The trading screens lit up first.
Before most people had finished their morning coffee, shares of old-school power plant companies were soaring — the kind of jagged green spikes traders don’t see often. Somewhere between Harrisburg and Wall Street, a single federal order had just made decades-old power plants the hottest new accessory in Big Tech’s AI arms race.[1]
Most Americans never heard the room go quiet in that Federal Energy Regulatory Commission meeting. They didn’t see Laura Swett, FERC’s chair, lean into the mic and call it a “critical step” toward solving “historic surging demand.”[1] They didn’t realize that, with one unanimous vote, the government had effectively opened the door for tech giants to plug their massive data centers directly into power plants — bypassing much of the grid everyone else depends on.[1]
But they will feel the impact, on their bills, their blackout risk, and the invisible infrastructure that now exists to keep artificial intelligence alive.
The New Power Couple: Big Tech + Power Plants
At the center of this shift is a word that sounds harmless: colocation.
In plain English, colocation means a giant power user — like a data center — sets up shop right next to a power plant and connects almost straight to the source.[1] Instead of feeding all electricity into the shared regional grid and then drawing it back out like everyone else, these facilities get a kind of VIP line.
FERC’s new order, aimed first at the mid-Atlantic grid run by PJM Interconnection, doesn’t just clarify the rules.[1] It legitimizes, structures, and accelerates this direct connection model — in a region that serves around 65 million people.[1]
On paper, the goal sounds sensible:
- AI and cloud computing are driving explosive electricity demand.
- The existing grid is old, overloaded, and slow to expand.
- Data centers want reliable, fast, predictable power — now.
Letting them tap power plants directly, FERC argues, could cut red tape, speed up AI and manufacturing growth, and “give investors and consumers more certainty” about how the U.S. will handle this demand surge.[1]
But certainty for whom?
A Deal Born in the Shadows of a Nuclear Plant
This order didn’t appear out of nowhere. It grew out of a quiet but high-stakes fight over a single deal: a proposed colocation agreement between Amazon Web Services and the owner of the Susquehanna nuclear power plant in Pennsylvania.[1]
The pitch: let Amazon’s cloud data centers plug in almost directly to the nuclear plant, giving it a private, ultra-stable power source for its AI and cloud services.
Power plant owners loved it. Utilities did not.[1]
Utilities argued that if big tech customers could sidestep them, they’d dodge paying for the transmission lines and grid maintenance that keep electricity flowing to everyone else.[1] Consumer advocates went further, warning that diverting existing plant output to private data centers could drive up prices and leave fewer resources available for regular ratepayers — households, small businesses, and critical services.[1]
What began as a one-off dispute turned into a test case. And that test case has now become a blueprint for the entire region — and possibly the country.
How the New Rules Actually Work
FERC’s order sets up new regulatory tracks for how these cozy plant–data center partnerships will function.[1] In practical terms:
- Custom pricing: PJM must design special rates and conditions for different colocation scenarios — including where a big user builds or co-funds new generation, or taps into an existing plant.[1]
- Cheaper transmission bills: A colocated data center may pay only for the transmission services it actually uses, instead of the broader grid costs a typical customer pays.[1]
- Payback for diverted power: If a data center taps into an existing plant, it may be required to cover the cost of replacing whatever energy is pulled away from the broader grid.[1]
To industry insiders, this looks like regulatory engineering to say: Yes, you can plug in — as long as we can show someone is paying for what’s lost.
To critics, it looks like the start of a two-tier power system: one for AI and hyperscale computing; another for everyone else.
One Family, One Bill, One Invisible Decision
Imagine a family in suburban Maryland.
Jasmine works nights as a nurse. Her partner, Miguel, runs a small HVAC business out of their garage. Their kids do homework on school-issued Chromebooks at the kitchen table.
Their town is served by the same regional grid now under pressure — a grid facing warnings of potential electricity shortages as data centers multiply faster than new power sources can be built.[1]
Over the next few years, they notice a pattern:
- Summer bills creep higher, even though they’re careful with the AC.
- Local news talks about “capacity warnings” and “peak alerts.”
- A substation upgrade is delayed again.
They never see the massive windowless buildings an hour away, lit inside by tens of thousands of servers training AI models that help power everything from streaming recommendations to automated customer service.
They don’t know that those facilities might be drawing power through special colocation deals that let them pay different transmission rates than their household does.[1]
All Jasmine and Miguel see is a bill.
All their kids see is a “please try again later” on the school portal during a brownout.
The Stakes: Growth, Grids, and Who Gets Protected
FERC insists the order will protect regular ratepayers, even as it clears a path for massive energy users to secure dedicated power.[1] Supporters say it sends a strong signal that the U.S. is serious about leading in AI and advanced manufacturing, not choking them with grid bottlenecks. Power plant owners certainly agree; their stocks surged on the news.[1]
But energy lawyers and grid analysts are already gaming out worst-case scenarios. One utility-side analyst put it this way in a recent briefing:
“If we carve off too much capacity for privileged loads and fail to build enough new generation, we don’t just risk higher prices — we risk reliability events that don’t show up on corporate balance sheets, only in people’s homes.”
Consumer advocates warn of a familiar pattern: private gain, socialized risk. Data centers can lock in firm power contracts. Power plants can secure premium customers. And when things get tight? The grid operator still has one primary lever — curtail demand from everyone else.
What’s Next / Could It Happen Again?
FERC’s order applies first to PJM’s territory, but its architecture is poised to become a national template.[1] Other regions, staring down their own data-center tsunamis, are watching closely.
Key open questions now hang over the entire system:
- Will regulators force enough new clean and firm power to be built, or will we simply reshuffle existing electrons toward AI?
- Will future rules lock in genuine protections for households and small businesses — or just model them on spreadsheets?
- And if colocation becomes the norm, do we end up with a de facto private AI energy network woven through public infrastructure?
Could it happen again? It already is. Each new hyperscale campus proposal, each nuclear- or gas-adjacent data center site plan, is now backed by a federal signal: There is a path for you.
The real question is not whether Big Tech can plug straight into America’s power plants.
It’s this: How much of our shared energy future are we willing to hand over, in the dark, to keep the machines thinking?
FAQ
What does “plugging data centers into power plants” actually mean?
It means Big Tech companies can build data centers directly beside power plants and connect almost straight to the source, instead of relying solely on the traditional shared grid and utility infrastructure.[1]
How does this federal data center power colocation policy affect my electric bill?
Colocation can change who pays for transmission and replacement power; if not carefully regulated, some of those costs can shift onto regular customers through higher rates for maintaining and upgrading the grid.[1]
Is this about AI data centers or regular cloud computing?
Both. The fastest-growing demand comes from AI training and inference, but cloud storage, streaming, and enterprise computing all increase the need for large, always-on data centers.
Are utilities being cut out of the picture?
Not entirely, but utilities lose some control and potential revenue when large customers bypass traditional grid connections and instead form direct agreements with plant owners under the new colocation framework.[1]
Does this federal data center energy ruling increase blackout risks?
Experts warn that if existing plant output is heavily redirected to private data centers without enough new generation, grid stress and blackout risks for ordinary users can rise, especially during peak demand.[1]
Can states or communities stop these colocated data centers?
Local permitting and land-use rules still matter, but the federal order gives powerful backing to colocation deals in interstate grid regions, making it harder for fragmented local opposition to override system-level decisions.
Is this the future of AI infrastructure and power in the U.S.?
Many analysts believe direct plant–data center colocation will become a dominant model for powering AI infrastructure, especially as companies chase guaranteed, stable energy supplies in an increasingly strained grid environment.[1]