Something is breaking in America’s electrical system, and AI is holding the hammer.

This week, North America’s grid reliability watchdog NERC prepared to issue only its third-ever Level 3 alert — the highest severity — because AI data centers are threatening power grid stability. Microsoft is reportedly reconsidering its ambitious 2030 clean energy pledge. And the operator of America’s largest electrical grid published a paper saying the entire energy market needs to be redesigned.

This isn’t a theoretical future problem. It’s a crisis unfolding right now.

The Scale Is Almost Incomprehensible

Microsoft alone adds roughly one gigawatt of data center capacity every three months. That’s enough electricity to power 750,000 homes — a mid-sized city — every single quarter.

Data centers consumed about 4.4% of total U.S. electricity in 2023. The Department of Energy says that could hit 12% by 2028. The Electric Power Research Institute projects up to 17% by decade’s end. Grid-power demand from data centers is expected to nearly triple by 2030.

These aren’t ordinary loads, either. AI workloads demand dense, continuous power with massive cooling. Multiple gigawatt-scale complexes are being planned across the country, each requiring the output of a major power plant running 24/7 with zero tolerance for interruption.

NERC’s Historic Warning

NERC doesn’t issue Level 3 alerts lightly — it’s done it exactly twice before in its entire history. The alert targets transmission owners, grid operators, and utilities with seven “essential actions” addressing what NERC calls “significant risks” to the bulk power system.

The trigger: data centers in Virginia and Texas unexpectedly disconnected from the grid, causing power swings that can physically damage infrastructure. A paper from scientists at Nvidia, Microsoft, and OpenAI warned last year that AI power swings can “cause physical damage” to grid equipment. Imagine suddenly unplugging an appliance the size of a city.

Here’s the kicker — data centers currently don’t follow the same rules as power plants when it comes to telling grid operators they’re coming online or going offline. Energy analyst Lee Shaver of the Union of Concerned Scientists called this a “huge obstacle” to reliability.

NERC is now laying groundwork for mandatory reliability standards that would require FERC approval. The era of data centers operating as invisible loads on the grid is ending.

Microsoft’s Climate Reckoning

In 2021, Microsoft announced its “100/100/0” goal: matching 100% of electricity use with zero-carbon energy, every hour of every day, by 2030. This was far more ambitious than the typical corporate play of buying renewable energy credits to greenwash annual usage.

Now, according to Bloomberg, Microsoft is considering whether to delay or abandon that target entirely.

The math doesn’t work anymore. You can’t match hourly clean energy demand when you’re adding a gigawatt of load every quarter in regions where clean supply doesn’t exist yet. Microsoft expects to spend $190 billion this year, with AI infrastructure consuming a growing share.

The company hasn’t abandoned clean energy — it’s signed agreements for 1.2 gigawatts of carbon-free projects in Wisconsin and is pursuing nuclear power, including the Three Mile Island restart deal. But retreating from hourly matching sends a devastating signal. If the company that pledged to be carbon-negative can’t keep up with its own AI appetite, what hope does anyone else have?

America’s Largest Grid Operator Says the System Is Broken

PJM Interconnection serves 67 million people across 13 states. On May 6th, CEO David Mills published a paper called “Powering Reliability Through Market Design” that essentially says: the current system is broken, and time is running out.

The core problem is a “credibility trap.” High electricity prices should signal the need for new power plant investment. But high prices hammer consumers, triggering political intervention to cap costs, which undermines the investment signal, which means nobody builds new generation capacity. Meanwhile, data centers get built far faster than the power plants needed to serve them.

Mills outlined three paths forward. The second one is quietly radical: accepting that reliability can no longer be rationed equally. That means a future where AI data centers get premium, guaranteed power while residential consumers face more uncertainty — a two-tier grid that would have been politically unthinkable five years ago.

Fossil Fuels Are Back, Whether Anyone Likes It or Not

AI is bringing fossil fuels back, and climate advocates are watching in horror.

Planned natural gas capacity jumped from 11.1% of new additions in 2024 to 18.1% in 2026. Non-renewable planned additions surged 71% from 2025 to 2026, while renewable growth flatlined at just 2%.

The economics are brutal. Natural gas interconnection costs average $24 per kilowatt — compared to $253/kW for solar and $335/kW for offshore wind. Gas plants permit faster, build faster, and deliver the reliable 24/7 baseload that AI demands.

Massive natural gas projects were approved in Texas and Pennsylvania in 2026 specifically to serve data center demand. The AI industry’s hunger for power is actively undermining the clean energy transition, even as the same companies publish glossy sustainability reports.

Voters Are Getting Angry

Morning Consult polling shows more voters than ever blame AI data centers for rising electricity prices. FOX News framed it as “voters paying the price” for AI’s energy appetite, while climate advocates see Big Tech backtracking on environmental commitments.

It’s the rare issue uniting populists on both sides. Elected officials in data center-heavy states like Virginia face increasing pressure from constituents watching utility bills climb. This political heat could accelerate NERC’s regulatory push through FERC.

What Comes Next

We’re watching a classic infrastructure mismatch in real time. AI builds demand at software speed. The power industry builds supply at hardware speed. Something has to give.

Near-term: more natural gas, more nuclear deals, more corporate climate goals quietly revised downward. NERC’s mandatory standards for data centers will likely move through the regulatory process.

Longer-term: this crisis could accelerate energy innovation — small modular reactors, advanced geothermal, grid-scale storage. AI companies have very deep pockets, and necessity drives invention.

But the harder question isn’t technical. It’s political: Who bears the cost of AI’s electricity consumption? Should residential ratepayers subsidize data centers? Should AI companies build dedicated generation? Should we accept a two-tier grid?

Those answers are coming sooner than anyone expected.