The Grid Is Drawing a Line: PJM Moves to Rein In the AI Power Boom – Miner Weekly

This article first appeared in Miner Weekly, a weekly newsletter by BlocksBridge Consulting curating the latest news in energy, compute, infrastructure, and data analysis from TheEnergyMag. Subscribe to receive in your inbox once a week.

The most important power market for U.S. data centers is rewriting the rules.

PJM Interconnection — the grid operator spanning from Illinois to Virginia — sits at the epicenter of the global AI infrastructure buildout. The region hosts Northern Virginia, the world’s largest data center hub, and already accounts for roughly 40% of U.S. data center capacity.

Late last month, PJM filed a proposal with federal regulators to overhaul how large electricity users — particularly data centers colocated with power generation — connect to and pay for the grid. The filing was made in response to a December directive from the Federal Energy Regulatory Commission (FERC), which concluded that PJM’s existing framework no longer reflects the realities of large-scale AI-driven demand.

The timing is not accidental. The proposal comes as the White House recently emphasized a “Ratepayer Protection” stance, signaling that households and traditional businesses should not bear the cost of infrastructure built to serve energy-intensive AI and data center growth. That message closely aligns with regulators’ concerns that some large users are benefiting from the grid without paying a proportional share.

The initial public comment period on that proposal closed on March 16, drawing formal objections from industrial power users and setting the stage for a regulatory decision in the months ahead.

PJM’s filing is more than a procedural update. It represents a structural shift in how the grid treats the fastest-growing source of electricity demand in the U.S.

At stake is not just a tariff adjustment, but the economics of where — and how — the next generation of AI compute gets built.

To understand why this matters, it helps to simplify how things used to work.

From “Self-Powered” to Grid-Dependent

For years, PJM allowed something called “behind-the-meter” generation.

In simple terms, this means a company produces electricity right next to where it uses it — for example, a factory with its own gas turbine, or more recently, a data center built next to a power plant.

Under the old rules, these users could “net” their electricity usage. Think of it like this:

• If a data center uses 100 MW of power
• And produces 70 MW on-site
• The grid only counts it as using 30 MW

That accounting matters because many of the biggest grid charges are based on how much demand you place on the system.

In PJM, large customers typically pay for several things tied to their load:

• Transmission costs (paying for high-voltage lines that deliver power across the region)
• Capacity costs (paying for the grid to have enough generation available during peak demand)
• Certain system and reliability charges

If your load is counted as 30 instead of 100, those charges are calculated on the smaller number.

That made sense when onsite generation truly reduced dependence on the system.

But the problem is that large facilities — especially data centers — still rely heavily on the grid in ways that aren’t captured by that simple math:

• They need backup power when onsite generation fails
• They rely on the grid for stability and frequency control
• The grid must still be sized to serve them at full demand

In other words, even if they “self-generate” most of their power, the grid still has to be there — fully built — just in case.

That’s where regulators see a mismatch.

The 50 MW Line

PJM’s proposal introduces a clear dividing line:

• Below 50 MW → treated like traditional, smaller onsite users
• Above 50 MW → treated as large grid users, regardless of onsite generation

In practical terms, this means large data centers can no longer use accounting to significantly reduce their grid costs.

Instead, they must choose:

• Pay for firm access to the grid (guaranteed power, higher cost)
• Or take non-firm service (cheaper, but can be curtailed)

The key shift is conceptual: You are no longer “off-grid” just because you have power next door.

Why This Is Happening Now

The rule shift only makes sense when you look at what PJM is seeing internally.

The operator is facing a surge in demand unlike anything in its history — driven almost entirely by data centers.

• PJM estimates up to ~30 GW of new data center demand by 2030
• It already represents ~40% of U.S. data center capacity

Even more striking:

• Total proposed large-load requests can exceed 100 GW before filtering

To put that in perspective, that’s comparable to adding multiple large states’ worth of electricity demand onto the grid.

PJM has had to heavily discount speculative projects and apply conservative assumptions just to make the numbers usable.

Still, the conclusion is unavoidable: this is not incremental growth — it’s a structural demand shock.

