The Flexible Megawatt

Every proposal an owner receives answers the same implicit question: how do we supply this load. Almost none of them ask the question that comes first and costs least to answer, which is how much of this load has to be supplied in every hour, and what the site would accept in the handful of hours when the grid is genuinely short.

Section 01The question that precedes the technology question

An organization discovers it needs more electric capacity than its service can deliver, or than the utility can deliver on the date it needs. The sequence that follows is nearly always the same. Someone estimates the new load, that estimate becomes a firm requirement, and the firm requirement becomes the specification against which generators, batteries, and utility upgrades are priced.

The load estimate is the least examined number in the whole exercise, and it is the number that sizes everything downstream. It is usually built from nameplate ratings, expansion plans, and a safety margin, and it is almost always expressed as a single figure that the site must have available continuously. That framing is a choice, not a fact. It converts a load with a shape into a load with a wall, and every path priced against that wall gets more expensive.

The alternative is not to assume the site can go without power. It is to establish, with measured data and operating knowledge, which portion of the load is genuinely inflexible, which portion could shift in time, and which portion could be reduced briefly under notice if something were paid for the privilege. Those three portions have very different costs to serve. Treating them as one is how owners end up buying firm capacity for loads that never needed it.

This paper is about the third portion. It is deliberately narrow, and it is not an argument that flexibility replaces building. It is an argument that flexibility is a priced, tariffed, quantifiable input to the sizing decision, and that a study which omits it has skipped a step rather than completed one.

Section 02What the system actually has

The strongest published evidence on this question comes from research led by Tyler Norris at Duke University, summarized in testimony to the House Energy and Commerce Subcommittee on Energy in March 2025. The study examined 22 of the largest United States balancing authorities, together serving 95 percent of the country's peak load, and asked how much new load the existing system could absorb if that new load agreed to be curtailed occasionally.¹

The stated findings are specific. Roughly 76 gigawatts of new load, equivalent to about 10 percent of the nation's peak demand, could be integrated if new loads accept curtailment for 0.25 percent of their maximum uptime. At a 0.5 percent curtailment rate the figure rises to about 98 gigawatts, and at 1.0 percent to about 126 gigawatts. The average curtailment event in the analysis lasts about two hours, and in nearly 90 percent of the hours where reduction is required, at least half of the new load can keep running.¹

Two caveats belong in the same breath as those numbers, and the authors state both. The work is described explicitly as a first-order estimate that does not resolve transmission constraints, and the testimony is direct that the findings should not be read to mean the country can meet near-term and medium-term demand without building new peaking capacity or expanding the grid.¹ An owner should treat these figures as evidence about the shape of the opportunity at system scale, not as a result that transfers to a particular meter on a particular circuit. What transfers is the underlying mechanism: systems sized for the extreme hour carry unused capacity in the ordinary hour, and a load willing to behave differently in the extreme hour is cheaper to serve.

Section 03What flexibility is already worth in the market

Demand flexibility is not an emerging idea. It is an established, metered, compensated resource class, and the federal record quantifies it annually.

The Federal Energy Regulatory Commission's 2025 Assessment of Demand Response and Advanced Metering, published in December 2025, reports that total demand response participation across the seven United States wholesale markets reached 33,272 megawatts in 2024, an increase of approximately 217 megawatts or 0.7 percent over 2023. Approximately 6.5 percent of the wholesale market peak demand across all regional transmission organizations and independent system operators could be met by demand response resources in 2024, against a sum of non-coincident peaks of roughly 515 gigawatts.²

The California figure is more relevant to a California site. In the same report, demand response capacity in the California Independent System Operator footprint rose from 4,154.3 megawatts in 2023 to 4,373.0 megawatts in 2024, an increase of about 219 megawatts or 5.3 percent, representing approximately 9.0 percent of CAISO peak demand.² California is meaningfully above the national average on this measure.

Two honest readings follow. The optimistic one is that a resource meeting nine percent of peak in this market is not a pilot; it is infrastructure, and an owner who ignores it is ignoring a real market. The sober one is that national growth of 0.7 percent in a year of extraordinary load-growth pressure suggests this resource is harder to expand than its arithmetic promise implies. Both readings are supported by the same table. A study should present both.

Systems sized for the extreme hour carry unused capacity in the ordinary hour. A load willing to behave differently in that extreme hour is cheaper to serve.

Section 04What a California tariff actually pays, and actually charges

General claims about flexibility are worth little. Tariffs are specific, published, and binding, and they are where an owner should look. Pacific Gas and Electric Company's Electric Schedule E-BIP, the Base Interruptible Program, is a useful worked example because every term below is in the filed tariff rather than in a brochure.³

The program requires enrolled customers to reduce load to a committed Firm Service Level when called. The terms, as posted in the tariff book, are as follows.

An illustrative calculation, and the reason it is only illustrative

Consider a site that can commit 2,000 kW of load reduction under the 30-minute option. At the posted rates, six months at $13.50/kW and six months at $10.50/kW would produce roughly $288,000 across a year. That figure is illustrative. It ignores the default adjustment values that reduce the potential load reduction on which payment is actually calculated, assumes uninterrupted enrollment and full-year participation, and reflects rates current as of the access date below rather than any future period.

