Interview: Amory Lovins on MIT's AP1000 Cost Report

Copyright © 2023 Energy Intelligence Group All rights reserved. Unauthorized access or electronic forwarding, even for internal use, is prohibited.

In our Apr. 29 edition, we invited the author of a recent MIT study on the economics of the AP1000, MIT’s Professor Koroush Shirvan, to elaborate on his findings. In an interview this week with Energy Intelligence's Stephanie Cooke, Amory Lovins responded to some of those findings. Lovins is adjunct professor of civil and environmental engineering at Stanford University. This interview was conducted via email, and was shortened and condensed. The hyperlinks are from Lovins.

Q: In his recent Q&A, MIT’s Professor Shirvan asserted that “we know historic nuclear power plants can operate safely to 80 years” and that in operating costs, or once their capital costs are paid off, they become competitive “with any source of electricity.” Do you think that’s possible now, and if not now, could it be possible for a new reactor in another 20 years for a reactor built today?

A: The world’s two longest-running power reactors, Beznau-1 in Switzerland and Nine Mile Point-1 in the US, are only in their 53rd year, so 80 is speculative. NRC [Nuclear Regulatory Commission] just suspended prior perfunctory license extensions from 60 years to 80 and will re-examine "aging management programs." In 2022, would you count on a 1942 car, however expertly maintained and refurbished? Can a reactor even run reliably and economically for 60 years? A chronic control rod drive seal leak closed Palisades 11 days early at age 51. The 40 US units closed by mid-2021 — 30% of total nuclear grid connections — averaged 22 years, and only eight had reached 40; the six closed in 2016–20 averaged 46 but were licensed for 60; and many if not most operating reactors cost more to run than they can earn in competitive markets.

In the actual competitive landscape, I don't see how nuclear newbuild of any type or size can compete in economic dispatch against unsubsidized renewables whose lifetime Levelized Cost of Energy (LCOE) in many countries has fallen through $30 and then $20 per megawatt hour — even to $10. By the time the next US AP1000 could be built, still-cheaper renewables will undercut Shirvan’s claimed $20-$40/MWh nuclear dispatch cost by even more. His critique of SMR [small modular reactor] economics seems directionally correct and implies neither SMRs nor traditional LWRs [light-water reactors] could pencil out: their bleak prospects of competing in operating costs after 2050 can’t justify their huge investments now.

No standard empirical dataset finds nuclear newbuild competitive with unsubsidized renewables. But nuclear probably becomes even less competitive if we count grid integration costs, which tend to be larger for big thermal plants because their forced outages are generally bigger, longer, more abrupt, and far less predictable than variable renewables’.

Q: You’ve criticized the MIT report for ignoring renewables. Asked about that in his interview with Energy Intelligence, Shirvan cited the fracking/natural gas boom as the reason nuclear was ignored back in 2008. Furthermore, he says that while nuclear can’t compete with renewables that will not be the case as markets approach "net-zero" emissions targets. He cites studies that he says conclude that at less than 5 grams of equivalent carbon dioxide per kilowatt hour, an AP1000 with an LCOE of $80-$120/MWh can compete with solar/wind. What is your response to that?

A: Shirvan compares AP1000s only with other reactors, ignoring demand-side and renewable competitors. Yet renewables have already taken around 95% of the world market for net capacity additions, versus nuclear’s less than 1% (and in seven of the past 13 years, less than 0%). PV [photovoltaic] and wind power are the cheapest bulk power source in over 91% of the world and rising (says BloombergNEF), with three to eight times (Lazard) or 5-13 times (BloombergNEF) lower LCOE than nuclear. IPCC [the Intergovernmental Panel on Climate Change] also says the demand side can provide 40%–70% of global decarbonization.

Investors and owners rejected nuclear newbuild because it had no business case, so in 2001-20, the world opened three fewer power reactors than it closed. But newbuild does worse against efficiency and renewables today than against gas in 2008. Grid models like MIT’s can make hypothetical $80-$120/MWh new nuclear compete with new solar and wind only by conjuring a need for "firm" power. They do this by constraining or omitting most of the 10 kinds of carbon-free grid-balancing resources — a portfolio so ample that the costliest kind, bulk electricity storage, is seldom needed.

Shirvan’s low AP1000 cost projections assume doubled construction productivity, near-record construction speed, unobserved learning curves, orders of at least 10 units to trigger those assumed cost drops (circularly assuming US market prospects that high early costs destroyed), 4%–8% cost of capital with no apparent nuclear risk premium, and decades-long PPAs [power purchase agreements] or equivalent utility reregulation. These "should cost" fantasy assumptions make AP1000s look prohibitively costly rather than astronomically costly, but wishing will not make it so. Especially in the nuclear industry’s advanced state of decay, such make-believe numbers merit scarcely more credence than Westinghouse’s original claims that simplicity and modularity would yield AP1000s buildable for $1 per watt in 36 months. Well, if we had some ham, we could have some ham and eggs, if we had some eggs.

