I've spent nearly three decades managing compute infrastructure. In that time, the thing that was never the constraint was power. You picked a data center, you co-located your gear, and the lights stayed on. The grid was a given.
That era is over.
On June 18, 2026, the Federal Energy Regulatory Commission issued show-cause orders to all six regional grid operators in the contiguous United States — PJM Interconnection, the Midcontinent ISO, Southwest Power Pool, the California ISO, ISO New England, and the New York ISO. In total, these organizations manage electricity delivery to roughly 200 million Americans across more than 30 states. FERC's message was unambiguous: justify your current tariffs for large power users within 60 days, or rewrite them.
This is not a routine proceeding. Section 206 of the Federal Power Act is the regulatory equivalent of a fire alarm. When FERC acts under Section 206, it shifts the burden of proof onto the utility or grid operator — rather than regulators having to prove a tariff is wrong, the operator must prove it's right. FERC Chair Laura Swett described it as a "national priority." That phrase doesn't appear in regulatory filings by accident.
How We Got Here
The numbers are not subtle. AI training clusters routinely exceed 100 MW. A single hyperscale inference facility can draw 200–500 MW continuously. When I started in this business, a 10 MW data center was considered large. Today, individual campuses are competing for power contracts that would have served mid-sized cities a decade ago.
The interconnection queue — the line of projects waiting to connect to the transmission grid — has ballooned to hundreds of gigawatts nationwide. Study processes designed for a different era are now bottlenecks measured in years, not months. As Intel's Prithpal Khajuria put it bluntly: "load growth is now outpacing the grid's ability to plan, price, and integrate it under the existing framework."
FERC's orders stem directly from a directive issued by Energy Secretary Chris Wright in October 2025, asking the commission to consider reforms for large-load interconnection. What might have taken years under normal rulemaking was compressed into targeted show-cause orders filed under Docket RM26-4. The urgency is intentional.
What the Five Reform Areas Actually Mean
FERC organized its show-cause orders around five categories of reform. These aren't vague aspirations — each one targets a specific friction point that anyone building or operating AI infrastructure has already encountered.
1. Efficient Transmission Study Processes
The interconnection study queue is broken. A single application can wait years before receiving a definitive interconnection cost estimate, and the queue is so backlogged that technical and cost assumptions change underneath projects before they even reach a study conclusion. FERC is pushing grid operators to incorporate alternative transmission technologies — advanced conductors, grid-enhancing technologies — and accelerate how studies are run. For data center developers, this means timelines for new large-load interconnections should start shrinking, but only after operators actually file acceptable reforms.
2. Preventing Cost-Shifting Through Transmission Cost Transparency
This one matters a lot, and it's often overlooked by people who focus on headline power costs. When a large data center connects to the grid, it typically triggers transmission upgrades — new lines, substation work, transformer replacements. Under current rules in several regions, some of those costs get spread across existing ratepayers rather than borne by the interconnecting customer. FERC wants that to stop. Jane Rueger, co-chair of Perkins Coie's data center practice, noted that "projects seeking transmission-level service should expect clearer requirements around applications, study procedures, readiness criteria, operating obligations, and transmission service terms."
If you're a hyperscaler that has been benefiting from cost socialization, this gets more expensive. If you're an existing ratepayer watching your bill climb to fund AI infrastructure you don't use, this is long overdue.
3. Accommodating Co-Location and Behind-the-Meter Generation
This is where I see the real structural shift happening. The model of "buy power from the grid, run your data center" is being replaced by hybrid architectures: build your own generation — gas turbines, nuclear microreactors, utility-scale solar plus storage — co-located with or electrically proximate to your compute. Microsoft, Amazon, and Google have all signed agreements with nuclear operators in the last 18 months. Anthropic just signed leases on more than 12 US data centers exceeding 1 GW of combined capacity, and you can bet generation co-location is baked into the planning.
FERC's explicit focus on co-location arrangements signals that this is no longer a niche development strategy. It's becoming the regulatory expectation for large loads that want expedited treatment.
4. New Transmission Services for Flexible Large Loads
This reform area is the most technically interesting to me operationally. The concept is demand-side flexibility at scale: a data center that can curtail or shift its load in response to grid conditions — much like a demand response program, but at the megawatt scale of an AI cluster.
