$136.53. That’s the number that changed the conversation.
Wholesale power pricing data from Monitoring Analytics, the independent market monitor for PJM Interconnection, the grid serving 65 million people across 13 states, is T1 authority for what power actually costs in the most data-center-dense grid in the United States. When that data shows a 76% year-over-year increase in Q1 2026, it’s not a projection. It’s a pricing signal from the infrastructure layer.
The mechanism is concentration. Data centers don’t distribute power demand evenly across a grid, they cluster geographically, near fiber routes, tax incentives, and existing power infrastructure. Northern Virginia alone hosts roughly 70% of the world’s internet traffic across its data center corridor. When AI infrastructure investment accelerates that clustering, localized power demand spikes faster than generation capacity can respond. The PJM Q1 2026 pricing data is what that spike looks like in market-clearing prices.
The NERC Alert in Context
NERC, the North American Electric Reliability Corporation, issued a reliability alert focused on computational load growth from AI data center demand. Reliability alerts are not routine communications. NERC issues them when grid conditions warrant formal operator notification. The specific classification level reported in some outlets has not been confirmed against NERC’s formal designation system; this analysis uses “reliability alert” only.
What matters operationally is what reliability alerts require. Grid operators in affected regions must assess whether peak load scenarios include AI data center demand in their planning reserves. Load-serving entities, the utilities that power data centers, face new reserve margin requirements. For data center operators, a NERC reliability alert translates directly into questions about power availability commitments and whether long-term power purchase agreements signed before the alert remain enforceable at the contracted delivery volumes.
The formal NERC publication will specify requirements. That document, not news coverage of it, is what data center infrastructure teams need. The alert’s existence is confirmed through credible sources. The details require direct verification from NERC’s official publications.
The Demand Projections: What EPRI and S&P Global Are Saying
Two T1 analytical authorities have published 2030 demand projections that frame where this goes.
EPRI’s 2026 analysis projects AI data centers consuming between 9% and 17% of U.S. electricity by 2030. That range, eight percentage points wide, reflects genuine uncertainty in AI infrastructure buildout pacing. The low end (9%) assumes roughly current trajectory. The high end (17%) assumes accelerated hyperscaler buildout and broader enterprise AI infrastructure adoption. Both endpoints are plausible given current capital deployment signals.
S&P Global projects AI data center demand reaching 134.4 GW by 2030. To calibrate: the U.S. total generating capacity is approximately 1,150 GW. A 134.4 GW demand figure from AI data centers alone would represent roughly 12% of current total U.S. generating capacity, consumed by a single end-use category that barely existed a decade ago.
These are projections. They carry the uncertainty that projections always carry. What distinguishes the Q1 2026 PJM data from these estimates is that it’s backward-looking primary data, not forward-looking modeling. It confirms that demand growth is already affecting pricing, the projections argue that what’s already happening will intensify.
Three Infrastructure Responses: How the Market Is Moving
The market hasn’t waited for policy to respond. Three distinct infrastructure strategies have emerged in the current cycle’s coverage.
Oregon’s Schedule 96, covered in this hub earlier this month, represents the regulatory response: a mandatory billing tier structure that charges data center operators at higher rates as their consumption exceeds defined thresholds. It’s the first state-level attempt to price AI infrastructure demand into utility economics explicitly. If Schedule 96 survives regulatory challenge, it’s a template other states will follow, the financial implication for hyperscalers with Pacific Northwest data center density is material.
The Bloom Energy Project Jupiter fuel cell deployment, covered in late April, represents the distributed generation response: siting AI data center power generation off-grid, adjacent to the facility, using solid oxide fuel cells that bypass grid congestion entirely. The economic logic is that at $136/MWh grid pricing, on-site generation at $70-90/MWh equivalent cost becomes immediately attractive. The constraint is deployment velocity, fuel cell manufacturing capacity is orders of magnitude smaller than grid power capacity.
The Utah “Stratos Project” illustrates the scale problem without a solution. According to reporting by Quartz and Newsweek, a large-scale data center development in Utah is facing community and regulatory opposition over a reported power requirement of approximately 9 GW. That figure has not been confirmed against project filings. 9 GW from a single development would be extraordinary, it exceeds the total generating capacity of several U.S. states. Whether or not the specific figure is accurate, the opposition pattern it’s generating reflects a dynamic that will recur: communities and utilities aren’t designed to absorb hyperscale AI infrastructure demand at the pace the investment is coming.
Who Pays the AI Energy Premium
The $136.53/MWh figure sits at the wholesale layer. How that cost reaches end-users moves through several intermediaries, and the pass-through dynamics matter for enterprise AI buyers.
Hyperscalers operate their own data centers under long-term power purchase agreements. If those PPAs were structured at $60-80/MWh for multi-year terms, the Q1 2026 spot pricing doesn’t immediately affect them, but it affects the PPAs they’re negotiating now. Next-generation hyperscaler data center capacity being brought online in 2026-2027 is being priced at current wholesale conditions, not 2023 conditions.
Colocation providers, the data centers that rent capacity to AI companies and enterprises without hyperscaler infrastructure, face more immediate exposure. Their power contracts have shorter terms and less favorable pricing than hyperscaler PPAs. Rising energy costs flow into colocation pricing within 12-18 months of spot market moves.
Enterprise AI buyers consuming through cloud APIs sit furthest from the direct power pricing. Their exposure is indirect: hyperscalers that face structurally higher energy costs across a four-to-six-year infrastructure investment cycle will need to recover those costs somewhere. Watch compute pricing in enterprise AI contracts for the signal, not the energy cost reports.
The Compound Risk: Grid Reliability Plus Demand Growth
The NERC alert and the pricing data together create a risk profile that goes beyond cost. Grid reliability constraints at the same time as demand acceleration means the power availability assumptions built into AI infrastructure plans from 2023-2024 may not hold through 2027-2030.
Data center operators that committed to hyperscale expansion based on available power projections that predate the current demand acceleration cycle need to reassess whether the utility infrastructure can deliver. This is the risk that’s harder to hedge than pricing: you can lock in a power purchase agreement, but you can’t lock in grid capacity that hasn’t been built.
Watch the NERC alert documentation. Watch utility interconnection queues for evidence of bottlenecks. Watch state-level energy policy responses to Schedule 96’s structure. The Q1 2026 PJM data is the earliest primary-source confirmation that AI data center demand is measurably moving major grid pricing. The infrastructure and policy responses to that data will define AI compute costs for the rest of the decade.
The real story is the asymmetry. Demand is growing faster than grid capacity can respond, and the cost of that gap will flow through the infrastructure stack in ways that aren’t yet fully priced into enterprise AI planning models. The $136.53/MWh number is where that asymmetry became measurable. Watch what comes next.