Far from gaining climate relevance, nuclear newbuild makes climate change worse, because it costs far more — probably an order of magnitude more — than other carbon-free competitors, so it saves proportionally less carbon per dollar, and does so more slowly. Why pay extra for less-effective climate solutions? The more worried we are about climate, the more vital it is to buy fast, cheap, sure options — not slow, costly, speculative ones.

Q: You’ve taken issue with the methodology used in the MIT report — specifically the use of overnight costs as compared to total costs, including finance. Could you explain why you don’t agree with that approach?

A: Traditional overnight cost metrics are economically meaningless because they exclude financing costs (and often major owner’s costs, too, which Shirvan estimates at $1.9/W initially, falling to $1.3/W). Financing costs are sensitive to two linked variables: construction duration and cost of capital. Nearly all the carefully analyzed modern nuclear programs worldwide have been significantly over budget and schedule: among 180 reactors, 92% showed cost overruns averaging 117%, while 175 units showed schedule overruns averaging 64% or 36 months. Capital markets therefore charge nuclear risk premia of typically several to many hundreds of basis points.

Omitting financing costs is thus likely to mislead. For example, Shirvan estimates Vogtle 3–4’s EPC [engineering, procurement and construction] overnight cost at around $21 billion or approximately $9.3/W as of February 2022 (in 2018 dollars), twice MIT’s 2009 estimate; but the May 2022 total cost estimate is around $34 billion or $15.2/W and rising, with no assurance of ultimate operation. The inferred financing cost, totaling in the neighborhood of $10 billion, would be far higher without $12 billion of federal loan guarantees and access to Treasury-window rates. Shirvan finds Vogtle’s costs are in line with other post-TMI [Three Mile Island] US nuclear construction costs, but those are financeable only in imaginary markets or with Vogtle’s conscripted capital.

Q: Based on 2018 dollars, Shirvan says the EPC cost of a new AP1000 "should" be $4.3/W — less than half Vogtle’s cost, which he puts at $9.3/W. Questioned about the impact of inflation and Covid-19 since then, Shirvan says the costs would rise by at least 20%. What’s your best estimate of the cost of a new AP1000?

A: Nobody knows: "at least" is vague and unbounded. That’s partly why the private capital market won’t finance them. Why even try to guess a number as imaginary as the weight of a hippogriff? Being less materials-intensive than an EPR (with some corresponding safety questions) is as irrelevant in today’s market as comparing hippogriffs to minotaurs. Vogtle 3-4 is America’s first-of-a-kind AP1000 only because so many others previously fell by the wayside, such as Bellefonte 3-4, Summer 2-3, Levy County 1-2, Harris 2-3 and Turkey Point 3-4. Vogtle continues only because taxpayers are holding the bag.

Q: Putting it slightly differently, how much does financing and decommissioning add to the cost of building and running a nuclear plant?

A: Financing can easily add plus 30% to bare EPC costs; for distressed projects, plus 50%. Shirvan’s paper shows that each $1 of Vogtle original EPC cost incurred $1.55 of site construction management cost (quadruple expectations), 45¢ of owner’s cost, 13¢ of interim payments and liens, thus $3.13 of overnight cost, plus 95¢ of project financing, raising total capital and financing cost to $4.07 per $1 of EPC cost. (This excludes 85¢ from Toshiba related to the Westinghouse bankruptcy, and imputed interest during construction, plus any local taxes, on compulsory pre-funding by ratepayers.) Financing cost rises with market capital, or long duration (the world average is nearly a decade but projects vary widely), or both.

No one really knows yet what decommissioning a big old US LWR will cost — only that it’ll probably exceed the escrowed funds. Global decommissioning experience with roughly 6 GW completed so far, mostly small reactors, reveals systematic underestimation of duration and cost; German decommissioning cost more than construction. Many retiring US reactors are being quietly transferred from accountable owners to special subsidiaries of disposal firms that can pay themselves the billions in the decommissioning accounts, then if the work isn’t finished, apparently leave the rest to the taxpayers.

Q: Could you explain the significance of net capital additions (NCAs) to nuclear plant costs? Shirvan puts them at $5/MWh, included in his LCOE estimates for an AP1000. Percentage-wise, how much do you estimate NCAs add to nuclear plant capital costs over a 40- or 60-year lifetime?