Inference workloads have real elasticity if you build for it. You can queue requests, throttle batch jobs, defer training runs during peak grid stress. FERC is laying the groundwork for data centers to operate as grid participants rather than passive consumers. That's a different operating model, and it requires infrastructure teams to think about power management at a level most of us haven't had to before. The 20 MW threshold that triggers the new requirements will capture essentially every serious AI compute facility being built today.
5. Study Processes for Electrically Proximate Generation
When you build a gas plant or solar farm adjacent to your data center, the regulatory treatment of that generation varies wildly by region. Is it "behind the meter"? Does it count as transmission service? How is it studied for reliability impacts? FERC wants consistency across regions. Right now, a behind-the-meter gas turbine in PJM territory gets treated differently from an equivalent setup in CAISO. That patchwork slows development and creates arbitrage on regulatory risk between regions.
The 60-Day Countdown
The show-cause orders dropped on June 18. That puts the tariff justification deadline in mid-August and the 30-day resource adequacy reports due in mid-July. PJM has already signaled it views the orders as "a continuation of work already underway," which is encouraging — they've been dealing with significant load growth from both data centers and electric vehicle charging for years and have been further along on structural reforms than most other ISOs.
CAISO and ISO New England are in more interesting positions. California has a complex generation mix and existing demand response programs that may need substantial restructuring to accommodate large AI loads. New England has geographic constraints and transmission limitations that make new large loads particularly complicated to site and interconnect.
Watch the regional filings carefully. The consistency or inconsistency of how these six operators respond will define the development landscape for large AI infrastructure over the next three to five years.
What This Means If You're Building AI Infrastructure Now
A few things I'd be thinking about:
- Co-location is no longer optional at scale. If you're planning a facility above 100 MW, the regulatory trajectory is clear: you need a generation story. That might be a PPA with a nearby nuclear plant, a behind-the-meter gas turbine, or a large-scale solar-plus-storage deployment. The grid cannot be your sole or primary power source at this scale without significant interconnection costs and multi-year delays.
- Flexible load capability is going to become a procurement criterion. As FERC pushes grid operators to develop large-load flexibility programs, the AI infrastructure operators who can participate in demand response — especially those running inference at scale — will have advantages in interconnection cost treatment and potentially in energy pricing.
- The cost transparency reforms will hurt some and help others. If your development proforma assumed you could socialize transmission upgrade costs across the broader ratepayer base, those numbers are going to change. If you've been a victim of cost-shifted upgrades from adjacent data center development, you'll see some relief.
- Regional variation is not going away immediately. FERC is not imposing a single national standard. Each of the six operators will file its own response. The most favorable regulatory environments for large-load development will diverge from the least favorable, and that should factor into site selection now.
The Longer View
What FERC did on June 18 is significant not just for its immediate regulatory impact, but for what it signals about the relationship between AI infrastructure buildout and the physical systems it depends on. The grid is not an infinite resource. It is an engineered system with physical constraints, economic structures, and regulatory frameworks that evolved over decades for a world where load growth was predictable and measured in single-digit percentage points per year.
AI training and inference at scale is a fundamentally different demand pattern. It's lumpy, geographically concentrated, and growing at rates the grid was not designed to accommodate. FERC is adapting, as it must. But regulatory adaptation is measured in months and years, while model deployments and infrastructure commitments are measured in weeks.
The operators who are already building generation co-location, flexible load management, and grid participation into their infrastructure roadmaps are ahead of this curve. The ones still treating power as a simple utility cost — something you order like bandwidth — are about to learn otherwise. Tom's Hardware's coverage frames it well: projects should bring their own power or plan to cut usage during high demand. That's not a threat — it's a design constraint.
For me, the shift from "pick a colo and plug in your servers" to "negotiate interconnection agreements, study behind-the-meter generation options, and build demand response participation into your control plane" represents one of the biggest operational changes I've seen in this industry. It's the kind of thing that gets outsourced until you realize the strategic implications are too important to outsource — and then you have to build the competency in-house, fast.
The grid is having its own AI infrastructure moment. FERC just fired the starting gun.