with more than 20 papers and 1 book.
