The Power Grid Is the New Chip Shortage — and It's Going to Be Worse

The Power Grid Is the New Chip Shortage — and It's Going to Be Worse

Let me tell you something that's been bothering me the last few weeks as I've been watching what's happening with the power grid in the United States.

For the last two years, every conversation I've had with fellow hosting providers has been about the same thing: chip shortages, GPU lead times, memory prices. But something shifted in the last month. The conversation has changed. Now it's about power. And let me tell you, the power problem is scarier than the chip problem ever was, because there's no TSMC equivalent for a power plant.

You can't outsource your way around a grid that can't keep the lights on.

The Grid Is Breaking — and Data Centers Are the Straw

Valley Forge, Pennsylvania — July 18, 2026 — The numbers out of PJM Interconnection this week are genuinely alarming. PJM — the grid operator covering 13 states and 67 million people from the Mid-Atlantic to the Midwest — just ran its capacity auction for the 2028-2029 delivery year and came up 6,831 megawatts short of its reliability target. That's the equivalent of almost seven nuclear reactors of missing capacity. And this isn't a one-off. It's the third consecutive year PJM has failed to secure enough supply.

PJM CEO David Mills said in a statement that "these auction results show that demand for electricity continues to grow faster than electricity supply." And the PJM independent market monitor found that power prices jumped 76 percent in the first quarter of 2026 alone, driven almost entirely by data center demand. PJM procured 149,182 MW of supply — still 6.8 GW below what's needed to guarantee reliability during peak demand events.

Let that sink in. The largest power grid in the country cannot secure enough electricity to reliably serve the people and businesses in its territory. And we're supposed to be deploying AI infrastructure at scale?

The Capacity Price Explosion Is Already Hitting Your Bottom Line

Here's where it gets concrete for anyone running servers. The price of capacity in PJM's auction hit the $325 per megawatt-day cap — the maximum allowed by FERC. To put that in perspective, in the 2024-2025 delivery year, capacity prices were $28.92 per megawatt-day. That's an increase of over 1,000 percent in two years. You read that right. Not ten percent. Not fifty percent. One thousand percent.

According to the Citizens Utility Board, electricity demand from new and proposed data centers is the primary reason for the spike. And the total additional utility costs attributed to data center load since 2024? Twenty-nine billion dollars. That's $29 billion in costs that someone has to pay. And that someone is ratepayers — which includes every business that rents servers, runs colo, or hosts anything in a facility within PJM territory.

If you're an independent hosting provider running servers anywhere from Virginia to Illinois, your electricity costs are going up. Not might go up. They're going up. And the capacity auction results tell us they're going to keep going up for at least three more years, because the supply isn't coming online fast enough.

New York Just Hit the Brakes — and Other States Will Follow

On July 14, New York Governor Kathy Hochul signed an executive order imposing a one-year moratorium on new large data centers — defined as facilities using more than 20 megawatts of power. It's the first statewide data center moratorium in the country, and it signals something important: the backlash against the AI infrastructure buildout has reached the statehouse level.

The New York State Legislature had already passed a moratorium bill in the final days of its session. Hochul's executive order makes it effective immediately. She's framing it as a pause to give lawmakers time to develop rules protecting the power grid, water supplies, and consumers. But let me tell you, when a state like New York — one of the biggest potential data center markets in the country — slams the brakes, every other state legislature is watching.

In the first three months of 2026 alone, grassroots groups blocked or delayed at least 75 data center projects worth a combined $130 billion, according to Data Center Watch. That's the most ever recorded in a single quarter. These aren't small projects getting stopped — these are billion-dollar campuses. And the reason isn't just NIMBYism. Communities are genuinely scared about what happens to their power bills and water supplies when a hyperscale data center moves in next door.

Virginia already has nearly 650 data centers, with Dominion Energy reporting a 5 to 6 percent annual increase in electricity demand driven entirely by data centers. Dominion originally projected residential ratepayer bills would increase by $2.90 a month from transmission upgrades — they later lowered that to $0.94 after pushback. But longer-term projections from Dominion show residential bills increasing by $255 per month by 2035. Two hundred and fifty-five dollars. Every month. And that's the utility's own projection, not some activist estimate.

The NextEra-Dominion Merger — $67 Billion Bet or $67 Billion Bailout?

This brings me to the other story that broke this month — the NextEra Energy acquisition of Dominion Energy in an all-stock deal valued at roughly $67 billion. The merger creates the world's largest utility company, and the framing from Wall Street could not be clearer: this is a bet on AI data center power demand.

"NextEra isn't buying Dominion Energy. It's buying Data Center Alley," Raj Brar, founder of Argus AI Labs, wrote on LinkedIn. And he's right. NextEra is one of the largest renewable energy developers in the world. Dominion controls the grid in Virginia — the data center capital of the United States. Put them together, and you have a utility that can build solar and wind farms, connect them to the highest-density data center market on earth, and charge premium rates for the privilege.