Now price the other side with the same discipline. If that site exceeded its Firm Service Level by 1,000 kW for a single six-hour event, the excess energy charge at $6.00 per kilowatt-hour would be approximately $36,000 for that event, again illustrative on the same assumptions. Eight such failures would erase the year. That asymmetry is the entire design of the program, and it is why the operational question precedes the financial one: the money is straightforward, and the ability to actually come down on 30 minutes' notice, repeatedly, in the hours the grid is worst, is not.

Section 05The honest case against

A paper that presented flexibility only as found money would be the same kind of document this series exists to correct. The case against is substantial and site-specific.

• Production risk is real and often understated. Curtailment events cluster in the hours when the system is stressed, which in California means hot summer afternoons. For many industrial and agricultural operations those are also the hours of highest production value or highest cooling need. The incentive is paid in dollars; the cost is paid in output, and the two are rarely compared in the same units.
• Restart is not free. Processes with thermal mass, curing cycles, batch integrity, or purge requirements do not resume where they stopped. A two-hour curtailment can cost considerably more than two hours of production, and in some processes it costs a batch.
• Committed reduction constrains growth. A Firm Service Level agreed today is a ceiling the site must still respect after it adds a line, a shift, or a tenant. Flexibility commitments should be sized against the operation the site expects to be, not the one it is.
• Flexibility is shallow before it is deep. The first increment of load reduction is usually available at modest cost through non-critical systems. Each subsequent increment reaches closer to the core process and gets sharply more expensive. Owners who extrapolate the first megawatt's ease to the fifth are projecting a linearity that does not exist.
• Meeting the commitment with a generator is a different project. Sites that plan to hold their Firm Service Level by running on-site generation during events have not avoided a capital decision; they have created one, along with permit conditions, run-hour limits, fuel arrangements, and testing obligations that belong in the analysis from the start.
• The measurement question is unavoidable. Performance is judged against a baseline or a committed level, and the definition of that reference decides what the site is actually paid. This is a contract term worth reading closely rather than delegating.

Set against those, the case for is equally concrete: the payment accrues monthly on committed capability rather than only on events, the annual exposure is capped and published, the resource requires no new generating equipment, and it can be implemented on a timescale that no construction path can match. For a site whose real constraint is the date on which capacity arrives, a resource that can be arranged in months rather than years deserves to be priced before a resource that cannot.

Section 06This is not the same thing as flexible service

Two distinct ideas are converging in the market under similar language, and conflating them produces bad decisions.

Flexible connection service is an arrangement with the utility to connect a new or expanded load sooner in exchange for accepting curtailment conditions on that service. It changes when the site gets power and on what terms.

Demand response is a market and tariff arrangement in which an existing load is compensated for reducing consumption when called. It changes what the site earns and how it operates, not whether new service exists.

They interact usefully. A site that has already demonstrated it can reduce load reliably is a more credible candidate for flexible connection terms, and the operating capability behind both is the same capability. But they are separately negotiated, separately governed, and separately priced, and an owner should never accept an analysis that quietly treats a demand response payment as though it were an interconnection outcome, or that treats a flexible connection offer as though it came with a demand response revenue stream. Those are two documents with two counterparties.

Section 07What a disciplined flexibility screen documents

A flexibility screen is a bounded piece of work that belongs early in any power decision, before technology selection. Six items make it defensible.

1. Load segmented by inflexibility, not totaled.Interval data divided into the portion that cannot move, the portion that can shift in time, and the portion that can be reduced under notice. A single annual or peak figure cannot support any of this analysis.
2. Depth and duration stated separately.How many kilowatts can come down, for how long, how often, and with what notice. A site that can shed deeply for ten minutes and a site that can shed modestly for six hours qualify for entirely different things.
3. Restart cost priced in production terms.The cost of a curtailment event expressed in output, scrap, labor, and recovery time, converted to dollars by the operations team rather than the energy analyst.
4. Program terms read against the site, not summarized.Notice, event caps, penalty rates, baseline definition, and enrollment duration compared against what the process can actually deliver in the hours events are called.
5. The reduced capacity requirement carried into sizing.If a portion of load is genuinely flexible, every downstream path is sized against a smaller firm requirement. This is where flexibility pays most, and it is the step most often skipped.
6. The commitment tested against the growth plan.A Firm Service Level or curtailment obligation checked against the load the site expects in five years, with the exit and amendment terms identified.

Section 08Where flexibility cannot substitute for capacity

The limits deserve as much clarity as the opportunity, because the failure mode here is an owner who defers a necessary decision on the strength of a resource that was never going to carry the load.

Flexibility does not create energy. It relocates or forgoes consumption, and a site whose fundamental need is more energy delivered over the year, rather than less demand in a few hours, will not solve that need with a curtailment agreement. It cannot substitute for a service upgrade when the constraint is the capacity of the wire or transformer serving the site in ordinary hours. It offers little to genuinely continuous processes with no tolerance for interruption, and pretending otherwise puts a production line behind a tariff penalty. And it does not deliver resilience: a program that requires the site to reduce load when the grid is stressed is close to the opposite of a resource that keeps the site running when the grid fails. Those are different problems with different equipment, and the same asset rarely does both without deliberate design.

What flexibility does reliably is change the size of the question. A firm capacity requirement reduced by a documented, contracted, operationally tested amount makes every remaining path smaller, cheaper, and faster to permit. That is worth establishing before capital moves, and it costs a fraction of what the capital costs.

The discipline this series argues for is not that any one answer is right. It is that the requirement should be measured before it is met, and that the cheapest megawatt on any site is usually the one nobody had to build.