Firm Power Is the Real Number, Not Nameplate Capacity

When a project lists hundreds of new megawatts, the figure describes installed capacity, not the power a grid can count on when demand peaks. For AI-scale load, that distinction is where reliability is actually decided.

The megawatt total attached to a generation plan is the easiest number to quote and the easiest to misread. A plant’s nameplate rating describes what it can produce under ideal conditions. It does not describe what a grid operator can depend on at the hour the system is most stressed. Those two quantities — installed capacity and firm, accredited capacity — diverge sharply for the resource types now being added fastest, and for loads that run continuously the gap is not academic. It is the difference between a portfolio that looks adequate on paper and one that holds up on a January evening.

Two numbers behind every megawatt figure

Every generating resource carries a nameplate: the maximum output it is rated to produce. For resource-adequacy purposes, what matters instead is accredited capacity — the portion of that nameplate a system operator will count toward meeting peak demand plus a reserve margin. For dispatchable thermal resources, accredited capacity sits close to nameplate, because the plant can be called on at will. For weather-dependent resources, accreditation is a fraction of nameplate, because output at the constraining hour cannot be guaranteed.

This is not a criticism of any resource type; it is an accounting reality of how reliability is measured. A region can hold a large quantity of installed capacity and still face a firm-capacity shortfall if too much of that installed total is credited at a discount during the hours that decide system adequacy.

Why AI-scale load makes the distinction load-bearing

The traditional planning paradigm assumed peak demand arrived on hot summer afternoons and receded overnight. AI training and inference load does not follow that curve. As the Readiness Brief notes, these facilities are generally planned as continuous loads that run around the clock, which means the resources serving them must be available at all hours rather than only during a predictable afternoon peak.

That changes which number matters. A resource that produces strongly at midday but little on a winter evening can look substantial on a nameplate basis while contributing modestly to the firm capacity a 24/7 load requires. The Brief’s central framing for this cluster — that ensuring adequate generation capacity is “only half the battle” — begins here: the headline megawatts are a starting point, not a measure of dependable supply.

How the regional operator decides what counts

Within MISO, capacity must be more than installed to be useful. It must be firm, controllable, and deliverable to the load it is meant to serve. The Readiness Brief is explicit that MISO grants capacity credit only to resources that pass a deliverability test — a demonstration that their output can actually reach load under peak conditions — and that where new transmission is needed to make a resource deliverable, its capacity credit may be reduced until that transmission is in service.

The Brief also records MISO’s own signal to utilities in its recent planning work: load growth through the mid-2030s, driven largely by data centers and electrification, will exacerbate an existing capacity shortfall, and “many new loads will require additional firm, controllable resources” to maintain reliability. In other words, the regional operator is drawing a distinction between megawatts that exist and megawatts that count — and asking that new load be matched with the latter.

What Wisconsin’s build-out reveals

The resource plans now before regulators reflect this logic. To serve expected data center demand, the state’s largest utility has proposed adding roughly 3 GW of new capacity. The composition is telling: on the order of 1,120 MW of solar, 180 MW of battery storage, and 1,400 MW of natural gas, according to figures compiled in the Readiness Brief. The single largest block is dispatchable gas, not because it carries the largest nameplate advantage, but because a continuous load requires resources that can be counted on at any hour, and the Brief notes utilities are currently leaning on fast-start gas to guarantee round-the-clock capacity for these loads.

The scale underscores why firm capacity is the operative measure. The Brief records that serving two proposed data centers alone would double one utility’s peak demand by 2030, and that a single gigawatt of continuous load is comparable to the consumption of roughly 700,000 homes. A load of that character cannot be met with capacity that is only intermittently available; it has to be met with capacity accredited as firm and shown to be deliverable to the site.

The distinction is local as well as firm

Firmness and deliverability travel together. The Readiness Brief describes how capacity credited toward a specific load must be deliverable to that load. A resource that is firm in principle but separated from the load by a congested transmission path may not count in full until the network is reinforced. A support resource sited near the load it serves strengthens local deliverability; distant capacity, even when firm, may be credited at a discount absent transmission headroom. The result is that “how much” and “where” are a single question. Installed megawatts in the wrong place, or of the wrong character, do not resolve a resource-adequacy requirement at a particular site.

The planning question

For anyone evaluating whether a region can serve a large new load, the nameplate total is the least informative figure in the packet. The more useful questions are how much of a proposed portfolio is accredited as firm capacity, whether that capacity is deliverable to the site under peak conditions, and how it performs in the season that now constrains the system. Those answers, not the headline megawatts, determine whether a large load can be integrated reliably.

The Readiness Brief frames its entire power analysis around this: capacity and deliverability are related but distinct, and both must be satisfied before a project’s power supply can be considered ready. Reading a nameplate figure as if it were firm, deliverable capacity is one of the more consequential ways an early-stage assessment can overstate readiness.

Installed capacity is a ceiling, not a guarantee. The number that governs whether AI-scale load can be served is firm, accredited, deliverable capacity at the constraining hour — a smaller and more demanding figure than any nameplate total. For a given site under consideration, what share of the proposed supporting capacity would MISO accredit as firm and deliverable at the seasonal peak, and how does that share compare with the nameplate figure being cited?

Source: the Wisconsin AI Infrastructure Readiness Brief, power capacity and grid deliverability — capacity accreditation and deliverability, MISO resource adequacy, and Wisconsin utility resource plans.