Maryland has put a number on the data-center cost shift: about $1.6 billion.

The paper's June 16 story on FERC making PJM decide whether data centers or households pay for the grid treated the AI power fight as an unresolved who-pays filing. Maryland's complaint turns that abstraction into a docketed bill. In its PJM filing, the state argues that Maryland customers are being assigned roughly $1.6 billion in transmission costs for projects driven primarily by data-center load outside Maryland, and asks FERC to make costs follow the loads that caused them. [1]

That is the difference between anti-data-center sentiment and a receipt. X is full of the broader charge: server farms use the power, households get the bill. The article uses one broader FERC/data-center X receipt rather than a Maryland/PJM-specific status, because the local search did not produce a better verified candidate. The divergence still matters because the public argument exists; Maryland's docket supplies the state-specific proof.

Utility Dive's April account of FERC's large-load work shows why the complaint matters now. FERC was already preparing a June decision on data-center interconnection reform, with Chairman Laura Swett emphasizing the line between federal and state jurisdiction. [2] State commissions, meanwhile, argue they are best positioned to protect customers from improper cost shifts and unfair interconnection processes. [2] Maryland's filing is that theory with a dollar sign.

The state is not asking FERC to decide whether artificial-intelligence infrastructure is good or bad. It is asking whether a regional transmission formula should assign costs to Maryland customers when the load growth driving the projects sits elsewhere. [1] That is why the phrase "cost causation" matters. It turns a culture-war fight over server farms into a utility-law fight over whether the beneficiary, the grid region, or the household customer pays for the next transformer and transmission line.

Maryland's filing also changes the time horizon. Data-center debates usually ask whether future load will require new generation or slower interconnection. This complaint says the cost shift is already visible in transmission planning and should be assigned before the bill hardens into ordinary rates. [1] That is why the refund-effective date is not a procedural footnote. It preserves the possibility that customers can be made whole if FERC later agrees that PJM's allocation method sent the wrong costs to the wrong people.

The $1.6 billion figure also reframes AI infrastructure. A data center is not only a building with servers. It is a grid claim: transmission upgrades, capacity planning, backstop procurement, interconnection rules, and rate design. If the cost is spread broadly, the data center becomes a private load financed partly through public utility bills. If the cost follows causation, the load pays more of its own way. The complaint asks FERC to choose.

The mainstream frame calls this planning. That is not wrong; the grid must be planned. But planning language can hide who pays. Maryland's complaint strips away the haze by naming customers, transmission projects, data-center drivers, and a refund-effective-date request. [1]

The jurisdiction fight is the other half of the story. Utility Dive reported that FERC and state regulators were already arguing over who gets to set the terms for large-load interconnection, with states warning that federal rules can miss local ratepayer harm. [2] Maryland's filing converts that warning into a case: if PJM's regional method socializes costs that a few large loads cause, then state customer protection depends on a federal forum willing to adjust the allocation. The complaint therefore makes one state the test case for a much larger AI-power question, and it does so before the next wave of data-center requests becomes routine across the regional grid.

FERC may reject Maryland, narrow the remedy, or push the matter into a broader PJM reform. Any of those outcomes would be a governing answer. What exists today is a filed claim that the AI grid build-out is already moving money from one set of customers to another. The useful number is not just $1.6 billion. It is the docket that makes the number contestable, refundable, and traceable to specific projects instead of to a vague fear that data centers will raise everybody's bill.

-- DARA OSEI, London