A: NCAs are maintenance, safety, or betterment investments that don’t nominally pay back within a year, so they’re capitalized rather than expensed. The Nuclear Energy Institute [NEI] says 2020 US NCAs totaled $4.2 billion and averaged $5.34/MWh (2020 dollars)—18% of $29.37 total generating cost (excluding various listed owner’s costs). Restoring seven omitted years, US NCAs in 2002-20 averaged $7.90/MWh (2020 dollars), which if sustained for 40–60–80 years at (say) a 5% annual real discount rate would have a present value of $122–$134–$139 per kilowatt plus financing. Future NCAs are speculative and unforecastable, but corrosion, fatigue, wear and radiation damage may well make them rise with age, much as they do for our bodies.

US NCAs peaked in 2012 at $12.42/MWh (2020 dollars) — 27% of that year’s total nuclear generating cost of $45.39. The high 2010-12 NCA values were raised by "enhancements" — presumably upratings and lifetime extensions if safety improvements are counted as "regulatory." NEI ascribes the post-2012 NCA decline chiefly to program completions; the influence of plant retirements and regulatory relaxations is unknown. The largest NCA type since 2016 has been "sustaining," implying repairs and upgrades to keep operating. NEI has concealed top-quartile operating-cost data since 2016; perhaps they’re embarrassingly high. If high enough, they can force closures.

Q: Nuclear advocates often argue that “learning experience” significantly lowers capital costs from an inaugural plant to an “nth of a kind” — or tenth plant — and that this has been demonstrated in France, South Korea and Japan. Shirvan argues that follow-on AP1000s in China realized significantly lower construction times and consequently lower costs. He says the same was true with the ABWR in Japan. Why don’t you buy that argument?

A: Shirvan’s evidence, if any, for a learning curve needs independent expert scrutiny. Competent literature reports no observed learning curve for modern reactors anywhere — specifically not in France, where real costs rose with experience. (The Flamanville-3 fiasco then confirmed the fragility of institutional learning: the 14-year ordering gap lost a generation of seasoned managers and supply-chain capabilities needed to sustain such a complex and finely tuned construction process.) The paper on which Shirvan relies for his claim of nuclear learning curves in "many countries" was demolished by three leading experts (and by another paper whose senior author later cited the paper he’d debunked as his own key reference for arguing the opposite). Several noted nuclear-cost scholars have found Kepco’s South Korean cost data unanalyzably opaque, shifting and unverifiable. I’ve seen no literature dissecting the inscrutable and unverifiable cost data from Japan — or from China, where AP1000 experience was mixed and BloombergNEF says new nuclear is two to three times costlier per kWh than new solar and wind.

Details matter: e.g. Shirvan cites his own co-authored paper as claiming that Japan’s four ABWRs were all built in under four years, but IAEA’s [International Atomic Energy Agency] Pris database says their duration from construction start to commercial operation rose from just over four years to 4.5 to 4.6. No power reactor on Earth has even achieved grid connection in under four years since 1970. But Shirvan measures construction time as ending with fuel load, not with commercial operation. He likewise celebrates that Haiyang-2's civil works took less than four years, without noting that its total completion took 8.3 years, versus 4.8 years planned.

Q: Often one hears that renewables provide only "intermittent" power. You say that adjective would best be applied to baseload power sources instead, and that renewables are better described as variable. Please explain.

A: No generator is 24/7/365, but their outage characteristics differ. Renewables are generally dispatchable except PVs and wind power, whose output varies with weather and the Earth’s rotation. But those variations, which Shirvan calls "unpredictable," are very predictable — often more so than electricity demand — making (for example) east Danish wind power biddable into the hourly day-ahead auction just like thermal capacity. In contrast, big thermal plants’ forced outages can lose a gigawatt in milliseconds, often for weeks or months, and typically without warning — behaviors properly called "intermittent." Well-designed PV and wind projects’ and portfolios’ variations tend to be milder, brief, shorter, and far more predictable, so they’re best called "variable."

The venerable honorific term "baseload" has at least five meanings — none useful in a world where, as Miso’s [Midcontinent Independent System Operator] strategist Jeff Bladen said, we’ll no longer be forecasting demand and scheduling supply, but instead forecasting [renewable] supply and scheduling demand. One generator does not serve one load; rather, all generators serve the grid, which serves all loads. Customers care about timely statistical deliverability of total resources, not single units. In an era when generators were less reliable than the grid, we built the grid to back up failed thermal units with working ones. Now in the same way, but often at lower cost, the grid can back up PV or wind output with each other, with other renewables in other places, or with demand response, storage (electrical, chemical or thermal), or other resources like industrial cogeneration.