The Hidden Constraint: The Grid Can’t Keep Up

Another key issue is timing.

• Data centers can be built in 1–2 years
• Grid infrastructure often takes 3–7 years

PJM’s proposed new service options reflect this mismatch. They effectively say:

• You can start operating early
• But you may be interrupted until the grid is fully upgraded

This introduces a tradeoff the industry hasn’t had to face at scale before: speed versus certainty.

The Pushback: Not From Big Tech (Yet)

Interestingly, though, the strongest opposition so far appears to be coming from industrial users, not hyperscalers just yet.

Manufacturers and combined heat & power operators argue:

• The 50 MW threshold is too low
• The rules are being rewritten for data centers
• But applied to everyone

Their concern is that longstanding onsite generation models — which genuinely reduce grid demand — could be penalized alongside large AI facilities.

Why This Matters for Bitcoin Miners

At its core, this is about cost allocation.

FERC’s position is increasingly clear:

If the grid must be built to serve you, you should pay for it — even if you don’t use it all the time.

That challenges a key strategy used by data centers: colocating with power to minimize grid costs while still relying on it as a safety net.

For Bitcoin miners, this dynamic is familiar because mining has long operated on a different model:

• Flexible demand
• Willingness to curtail
• Optimization around power pricing

In many ways, PJM’s framework formalizes that distinction:

• Firm power → expensive but reliable
• Flexible power → cheaper but interruptible

The difference is that AI workloads are far less tolerant of disruption. You can shut down ASIC miners any time – bitcoin doesn’t care. But you can’t turn off AI compute hardware without jeopardizing long-term service revenues from customers.

The Bottom Line

PJM is not just tweaking rules — it is redefining what it means to be a large power user and implementing a regulator-driven reset of how large loads interact with the grid.

The old idea was:

“If I generate my own power, I pay less for the grid.”

The new reality is:

“If the grid needs to exist for you, you pay for it.”

The first round of public comments has already been filed, and FERC is now deciding whether PJM’s proposal meets the mandate it laid out in December. Even if approved, the rule would be phased in over several years, with legacy users getting some protection and newer large-load projects facing a much stricter framework almost immediately.

That means the fight is no longer about whether the old model survives. It is now about how quickly the new one takes hold — and how expensive it becomes for the next wave of AI infrastructure to secure reliable power.

Hardware and Infrastructure News

• Nscale in Talks to Acquire 8GW West Virginia AI Data Center Site: Report
• DOE Targets AI-Driven Power Demand With $1.9B Grid Upgrade Initiative
• Bitdeer Launches Scrypt Miner as Bitcoin Mining Hardware Market Tightens
• DOE Issues Second Emergency Order to Keep Centralia Coal Plant Online
• HIVE Launches GPU Cloud Cluster in Paraguay Supporting Columbia University AI Research
• Bitcoin Difficulty Set for ~8% Drop, Exceeding 2022 Capitulation
• Fluidstack Abandons $11B France AI Data Center Projects, Pivoting to US Expansion

Corporate News

• AMC Robotics Taps HIVE GPU Cloud to Power AI Robotics Development
• HIVE Expands AI Data Centers in Canada as Sweden Bitcoin Hashrate Winds Down
• Bitcoin Miner Creditors Claim $13.6M in BlockFills Bankruptcy
• NVIDIA releases new open models to support autonomous and healthcare AI application
• Cango Reports $452.8 Million Net Loss for 2025 Amid AI Infrastructure Pivot
• Constellation agrees to divest PJM assets to LS Power amid regulatory settlement

Financial News

• Bitdeer Reaches 68 EH/s Hashrate and Completes $375M Note Offering
• Terawulf Adds $500M Bridge Loan for Kentucky AI Data Center as Debt-Fueled HPC Pivot Deepens
• Nebius Signs $27 Billion AI Infrastructure Deal with Meta
• Southern Company Prices $1.3 Billion Junior Subordinated Note Offering
• Nebius Prices Upsized $4 Billion Convertible Debt Offering

Feature

• How Eric Trump Became an Ally of One of China’s Biggest Crypto Companies