EPRI’s FlexMosaic Grades Load Flexibility as the Grid Runs Short on Dependable Capacity

Of the roughly 445 gigawatts of new generating capacity ICF expects to connect across the United States through 2030, only about 191 gigawatts will count toward meeting peak demand.

The additions are mostly solar and storage. After the firm adjusts for what each technology can reliably deliver when the system is most stressed, less than half of the nameplate total remains. The 191-gigawatt figure is the number that now governs grid planning, and it is the reason a measurement framework published this month carries more weight than the raw capacity coming online.

The shortfall. ICF’s June 25 analysis projects United States electricity demand rising 21 percent by 2030, with peak demand up 14 percent over the same period. The firm warns that PJM and ERCOT have no spare capacity to support demand growth beyond next year, and that the Southeast and New York are approaching the same limit.

ICF frames the constraint as a matter of timing rather than supply. The generation is planned; what it cannot do is interconnect, and the transmission to carry it cannot be built, fast enough to keep pace with the load arriving on the system. In that gap, the report argues, demand-side management and grid-enhancing technologies have to provide the reliability the wires cannot yet deliver.

The framework. That is the gap EPRI’s FlexMosaic is built to fill. Released as FERC’s large-load interconnection order took hold, the framework abandons the binary distinction between firm and interruptible load and grades flexibility into five classes. Each class is defined by the notification time a resource needs, the duration and frequency of its response, and the depth of the cut it can deliver. The classes span a spectrum, from rare and deep reductions invoked only under extreme system stress to frequent, shallow adjustments suited to routine balancing.

The taxonomy exists so that a flexible load can substitute, on paper and in a contract, for dependable capacity that cannot be built in time. FlexMosaic aligns the incentives by linking faster and larger interconnection to contractual commitments that protect the utility when the system is tight. For a data center waiting in an interconnection queue, the offer is concrete: accept a defined obligation to curtail, and move up the line.

The evidence. EPRI’s analysis puts numbers to the value. A 1 to 2 percent reduction in peak demand could lower electricity rates by 0.5 to 2.8 percent. AI workloads, by EPRI’s estimate, can offer between 18 and 55 percent flexibility relative to average power draw while still meeting quality-of-service requirements. In pilots run with NVIDIA, Portland General Electric, and Salt River Project, Emerald AI demonstrated load reductions of 20 to 33 percent. The framework names onsite backup power and stored energy as one of three core flexibility pillars.

The scoring. A grading system is not neutral. Once flexibility is scored on notification time, duration, frequency, and depth, the score sets the value, and the value sorts the resources. That sorting is the part of FlexMosaic its data-center framing does not foreground.

Measured against those four axes, stored energy rates well across all of them. A battery’s notification time approaches zero, its depth of response is controllable down to the inverter, its duration is bounded only by energy capacity, and it can cycle daily without a person in the loop. A curtailed compute workload scores differently, because it trades against the quality of service it was installed to deliver, and a managed workload depends on there being slack to manage in the first place. On a scale built to let hyperscalers buy interconnection speed with curtailment, dispatchable stored energy ranks above demand response and workload-shifting alike.

Where it lands for commercial storage. The open question for behind-the-meter batteries has never been whether they can shave a peak. It is whether the system credits them for it. Capacity accreditation has historically been built around generators, and the effective load-carrying capability methods that govern PJM and other markets were not designed to read a customer-sited battery as firm. A standardized flexibility vocabulary changes the terms of that argument. When a utility can place a resource in a flexibility class using the same axes it applies to a data center, a commercial battery is no longer asking for a credit the tariff has no language to express.

The cost-allocation half of FERC’s order points in the same direction. As large-load cost responsibility is re-sorted across the rate base, the capacity and demand components of commercial bills are where the volatility settles, and those are the components a graded, dispatchable resource is positioned to absorb. A framework written for hyperscalers ends up describing, in the same terms, what a much smaller customer-sited battery already does.

The shift underneath. For a decade the grid tried to build its way past its peaks. The 191-gigawatt figure is the acknowledgment that construction cannot keep pace with demand through 2030. FlexMosaic is the corollary: the next increment of reliability has to come from grading what is already connected rather than from pouring more concrete. On the scale the framework establishes, the resources that grade highest are the ones that can be instructed and will respond on command.

Sources