Data centers accounted for an estimated $6.3 billion of the $16.4 billion in charges from PJM Interconnection's latest capacity auction, according to Joseph Bowring, president of the grid operator's independent market monitor. The estimate assigns 38 percent of the auction's charges to that load. Monitoring Analytics has not yet published its full analysis. [1]
The number advances July 18 position 4, which said national data-center protests still required project-specific power, water, tax and permit records. It also follows July 19 position 26, where Stack cancelled its Hesse project without one documented cause. Those stories made opposition and withdrawal visible. Neither produced a regional capacity-cost estimate.
July 19 position 48 recorded Duos's five-year agreement for 10 megawatts of colocation capacity. That contract represented demand before construction, energization, use, revenue or cash. Bowring's figure works at another stage. It estimates auction charges caused by forecast data-center load rather than power already consumed by one named campus.
The distinction begins with what PJM buys. Capacity markets pay resources to be available in a future period. PJM normally procures capacity three years ahead using a demand forecast. A forecast influences today's procurement before every expected data center is built or energized. [1]
Bowring told Utility Dive that data centers drove $29.4 billion in capacity charges across PJM's last four base auctions. Total charges over those auctions were $63.6 billion. The attributed data-center share was 46 percent. [1]
Those two ratios describe different windows. The latest auction produced the 38 percent estimate. Four auctions produced 46 percent. Neither percentage is a measured share of electricity consumed at this moment.
The $6.3 billion is also not a household bill. It is an attributed portion of regional auction charges. Retail bills reach customers through utilities, tariffs, state regulation, customer classes and billing periods. A regional estimate does not disclose which household paid how much.
The forecast enters before the server
PJM spans 13 Mid-Atlantic and Midwestern states and the District of Columbia. Its auction therefore joins places with different utilities and retail rules. The same regional procurement cost can reach customers through different local arrangements. [1]
Data-center demand creates a particular forecasting problem. A large campus can request substantial power years before operation. Some projects will arrive as planned. Others will shrink, move or disappear. PJM must decide how much capacity to secure before those outcomes are known.
Buying too little risks reliability. Buying against load that never materializes risks charging customers for capacity that was not needed. Bowring argues that PJM's existing process does not sufficiently isolate that uncertainty. [1]
His estimate therefore depends on a counterfactual. Monitoring Analytics must determine what the auction would have procured or charged without data-center demand. The complete method remains unpublished. Utility Dive said the market monitor planned to issue its analysis within several weeks. [1]
That publication stage matters. An emailed number and interview can be newsworthy when the speaker runs the independent market monitor. The eventual report should expose forecast years, load assumptions, price effects and allocation. Until then, readers cannot reproduce the $6.3 billion from the article alone.
Bowring said other customers bear costs created by data centers. He included capacity, energy and transmission. Those are three different products. This article's headline concerns his capacity-auction estimate. It does not combine every grid cost into the same total. [1]
Energy charges pay for electricity produced and used. Capacity charges pay for future availability. Transmission charges pay for moving power across the network. A customer can face all three without any one number explaining the whole bill.
A proposed separate lane
Bowring supports a separate procurement system for large loads. He said data centers should first contract for their own generation. He proposed separate auctions with 15-year contracts for loads that cannot bring generation. [1]
The proposed division would remove data-center uncertainty from the base auction. Monitoring Analytics argues that this would protect other customers from paying for unneeded capacity. It would also give hyperscalers a defined route to reliable supply. [1]
That is a proposal, not the rule governing the latest auction. PJM's board was developing a backstop-auction filing for the Federal Energy Regulatory Commission. Utility Dive said the board aimed to file in July and hold an auction in September. Those were future stages at Monday's cutoff. [1]
PJM staff proposed procuring an approximately 6.8-gigawatt shortfall from the latest base auction in a one-time process. Stakeholders had supported a structure in which utilities, load-serving entities and potentially data centers would request capacity. A supported design is not a filed tariff or an approved auction. [1]
The difference determines who carries risk. A separate long-term contract can place more development risk on the large customer requesting service. A base-auction allocation can spread charges through existing load-serving arrangements. The exact result depends on tariff text that had not completed federal review.
Google, Meta, Microsoft and other data-center companies made a White House pledge in March to protect consumers from price increases caused by their energy and infrastructure needs. Bowring told Utility Dive that PJM's current rules made that pledge impossible to satisfy. [1]
The pledge is a statement of intent. It does not itself assign one auction charge. The enforceable answer must appear in contracts, tariffs, collateral, minimum payments and regulator-approved allocation.
What the $6.3 billion does not buy
The estimate does not prove that $6.3 billion flowed directly to operating data centers. Capacity charges support resources available to meet forecast demand. The causal claim concerns how added load changed procurement and price.
It does not prove that all forecast data-center load will energize. The source expressly describes uncertainty over how much will materialize. A megawatt in a forecast is not a server drawing power. [1]
It does not prove that every data center imposes the same cost. Projects differ by location, timing, interconnection, load shape, generation support and contract terms. A regional estimate cannot replace each project's docket.
It does not identify a final rate-class allocation. Industrial, commercial and residential customers enter retail tariffs differently. Utility Dive reports Bowring's claim that other customers pay. It does not provide a bill comparison by class or jurisdiction. [1]
It does not settle whether a separate auction is lawful or optimal. FERC must receive and decide an operative filing before the design becomes binding. Long contracts can shift risk. They can also preserve bad forecasts for longer if their assumptions fail.
The prior project stories help hold the sequence straight. Protest can expose public concern before a permit changes. Cancellation can close a project without identifying the fatal condition. A colocation agreement can create contractual demand before the campus operates. A capacity auction can procure against aggregate forecasts before all projects become real.
Each stage leaves a different stranded-cost question. A cancelled campus may leave planning expense. A delayed campus may leave utility infrastructure. A regional forecast may alter procurement. The person who pays depends on the instrument governing that stage.
No authorized X post was recovered through the documented search. The paper therefore cannot promote AI-boom triumph or ratepayer-bailout outrage into a verified platform frame. Utility Dive supplies an attributable institutional claim and its limits. [1]
The $6.3 billion estimate deserves attention because it converts a broad fight over AI infrastructure into a regional cost claim. Its usefulness depends on keeping the nouns precise. It is capacity-auction cost attributed to data centers by the independent market monitor.
It is not energized load. It is not a final methodology. It is not a FERC-approved allocation. It is not a named customer's bill.
The full Monitoring Analytics report can test the first claim. A tariff can decide the allocation. Compatible retail bills can show the consequence. Until those records arrive, the auction has a large number and the public still lacks an address for it.
-- DARA OSEI, London