by: David Riester
This article was originally published on DealFlow.Energy and has been republished with permission.
“A 10% gap in energy terms produces a gas plant with the same nameplate as the data center.”
Solar, wind, and storage are the cheapest and fastest new power in America. So why does a hyperscaler with a 500 MW campus end up commissioning a 500 MW gas plant?
The stock explanation – long queues, and intermittent resources that don't meet firmness requirements – is sort of right. But data center load is just about the only tailwind clean energy has right now, so it's worth understanding precisely where we lose.
The answer is not ideology or impatience. It's a stubborn marginal cost curve, and a market where both buyer and seller look at a 90%-complete clean plan and say, "Sorry, close doesn't count."
Why Clean Energy Rarely Gets to 100%
Take a routine configuration: a few hundred MWac of solar and wind plus four-hour storage, all electrically deliverable to a new large load. Sized sensibly, it covers roughly 90% of the load's annual energy at a competitive price, and faster than any honest service date for a new gas build. Up to that point coverage is cheap, because batteries solve the diurnal problem beautifully. Exciting!
Then the marginal cost curve goes vertical (and the marginal benefit plateaus). The remaining hours are not scattered randomly through the year; they are clustered, correlated, and seasonal – a run of overcast January days, the same stretch when everyone else is short too. Covering them from a single site means either i) absurdist overbuilds, or ii) long-duration storage that doesn't yet exist at scale. Push from 90% to 95% and the plan's cost roughly doubles; push to 100% and it's a multiple of that.
This is not a flaw in the technology, it is the reason the grid exists. Samuel Insull's founding insight was that one generator per customer was holding electricity back: pool diverse loads onto shared plants, and the coincident peak is far smaller than the sum of individual peaks. The last 10% is ruinously expensive for an individual and nearly trivial for most systems – imports, underutilized thermal plants, hydro, and demand response are already standing there, waiting, in exactly those hours.
The Knock-On
Here's where clean energy gets smoked. The load with a 90% plan turns to its LSE to buy the last 10% and firm the rest. The LSE isn't game. There is nothing on the shelf: no tariffed product, no obligation, no standard. Few LSEs are keen to stand behind 500 MW of new load while selling only ~50 MW of energy – the hardest 50 MW – at a standard industrial tariff rate. No market mechanism or regulatory compulsion currently corrects for that. The LSE: "Sure, I'll give you load service… if you come with 100% of the power you need." Sound familiar?
The hyperscaler wants certainty and control of its own fate, so this sequence of logic unfolds: build 100% yourself, maybe even behind the meter if the RTO is cramping your style. If you must cover all 8,760 hours, the thing you build has to carry the full load for the worst 25-50 of them – 500 MW of thermal, because on the gray January night the solar is dark and the battery ran dry hours ago. Once that 500 MW turbine is sunk, the solar and storage become the redundant asset, not the other way around. They get shrunk, deferred, or cut entirely. Nobody drew this up on a whiteboard, but the sequence keeps ending with a purchase order for 500 MW of Caterpillar recips.
A 10% gap in energy terms produces a gas plant with the same nameplate as the data center – roughly ten times the capacity the gap actually required – running at a capacity factor that would embarrass a peaker and parked behind a meter where it contributes nothing to anyone else's reliability. Multiply that by a few hundred campuses and you have the current moment: America de-electrifying the grid one data center at a time, and calling it progress because some coder's agent's sub-agent didn't suffer any latency.
The Fix: Sell the Last 10%
The idea is a special residual-firming tariff. A load – or a load-plus-generation pairing – that has demonstrably secured at least 80% of its expected annual energy from resources deliverable to it, under defined shape and performance criteria with financial commitments posted, is entitled to purchase firming and shaping for the remainder at an openly priced, tariffed premium.
Three features matter. First, the premium is a premium. Standing firmly behind an irregular residual is real work and deserves real compensation – and unlike islanded self-supply, every dollar of it finances resources that make the shared grid more robust for everyone. Conveniently, this is a political layup in that it very visibly makes data centers “pay their own way”. Second, it is resource-neutral. The residual can be served by storage of any duration, demand response, imports, a neighboring peaker, or the utility's existing fleet. If gas is truly the cheapest way to cover the residual, fine – at 10% scale, visible to the system, instead of 100% scale behind a fence. Third, it needs no new statute and no new technology. FERC is presently deciding how large loads interconnect and pay; those dockets are the natural vehicle for requiring RTO/ISO tariffs to carry a residual-firming product. The implementation specifics belongs to the states, but a federal model tariff gives them something to adopt. LSEs name the price; PUCs hold them to it – both to offering the tariff wherever the criteria are met and to pricing it honestly.
Is this a thumb on the scale? No. A subsidy pays someone to do something uneconomic. This completes a market: it requires that a price exist for a service the system can physically provide but structurally refuses to quote. The obligation being created is the obligation to name a price – the minimum a functioning market owes a willing buyer.
The skeptic's rejoinder is my favorite argument for it: if a 90% plan is really as good as I claim, the market should price it. Agreed. That is the entire idea.
