Maryland Would Freeze Data Centers Rather Than Price Electricity

Maryland Governor Wes Moore said on Thursday, September 17, 2026, that he would "absolutely sign" a statewide moratorium on new data center development if the General Assembly sent him one. He has no bill in front of him. The legislature does not reconvene until January 2027, and Maryland's 90-day session means nothing reaches his desk before April. What he does have is a general election on November 3, roughly six weeks out, and an electorate that has spent two years watching utility bills climb. The moratorium is not a policy yet. It is a position, and it tells you which way the political wind is blowing across the PJM territory.

The Thursday Remark

Moore made the comment in response to a reporter's question, which is worth holding onto: this was not a prepared rollout with a bill number and a fiscal note attached. It was an answer. But "absolutely sign" is a strong verb choice from a governor who spent 2025 pushing in the opposite direction. Maryland's Critical Infrastructure Streamlining Act, passed that year, was designed to speed permitting for exactly this class of project. The Next Generation Energy Act, from the same session, was pitched as a response to rising bills and in-state generation shortfalls. A governor who signs a moratorium in 2027 would be reversing his own 2025 legislative program inside two years.

The projects in question are real and already moving. Quantum Loophole assembled roughly 2,100 acres at the former Alcoa Eastalco aluminum smelter site in Frederick County, marketing it as a campus for gigawatt-scale tenants. Aligned Data Centers has been building there. The smelter is the tell: that land has high-voltage interconnection because it used to host an industrial load that drew hundreds of megawatts. Data center developers are not conjuring demand out of nothing. They are buying the corpses of American heavy industry for the wires attached to them.

Moore is not alone. The source reporting places him alongside other governors and elected officials pushing for a freeze, and the pattern is consistent across the Mid-Atlantic and the Midwest. The politics of this have flipped fast. In 2023, a data center announcement was a ribbon-cutting. In 2026, it is a liability.

What Maryland Actually Imports

Maryland generates well under half of the electricity it consumes. The rest comes across transmission lines from Pennsylvania, West Virginia, and Virginia, which is to say it comes from coal and gas plants that Maryland's own political class has spent a decade trying to shut down. That is the structural fact underneath every argument in this fight, and it predates ChatGPT by years.

Two coal units make the point. Brandon Shores and H.A. Wagner, both near Baltimore, were slated for retirement in 2025. PJM Interconnection determined that the grid could not reliably lose them until transmission upgrades were finished, and they were kept running under reliability-must-run arrangements. The cost of keeping unwanted coal plants alive for reliability, on the order of hundreds of millions of dollars, lands on ratepayers. Maryland households are paying today for the gap between the generation the state retired and the transmission it did not build.

Data centers did not create that gap. They walked into it and made it visible. A 300 MW campus in Frederick County is a rounding error against national consumption and a serious problem against a state that imports 40 percent of its power. The moratorium instinct treats the new load as the cause. It is closer to a stress test.

The PJM Number That Explains Everything

PJM runs a capacity market, an auction where generators are paid to promise availability years in advance. For the 2024/2025 delivery year, capacity cleared at $28.92 per megawatt-day. For 2025/2026, it cleared at $269.92 per megawatt-day, taking the total cost from roughly $2.2 billion to roughly $14.7 billion in a single auction. The next auction, for 2026/2027, cleared at $329.17 per megawatt-day, effectively at the administrative price cap, for about $16.1 billion.

That is a price signal doing exactly what a price signal is supposed to do. Capacity got scarce, the price went vertical, and the message to anyone who can build a generator is: build one, immediately, we will pay you. The problem is that the signal arrives at consumers through a regulated retail rate that they cannot escape, cannot hedge, and did not agree to. A Baltimore household on a standard offer service tariff sees a bill increase and has no mechanism to respond except to vote.

So they vote. And the politician's available lever is not "let the price work," because the price is already working and it hurts. The lever is quantity restriction. A moratorium caps demand by decree because the retail market structure forbids demand from responding to price on its own.

Nationally, the scale is not in dispute. Lawrence Berkeley National Laboratory put US data center consumption at roughly 176 terawatt-hours in 2023, about 4.4 percent of national electricity, and projected a range of 325 to 580 TWh by 2028, or somewhere between 6.7 and 12 percent. Hyperscaler capital expenditure from Microsoft, Amazon, Alphabet, and Meta has run in the hundreds of billions of dollars annually. Meta's Hyperion campus in Louisiana was announced as a multi-gigawatt, multi-billion-dollar build. The Stargate program associated with OpenAI and Oracle was announced at $500 billion. These are not marginal projects.

Industry's Case and Its Weak Spot

The developers make a serious argument. Data centers sign long-term contracts, often 15 years or more, at high load factors. A campus running at 90 percent utilization around the clock is the best customer a utility has ever had, because it spreads fixed transmission and distribution costs across enormous volume. In principle, a large new industrial load should lower everyone else's rates per kilowatt-hour, not raise them. Northern Virginia's Loudoun County built an entire tax base on this logic, and the data center property tax revenue there runs into the high hundreds of millions of dollars annually, funding schools without touching residential rates.

The weak spot is cost allocation. If a data center triggers a transmission upgrade or a new peaker plant, and that cost enters the utility's rate base, every customer pays for it through the general rate, not just the customer who caused it. That is the actual grievance, and it is a legitimate one. The industry answer is large-load tariffs: dedicated rate classes with minimum take-or-pay commitments, exit fees, and collateral requirements so the hyperscaler eats its own capacity cost. Several utilities have filed versions of these. They are the correct fix.

