Are AI Data Centers Driving Up Electric Bills?
What the PJM vs. ERCOT Divide Means for Power Infrastructure Strategy
In recent months, headlines have suggested that AI data centers are driving residential electricity bills sharply higher — particularly in PJM markets across the Mid-Atlantic and Midwest. In New Jersey, for example, residential rates jumped nearly 20% in mid-2025, and some observers were quick to point to AI facilities as the culprit.
But a deeper analysis tells a different story.
According to a recent joint report by SemiAnalysis and Archer Daniels Midland Investor Services (ADMIS), the increase in electricity bills in PJM appears to be driven less by realized AI data center load and more by the design of PJM’s forward capacity market — specifically the Base Residual Auction (BRA).¹
Understanding that distinction matters for data center developers, utilities, hyperscalers, and infrastructure investors alike.
Energy vs. Capacity: The Structural Difference
Electric bills in deregulated markets typically include:
- Energy charges (real-time supply/demand pricing, $/MWh)
- Capacity charges (payments to ensure standby generation is available during peak events)
- Transmission & Distribution charges
In PJM, capacity is procured through a forward auction mechanism that clears years ahead of actual delivery. In the 2025/26 auction, clearing prices increased 9.3x year-over-year — from $29/MW-day to $270/MW-day, translating into approximately $16 billion in total capacity payments.¹
That spike flows directly into retail electric bills.
By contrast, ERCOT (Texas) operates an energy-only market without a forward capacity auction. Scarcity pricing during real-time grid stress incentivizes generation investment, but there is no centralized capacity pricing mechanism that can amplify forecast assumptions into multi-billion-dollar cost shocks.
The result?
- PJM residential bills projected to increase ~15% in 2026 vs. pre-AI era
- ERCOT forward power prices up only ~11–17%, despite comparable AI load growth¹
Same AI buildout. Very different price outcomes.
The Role of Forecasting — and Its Risks
In PJM, the capacity clearing price is driven by a simulated demand curve (the Variable Resource Requirement or VRR curve) based on load forecasts produced by the grid operator.
SemiAnalysis highlights that relatively small forecast deviations — particularly around data center load growth — can materially change clearing prices. The Independent Market Monitor (IMM) estimated that removing projected incremental data center load would have reduced total capacity payments by billions of dollars in recent auctions.¹
At the same time, PJM has revised its own data center load forecasts downward in successive years, suggesting uncertainty around construction timelines, GPU supply constraints, and interconnection delays.¹
The implication is not that AI demand is weak. It is that forward market structures amplify uncertainty.
ERCOT, meanwhile, applies significant “haircuts” to speculative large-load interconnection requests and does not embed its forecast directly into pricing mechanisms.²
Different governance models. Different cost volatility profiles.
Winter Storm Fern: A Stress Test
The divergence became even more visible during Winter Storm Fern (January 2026).
- ERCOT maintained grid stability, with real-time prices peaking around $300/MWh.²
- PJM lost approximately 21 GW of generation capacity during extreme weather, with system prices averaging $700/MWh and localized spikes much higher.¹
Despite record-high capacity payments, PJM experienced operational failures largely tied to winterization and fuel delivery issues.
One underappreciated insight from the event: the U.S. Department of Energy identified roughly 35 GW of backup generation at data centers and industrial sites that could be leveraged as emergency resources.¹
This highlights an important shift – Data centers are not just large loads — they are potential grid assets.
What This Means for Data Center Developers
The debate is no longer simply about megawatts.
It is about market design risk, regulatory agility, forward pricing volatility, interconnection timing, onsite generation strategy, and grid-interactive resiliency.
For developers operating in PJM, capacity pricing volatility can materially affect long-term operating costs and political optics. For developers in ERCOT, scarcity pricing and regulatory flexibility create a different risk profile.
In both cases, the need for robust, flexible, and strategically designed power infrastructure has never been clearer.
Onsite Power Is No Longer Just Backup
As AI workloads scale and regional market structures diverge, onsite power solutions are evolving from redundancy assets to strategic infrastructure components.
Integrated solutions such as gas-fired peaking systems, multi-fuel capable generators, battery energy storage, grid-interactive backup capacity, and fuel cell systems can materially strengthen a facility’s power strategy. When properly designed and integrated, these systems enhance operational resilience, reduce exposure to capacity pricing volatility, accelerate energization timelines, and provide greater optionality in constrained grid environments. In addition, they can create opportunities to participate in demand response programs or other grid services, turning backup infrastructure into a strategic asset rather than a passive safeguard. In short, the power strategy itself is becoming a competitive differentiator.
The Takeaway
AI data centers are not the sole driver of rising electric bills. Market structure, regulatory design, and forecasting methodology are playing an outsized role in price formation — particularly in PJM.
For data center operators and infrastructure investors, that means power planning can no longer be treated as a commodity input. It must be treated as a strategic variable.
Jurisdictional differences in grid governance will directly influence development risk, long-term cost predictability, community acceptance, capital structuring, and overall project timelines. These structural factors increasingly shape not only where projects get built, but how they are financed, permitted, and delivered.
Power infrastructure decisions made today will determine operational flexibility tomorrow.
How Network Environments Can Help
Network Environments partners with data center developers, hyperscalers, and mission-critical operators to design and deploy integrated power infrastructure solutions built for today’s market realities – not yesterday’s assumptions.
If you are evaluating new markets, expanding capacity, or assessing onsite generation strategies in light of evolving grid dynamics, our team can help you structure resilient, scalable power systems aligned to your development objectives.
Connect with Network Environments to discuss your power infrastructure strategy
Sources
- SemiAnalysis & Archer Daniels Midland Investor Services (ADMIS), Are AI Datacenters Increasing Electric Bills for American Households?
- ERCOT Long-Term Load Forecast (2025) and Biennial ORDC Report (2024); PJM Base Residual Auction Reports (2025/26).