So what’s really intermittent? This year, France is officially forecast to have more nuclear capacity shut for repairs and safety checks than total German nuclear closures throughout 2000-21. The 2022 French nuclear capacity factor is forecast at 55%–59% (versus 2020’s 61% average) but has lately been 42%–52%. Like Japan’s post-Fukushima shutdowns, a nuclear monoculture can suffer large-scale common-cause failures. Germany’s diversified, competitive, half-renewable power supply now looks not just cheaper than France’s (beating French wholesale annual-average costs in all but one year since 2007) but also more reliable and resilient — fortunately, since Germany’s long-standing net power exports to France lately hit new highs.

Q: The American Nuclear Society says that the US fleet has "maintained a median capacity factor near 90% for 20 years." How does that compare with the output relative to capacity of wind power in the US? Or solar?

A: An unknowable combination of improved operations (with all due credit to Inpo and Naval reactor culture), culling lemon plants, and laxer regulation has indeed kept the average capacity factors of the surviving US units around a commendable 90% (versus IAEA’s global cumulative average of about 75%), falling below 90% in only three years since 2007. The US 2020 nuclear average of 92.4% compared with utility-scale 2020 averages of 69% geothermal, 65% "other non-fossil gas," 63% nonwood biomass, 58% wood, 41% hydro, 35% wind and 24% photovoltaic. All resources’ LCOEs account for their respective capacity factors and shouldn't be double-counted. Grid integration costs are generally small for variable renewables but larger for big thermal generators.

Risk-conscious investors are also aware of the 92 operating US power reactors’ survivor bias. Of 259 units ordered, by mid-2017 only 28 — some now slated for closure — had been built that remained competitive in their regional markets and did not suffer at least one outage of a year or more.

Q: Finally, what is your model for getting to "net zero" in 2050, or could it be sooner?

A: Let me summarize a few conclusions from a recent review paper and its terse summary. Global fossil-fueled power generation peaked in 2019 and, by a fluke, in 2021, but renewables can already or will imminently meet all the further demand growth. IEA [International Energy Agency] says the world added about 294 GW of renewables in 2021, and expects about 305 GW more in each of the next five years, but so far has underestimated their growth.

Renewables’ business case is so decisive that if markets and policies plausibly overcome or bypass obstructions, the US could cut its fossil-fueled power generation to about zero in the 2030s, with huge financial savings (even bigger in India and China). The energy transition could be even faster and cheaper if demand-side resources too were systematically competed or compared with supply-side resources — especially as integrative design makes electrical savings severalfold larger, yet cheaper, often with increasing returns.

With its roughly three-to-thirteenfold higher LCOE (and probably higher grid-integration costs) than unsubsidized renewables, nuclear power has no business case and hence no climate case. It’s therefore winning about 10–20 times less global investment and adding hundreds of times less annual output. The US government is lavishly subsidizing and promoting virtually every kind of reactor— shredding decades of patient nonproliferation efforts, so DOE [Department of Energy] is undermining DOD’s [Department of Defense] national-security mission — but even these tens of billions of dollars of new largesse are unlikely to prevent the stagnant nuclear industry’s slow-motion collapse. Nuclear trainwrecks will multiply as taxpayers’ billions are spent, deceived customers are fleeced, and rosy claims prove false.

Nuclear advocates tout new roles for process heat, desalination, hydrogen, etc., but new uses can’t remedy grossly uncompetitive prices. Nor can putatively soaring electricity needs to electrify transport and heat — doubtful if energy efficiency and materials efficiency are competed or compared with supply and storage — build a nuclear business case. I doubt any vendor has cumulatively made money selling reactors—only fueling and fixing them. A modern grid also has no reliability or operational need for “firm” generation. Renewables’ global potential is orders of magnitude greater than plausible long-run need. Nuclear power has finally run out of reasons to exist, except as an enabler and cover for bomb programs and a sinecure for lobbyists.

The famous Nuclear Renaissance (boosted by a faulty MIT analysis) has saved no carbon. Maybe it never will — but its $40 billion-plus cost could have saved far more carbon sooner by buying efficiency and renewables. Should we now repeat that error?

Nuclear Policy
Wanda Ad #2 (article footer)
OPG launches SMR innovative SMR delivery model; France's regulator calls for thorough nuclear program review; Oklo prepares to license a US nuclear fuel recycling plant.
Fri, Jan 27, 2023
A recent carbon neutral LNG trade between Shell and Taiwan's CPC brings back scrutiny over the accuracy of carbon offsets in these transactions.
Tue, Jan 31, 2023
In Brussels nuclear is the energy that dare not speak its name, but it still may benefit from electricity market design changes under consideration.
Fri, Jan 27, 2023