A moratorium is not that fix. It is a blunt instrument that stops the good projects and the bad ones with equal force, and it will not lower a single Maryland bill in 2027, because the Brandon Shores problem and the capacity auction outcome are already baked in. The rate increases Marylanders are angry about were caused by retirements and transmission underbuild, not by a Frederick County campus that is not energized yet. Freezing new construction treats a symptom that has not appeared and leaves the disease.

There is a second cost, harder to see. The states that freeze will export the load, not eliminate it. Texas, under ERCOT, has a competitive retail market and an interconnection queue that moves in months rather than years. Louisiana, Wyoming, and the Dakotas are actively recruiting. Maryland can stop a building in Frederick. It cannot stop the compute. It can only decide whether the tax base sits inside its borders or outside them.

The Flexible Load Bitcoin Already Built

Here is where the debate gets sloppy, and where the Bitcoin industry has a claim nobody is making loudly enough. Bitcoin mining and AI inference are both called "data centers" in these bills, and they are opposite kinds of grid citizen.

An AI training cluster is firm load. You cannot pause a training run for four hours during a heat wave without destroying checkpointed work worth millions. It demands 99.99 percent uptime, which means the grid must build capacity for the worst hour of the worst day.

A Bitcoin miner is interruptible by design. Hashrate is fungible across time. A miner that shuts off during a peak event loses revenue and nothing else. ERCOT formalized this through its Controllable Load Resource and large flexible load programs, and mining operations in Texas, including Riot Platforms' Rockdale facility at roughly 700 megawatts of capacity, have curtailed heavily during summer peaks and been paid for it. In August 2023, Riot disclosed more in power and demand response credits than in Bitcoin revenue for the month. That is a load acting as a shock absorber, buying capacity that would otherwise go unbuilt and then handing it back when the grid needs it.

Lumping the two together in a moratorium is a category error with real consequences. If Maryland wants to cap firm load, cap firm load. Write the bill in terms of curtailability, not in terms of what the servers compute. A statute that says "any facility over 100 MW must accept interruption for X hours per year or pay a firm-service premium" would solve the reliability problem, protect ratepayers, and let flexible operators in the door. That bill is harder to write and less satisfying to announce. It is also the one that works.

Energy, Money, and the Malinvestment Question

Take the position plainly: the AI buildout is being financed by monetary conditions, not by savings, and a moratorium is the political system reacting to a symptom of its own currency policy. Hundreds of billions of dollars in annual capital expenditure on depreciating silicon, funded increasingly by corporate debt issuance, on revenue projections nobody can underwrite, is the textbook profile of a credit-driven boom. Cheap money lengthens the production structure and pulls capital into projects that only pencil out at artificially low discount rates. When rates normalize or the revenue fails to materialize, the buildings remain and the assumptions do not.

Bitcoin sits in a different position in this cycle, and the distinction matters. Mining capex is disciplined by a fixed issuance schedule that nobody can lobby. The subsidy halves every four years regardless of who is in office, the difficulty adjusts every 2,016 blocks, and unprofitable hashrate switches off within days. There is no central bank to backstop a miner and no appropriations committee to bail one out. That is what hard money does to an industry: it produces continuous, brutal, honest liquidation instead of a decade of accumulated distortion followed by a crisis. The AI sector has no such mechanism, which is precisely why its energy demand curve looks like a straight line up and to the right in every projection deck.

The deeper point for anyone who cares about monetary sovereignty: when the state controls the money, every capital allocation decision eventually becomes a political decision. Electricity is the base commodity of the digital economy, and once the government is in the business of rationing who may buy it, it is in the business of deciding which computations are legitimate. Today it is a moratorium aimed at hyperscalers. The same statutory authority reaches a mining operation, a self-hosted node, or a private cluster somebody built for reasons the state does not like. Price rationing is impersonal. Quantity rationing requires a list, and somebody has to write it.

What to Watch

The bill text in January 2027. Watch whether the Maryland General Assembly's version distinguishes flexible load from firm load, and whether it carves out or includes cryptocurrency mining explicitly. Most state drafts to date have used a raw megawatt threshold with no curtailment provision. If Maryland writes a curtailability standard instead of a flat cap, that is a genuinely new template and other PJM states will copy it. Expect a threshold somewhere between 50 and 100 MW.

Whether "absolutely sign" survives November 3. Campaign positions and signed legislation diverge routinely. If Moore wins and the moratorium bill arrives in April 2027 with a fiscal note showing eight or nine figures of forgone property and income tax revenue in Frederick County, watch for the language to soften into a study commission or a two-year pause with exemptions. That is the most probable outcome.

Large-load tariff filings at the Maryland Public Service Commission. These are the real fight and they get no coverage. If BGE or Potomac Edison files a dedicated data center rate class with minimum demand charges and exit fees, and the PSC approves it, the moratorium loses most of its rationale. Watch the docket, not the podium.

PJM's next capacity auction result. If the clearing price stays pinned near the cap, the political pressure for quantity restrictions intensifies across all 13 PJM states plus DC. If new generation interconnects and the price breaks downward, the moratorium movement loses its fuel within a year.

Migration to competitive markets. Track announced project megawatts in ERCOT, Louisiana, and the Southeast against the Mid-Atlantic over the next four quarters. The load does not disappear. It relocates to jurisdictions where price, not permission, determines who gets electricity. Maryland is about to run that experiment on itself.

---

Go deeper: How Bitcoin Mining Works

Source: The Hill