Victoria Yang is an energy analyst at Latitude Intelligence.
Data center flexibility has become the grid’s favorite compromise. Let large loads curtail during a few hours of peak demand, the argument goes, and in exchange potentially connect to the grid years sooner.
Hyperscalers have begun to broker flexibility agreements directly. In March, Google said it had signed one gigawatt of demand response into long-term contracts with Indiana Michigan Power, TVA, Entergy Arkansas, Minnesota Power, and DTE. Every one is a bilateral deal with a utility, negotiated site by site.
Now the Southwest Power Pool has written the first version of a market policy ostensibly to encourage flexibility, but one that could instead end up complicating things. In June, federal regulators approved SPP’s conditional high impact large load service, or CHILLS. It gives data centers access to non-firm transmission — meaning SPP can cut them off whenever the transmission system is constrained or in an emergency — for up to seven years while they wait for firm service.
SPP’s conditional service won’t kill data center flexibility. Rather, it risks moving flexibility into bilateral contracts priced and dispatched out of public view, where its value isn’t discovered and its terms can’t be replicated by other loads. A CHILLS load can be cut off at any moment if needed by the grid, with no ceiling on the hours of curtailment.
So in order to avoid the unbounded risks that CHILLS present, the rational move for hyperscalers is to keep negotiating flexibility bilaterally with utilities that will write call limits into a contract — even if they have non-firm transmission.
There’s a public cost to that approach. Loads that hold out for firm service need transmission upgrades built to serve them, and depending on how a region allocates those costs, ratepayers could end up covering part of them. Private terms also set no public price, so smaller developers never get the same deal as massive hyperscalers. And the easiest way to route around the grid is on-site gas, which comes with more local pollution and greenhouse gas emissions.
SPP’s July compliance filing states that the grid operator and its transmission owners should undertake no obligation to plan the transmission system to accommodate CHILLS load, and that the load must fully respond to any curtailment directive. The tariff sets no annual hour cap and guarantees no minimum notice.
A stakeholder revision that SPP’s Markets and Operations Policy Committee approved on July 14, RR767, goes further, requiring CHILLS to reach its curtailment target within 15 minutes of an instruction from SPP or the local transmission operator, and placing CHILLS curtailments ahead of SPP’s emergency demand response programs. It does not limit how often curtailments can be called.
SPP has a point. CHILLS curtailment is triggered by transmission constraints such as forced outages, line trips, and other contingencies that happen in real time. Those appear in minutes, not a day ahead. A guaranteed notice period would force SPP to hold capacity for the load as if it were firm, which is exactly the service CHILLS exists to defer. Other programs, including in both ERCOT and Pennsylvania, have taken different approaches to the same problem, but only CHILLS is FERC-endorsed.
The numbers behind the case for flexibility
The case for flexibility was built on bounded numbers.
Duke’s February 2025 Rethinking Load Growth study, one of the most-cited arguments for curtailable data centers, found that the 22 largest U.S. grid operators and utilities could together integrate around 76 GW of new load if that load agreed to curtail 0.25% of the time, which is “equivalent to 10% of the nation’s peak demand,” according to the study. It found that events would average under two hours; for 88% of those curtailment hours, at least half the new load would keep running, the scenario studied found.
In the 21 months since the study was released, a Latitude Intelligence analysis found that utility retail commitments have multiplied, especially in SPP’s territory.
A hyperscaler can live with real-time curtailment. What it cannot underwrite, however, is not knowing whether the number is 20 hours or 400 hours each year for up to seven years. A rational hyperscaler will only be willing to experiment with flexibility in contexts where it can set the terms. That means that bilateral utility contracts with defined call limits and on-site generation will continue to be the default. The flexibility still exists — just in private deals that the public market never hears about.
Every demand response arrangement Google has announced sits outside an RTO tariff, Latitude Intelligence found. The first one, signed in 2024, is for a demonstration at Omaha Public Power District, which is inside SPP’s own footprint. CHILLS gives the hyperscaler no reason to change its ways.
CHILLS matters beyond SPP because it is the first answer to a question the Federal Energy Regulatory Commission has now put to every RTO. In June, FERC issued show-cause orders to all six grid operators under its jurisdiction. Among other reforms, they ask each operator to adopt an interim non-firm transmission service for large loads, or explain why its tariff is just and reasonable without one.
PJM and MISO are still settling their large-load rules: FERC rejected most of PJM’s plan, and MISO has its own proposal pending. By Duke’s estimate, those three regions have room for 43 GW of flexible load at a 0.5% curtailment rate, and SPP has room for another 10 GW.
If SPP’s design becomes the default, “flexible” will come to mean non-firm with no ceiling. And hyperscalers will simply route around it.
That leaves the rest of the grid worse off. Flexibility written into a private contract is dispatched against one utility’s needs, at sites a hyperscaler chose for its own reasons, and the market operator can’t call on it when a constraint binds somewhere else. Meanwhile, the potential headroom that Duke counted would largely be left idle, even as firm requests wait on upgrades.