But here's the question nobody on Wall Street is asking: what happens to the ratepayers? The merger was announced in May, but the companies only filed for regulatory approval on July 15. The Daily Signal reported that Virginians are deeply concerned about what a combined NextEra-Dominion means for already-rising electricity bills. A $67 billion acquisition doesn't pay for itself. That cost gets recovered through rates — and rates are paid by people and businesses.

And let me tell you something about that $255 per month projection. If you're a hosting provider in Northern Virginia, that's not just a residential electricity bill issue. Commercial and industrial rates are linked to the same infrastructure costs. Your colo power bill is going up because the grid upgrades cost the same whether you're a hyperscaler or a hosting company running 50 racks.

What This Actually Means for Independent Hosting Providers

I've been running hosting infrastructure for over a decade, and I've never seen a structural constraint like this. Chip shortages you can work around — buy older gear, extend refresh cycles, use different suppliers. Power is different. You cannot run a server without electricity, and unlike chips, you cannot buy power on the open market from a different supplier. You're at the mercy of your local grid operator.

Here's what I'm telling hosting providers who ask me about this:

First — location matters more than it ever has. If you're planning a colo expansion or a new cage deployment, the single most important factor is no longer bandwidth or latency. It's power availability and power cost. Look at the PJM footprint — it's 13 states and over the next three years, every one of them is going to see capacity costs climb. Consider markets outside the traditional data center hubs. The QTS Digital Gateway collapse and the NY moratorium should tell you that the path of least resistance is shifting away from the East Coast.

Second — lock in power contracts now. If you have the ability to sign multi-year power purchase agreements at fixed rates, do it. The capacity auction results tell us that spot pricing is going to be brutal. Every hosting provider I know who locked in rates before 2025 is sitting pretty right now. Everyone who floated month-to-month is getting hit hard.

Third — efficiency is no longer optional, it's survival. I know we all talk about PUE and power efficiency at conferences. But the math has changed. Every watt you save in cooling, every server you consolidate, every old switch you replace with a more efficient model — that's not an environmental play anymore. It's a cost-of-doing-business calculation with a very concrete dollar figure attached. A 10 percent reduction in power consumption at PJM rates saves you more in a year than most hosting companies make on a dozen VPS plans.

Fourth — watch the FERC interconnection rules. In June 2026, FERC issued orders requiring grid operators to fast-track AI data center interconnections, with 30 to 60 day deadlines. But here's the catch — FERC also ordered that costs cannot be shifted to residential ratepayers. That means the interconnection costs for new data centers are going to land somewhere. And if the rules shield residential customers, the remaining pool of commercial and industrial customers — which includes independent hosting providers — will absorb a larger share. The fast-track is great for hyperscalers. For everyone else, it's a cost transfer.

FERC Commissioner Says It — the Status Quo Is "Untenable"

Last week, FERC Commissioner David LaCerte said the status quo at PJM was "untenable." That's a federal regulator using the same word PJM's CEO used. When the regulator and the regulated agree that the situation cannot continue, you know it's bad.

The PJM shortfall is 6.8 GW. For context, a typical nuclear reactor generates about 1 GW. So PJM needs the equivalent of seven new nuclear plants, and it needs them by 2028. The last time the United States built a new nuclear reactor from scratch, it took 43 years and went $17 billion over budget. Even if you factor in gas plants, solar, and wind as alternatives, the interconnection queue is backlogged by three to seven years. The data center growth is happening now, and the generation capacity to serve it won't arrive until the 2030s.

This isn't a shortage. It's a structural gap that will take a decade to close.

Australia's Approach — A Preview of What's Coming

Meanwhile, Australia announced on July 15 that it will require large data centers to generate as much power as they consume. That's right — new AI data centers in Australia will need to have on-site renewable generation matching their full load. It's the first national-level policy of its kind, and it is a direct response to the power grid crisis that every developed country is facing.

The question isn't whether other jurisdictions will follow. The question is how fast. If New York's moratorium and Australia's generation mandate are any indication, the era of data centers connecting to the grid with no questions asked is over. Every new facility will face scrutiny on power sourcing, water usage, and community impact. And that scrutiny will apply to smaller hosting facilities too, not just hyperscale campuses.

If you're an independent hosting provider, you need to prepare for a world where getting a new facility online takes not 6 months but 18 to 24 months because of power interconnection timelines. That changes your expansion planning fundamentally.

The Bottom Line — This Crisis Has Years Left to Run

I've been watching this industry long enough to know the difference between a cyclical hiccup and a structural shift. The chip shortage we've been dealing with since 2022 is a cyclical hiccup — painful, but eventually fabs will come online and supply will catch up. The power crisis is structural. Grids cannot be expanded quickly. Power plants take years to permit and build. And the hyperscalers have balance sheets that can outbid everyone else for every megawatt that comes available.

For independent hosting providers, the window to plan ahead is closing. If you haven't thought about where your next facility will be, what your power costs will look like in 2028, and how you'll absorb capacity price increases, now is the time. Not next quarter. Not when PJM runs its next auction and prices go up again. Now.

The power problem is the new chip problem. And unlike chips, there's no TSMC that can manufacture more grid capacity. You can't fab your way out of this one.

— Allan Ali, Founder