Here’s the thing about building AI infrastructure in 2026: you can have all the funding, all the chips, and all the engineers in the world — but if you can’t plug into the power grid, you’re stuck.
.jpg){kind=link}
Last Thursday, the Federal Energy Regulatory Commission (FERC) tried to fix that. In a unanimous vote, the commission ordered six major US grid operators to fast-track interconnection requests from data centers and other large electricity users. The message was clear: AI isn’t going to wait for bureaucracy.
But here’s the catch that nobody’s talking about — FERC gave data centers a faster lane to connect, without actually building more road.
What FERC Actually Ordered
The specifics matter here. FERC didn’t just send a strongly worded letter. The orders are binding, and they come with deadlines:
- 30 days: Grid operators must submit reports detailing how much spare generating capacity they actually have.
- 60 days: Grid operators must “defend or revise” electricity rates within their regions.
- Ongoing: Data centers must be able to connect “in a timely and orderly manner” — and they’re paying for it.
The commission also told grid operators to consider “alternative transmission technologies” — things like solid-state transformers and superconducting transmission lines. That’s a nod to the grid tech startups that have been pitching solutions for years while the traditional utilities dragged their feet.
On paper, this looks like a win. Data centers have been choking on connection delays for years. Some projects are stuck in queues that stretch five to seven years. The backlog is so long that at the end of 2023, grid connection requests for new power plants already exceeded the total capacity of the entire existing US power fleet. Read that again — the line to get on the grid was literally longer than the grid itself.
The Math Doesn’t Work
Here’s where FERC’s move starts to feel like rearranging deck chairs. The commission ordered faster connections, but it didn’t address the fundamental problem: there isn’t enough electricity.
Electricity demand from data centers is expected to nearly triple through 2035. Meanwhile, grid operators — which spent the last two decades dealing with near-zero demand growth — are buckling under the load. PJM, the country’s largest grid operator, has descended into what analysts politely call “operational challenges” and what utilities threatening to withdraw call something considerably less polite.
The result? Wholesale electricity prices have soared as much as 267% compared to five years ago, according to Bloomberg. That’s not a minor uptick. That’s a structural shift that changes the economics of running AI workloads at scale.
And it gets worse. Tech companies, unable to connect to the grid in reasonable timeframes, have been turning to behind-the-meter power — basically, building their own on-site generation. It’s more expensive, more complicated, and less reliable than grid power, but at least it exists. FERC’s order directs grid operators to be more accommodating to these arrangements, which is a polite way of admitting the grid itself can’t serve them.
The Wind Energy Paradox
What makes this story genuinely fascinating — and a little absurd — is the timing. The day before FERC’s data center fast-track order, the Trump administration announced it would pay $765 million to wind developer Invenergy to cancel offshore wind leases near California, Maine, and New York. Those projects would have generated as much as 2.4 gigawatts of power — enough to supply roughly 1.8 million homes at peak output.
In total, the administration has now spent about $2.6 billion to cancel offshore wind developments. That’s 2.6 billion dollars spent to remove generating capacity from a grid that can’t keep up with demand.
Think about that for a second. The government is simultaneously telling data centers “connect faster” while removing the power sources they’d connect to. It’s like telling someone to run a marathon while tearing up the road in front of them.
Secretary of Energy Chris Wright prodded FERC to act, arguing that grid delays threatened US competitiveness in AI. He’s not wrong about the urgency. But the administration’s parallel decision to cancel gigawatts of clean energy capacity undermines the very goal it’s trying to achieve.
What This Means for AI Companies
If you’re building AI infrastructure — training clusters, inference farms, data centers — this order changes your planning timeline but not your fundamental constraints.
The fast-track order means your interconnection application moves from a seven-year queue to something closer to a three-to-five-year queue. That’s better, but it’s not “plug in next quarter” territory. Data centers will still need to secure power purchase agreements, navigate local permitting, and often build their own substations.
The real impact is on the economics. With wholesale rates up 267% and grid operators now tasked with revising their rate structures within 60 days, the cost of power for AI workloads is becoming a major competitive differentiator. This is exactly why Amazon is pushing its own AI chips — as I covered in my analysis of Amazon’s chip strategy, companies that can optimize their hardware for lower power consumption will have a massive advantage in a world where electricity is getting more expensive, not less.
It also explains the explosion of battery storage investments in the AI space. When grid power is expensive and unreliable, batteries become essential infrastructure, not optional backup. The companies that figure out energy storage first will have a structural advantage that no amount of chip performance can overcome.
The Bigger Picture
FERC’s order is really a symptom of a much larger problem: the United States is trying to build a 21st-century AI economy on a 20th-century power grid. The grid was designed for a world of slow, predictable demand growth. AI is neither slow nor predictable.
The alternative transmission technologies FERC mentioned — solid-state transformers, superconducting lines — are real innovations that could help. But they’re years away from large-scale deployment. In the meantime, data centers will continue to be built faster than the grid can serve them.
This creates a fascinating strategic landscape. Companies like Google, Amazon, and Microsoft are increasingly looking at nuclear power, geothermal, and even fusion as long-term solutions. Smaller AI companies are exploring distributed computing architectures that reduce the need for massive centralized data centers. And grid tech startups are finally getting a regulatory tailwind from FERC’s directive to consider alternative technologies.
For the Philippines and other Southeast Asian nations watching this unfold, there’s a lesson here. As AI adoption accelerates globally, the countries that will benefit most are those that build energy infrastructure proactively — not reactively. The US is learning this lesson the hard way, spending billions to fix a problem that was predictable a decade ago.
The Bottom Line
FERC’s data center fast-track order is the right instinct at the wrong time. The commission is trying to remove bottlenecks in a system that doesn’t have enough capacity to serve existing demand, let alone the tripling that’s coming.
The real fix requires building more power generation, upgrading transmission infrastructure, and rethinking how we plan for explosive demand growth. FERC’s order handles none of that — it just asks grid operators to be faster about connecting to a grid that’s already overloaded.
For AI companies, the message is clear: power is the new bottleneck, and it’s not going away anytime soon. The ones that solve the energy problem — through efficiency, storage, or alternative power sources — will define the next era of AI infrastructure. The rest will be stuck waiting in line.