Sizing a Risk Is Not Assigning It

Two nuclear projects were under construction at the same time in the American Southeast, on the same reactor design, under the same designer-constructor.

One finished. Vogtle Units 3 and 4 were expected to cost $14 billion and to enter service in 2016 and 2017. They entered service in 2023 and 2024, with total spending nearing $35 billion.

One did not finish. V.C. Summer Units 2 and 3 were abandoned in July 2017, four months after overruns at both projects drove the reactor vendor into Chapter 11.

The pair is normally read as the verdict on American nuclear construction. The more useful reading is comparative. Between them, the two projects sort their own causes of failure into two categories, and only one of those categories still has bite.

What actually went wrong

The Department of Energy names what Vogtle brought into its construction. An incomplete design. An immature supply chain, in a country that had not built a reactor in three decades. And an untrained workforce, demonstrated by the roughly thirty thousand people Vogtle trained during the build.

To that list add the contracting itself. Both projects ran under substantially fixed-price EPC arrangements with Westinghouse, and the parties holding at-risk construction work did not stay around. Shaw Group and then Chicago Bridge and Iron shed their at-risk positions to escape mounting losses, after which Westinghouse brought in Fluor on terms of carrying no such exposure. When that arrangement collapsed, Southern Nuclear assumed project management at Vogtle and Bechtel came in as constructor, cost-reimbursable plus fee. At that point, the contractual risk transfer had ended, and completion depended on whether the ownership group could continue supplying capital.

The cost decomposition matters enormously. Vogtle Units 3 and 4 carried an overnight capital cost near $11,000 per kilowatt as spent, restated at roughly $15,000 per kilowatt in 2024 dollars. DOE separates that figure into parts. Roughly $2,900 per kilowatt is true first-of-a-kind cost. DOE identifies roughly $3,800 per kilowatt of that as Vogtle-specific inefficiency that a future build avoids, covering the incomplete design, Part 52 licensing against an unfinished design and the Westinghouse bankruptcy with the EPC change that followed, and places the next AP1000 near $8,300 per kilowatt. The entire gap between those two figures is cost the next build does not carry.

What cured itself, and what did not

Most of those causes cured themselves. Incomplete design, immature supply chain and untrained labor were conditions of the first build alone. They were paid for in cash by a project which will not need to pay for them again.

Today's context differs dramatically. The AP1000 design is complete, supply chain infrastructure now exists and the workforce has been trained. Westinghouse has since filed to make the as-built Vogtle Unit 4 the standard reference unit for all future AP1000 licensing, a submittal now under NRC review, which on approval would convert a hard-won project record into a repeatable product.

Further, the cost trajectory can now be observed rather than only modeled. DOE reports Vogtle Unit 4 construction running roughly thirty percent more efficiently and roughly twenty percent cheaper than Unit 3, on the same site with the same crews. Key testing milestones completed 38 to 76 percent faster and engineering service requests fell by roughly half. DOE cautions that shared infrastructure between the two units makes a full disaggregation difficult, which bounds how precisely the cost figure can be read. Corroborating evidence, and the only evidence on dispersion, comes from China. Across the second series of AP1000-derived units, average duration from first nuclear concrete to milestone completion fell 48 percent against the first series and average duration between milestones fell 42 percent, with the corresponding variances down 60 percent and 46 percent. Those schedule figures are drawn from milestones reached so far in units still under construction, describing a trend in progress. The variance reduction matters most, as dispersion is what an underwriter prices.

Projections built on that experience place the overnight cost of the next two AP1000 units at a new site near $8,300 per kilowatt-electric. A four-unit build at a single site falls to roughly $7,500 per kilowatt-electric on co-location and learning, and the eventual nth-of-a-kind cost is put near $4,700 per kilowatt-electric.

Every one of those figures moves the expected cost downward. Not one of them determines who funds the outcome in which the improvement fails to repeat.

The contracting failure did not cure itself. What the industry took from Summer and Vogtle is that unbounded completion exposure cannot be handed to a single vendor. That lesson also removed the only allocation mechanism either project had, with nothing to replace it. Bechtel finished Vogtle on cost-reimbursable plus fee, which does not allocate overrun risk so much as leave the owner holding all of it. The industry now has less overrun allocation machinery than it had in 2013.

That unassigned exposure is what remains. Neither project had a plan for cost overrun management, and no amount of design completion touches it.

Why the unassigned exposure survives

It survives because it was never the kind of problem the industry believed it was solving. Contingency estimating is a mature discipline and it answers one question: how large the exposure might be. It leaves five others open. Who funds the exposure when it materializes? When does each funding obligation activate? Is the obligation capped? How does the funding party recover? What remains unfunded beyond the committed layers?

Estimated exposure and funded exposure are different objects, and a project can hold an excellent answer to the first while holding none at all to the second.

An overrun's funding need arrives during construction and is concentrated in the years of peak spend. The project's capacity to repay arrives only after commercial operation. The exposure is a mismatch between when capital is needed and when it can be earned back — a liquidity problem wearing a cost problem's clothes. Left unfunded, it converts a cost problem into a completion problem at the exact moment the project's bargaining position is weakest and its perceived risk is highest.

Two ways of handling that gap have now been tested at full scale in the United States. The first was contractual transfer. It failed on its own terms at both sites, as there is no scenario in which the contractor absorbing the exposure fails while the project it is building succeeds. The second was not chosen. When the transfer collapsed, whatever exposure the contractor could not absorb had no named bearer, so it was assigned after the fact, with the project already in trouble. At Vogtle that meant the owner taking over project management, a reset budget, regulatory accommodation and federal support negotiated across several years. That worked once, at a cost and duration no financing plan would choose in advance. Neither is a repeatable financing strategy.

The private market does not close the gap on its own either, as capital arrives to nuclear in two forms and neither can hold this exposure. Venture and strategic capital funds reactor designs, prices for total loss and exists to manufacture optionality across a portfolio. Project finance capital funds construction, prices for close to zero loss and will not commit until exposure has already been assigned elsewhere. Overrun risk on a proven design is too large for the first and too unbounded for the second.

It belongs to neither, so it belongs to nobody.

What leaving it unsolved costs

The AP1000 could plausibly be the lowest-cost generator in the United States, conditional on scale across dozens of units, with each build handing its learning to the next. This is the condition under which every other industrial product becomes cheaper.

Nothing yet built reaches that scale. Two efforts are live. Santee Cooper selected Brookfield to complete the abandoned Summer units, now managed through a joint venture Brookfield formed with The Nuclear Company, and the federal government entered a partnership with Westinghouse and its owners directed at eighty billion dollars of AP1000 deployment, taking a participation interest in the upside. Neither has reached a final investment decision, and at Summer the combined licenses terminated in 2019 must be replaced through fresh applications.

Look at what the two have in common. In the restart, the firm completing the units also owns the reactor vendor and holds a position in the venture managing the construction, so one balance sheet spans positions that would otherwise sit across a contract. In the fleet program, the federal government holds a participation interest in the vendor's upside. Each arrangement is available to its particular participants rather than to the market generally. Neither has published an allocation of overrun exposure. Scale has been announced, but who pays if it goes over has not been settled.

Vogtle paid the price of being first once, and what that bought was a second build that does not repeat it. The advantage is perishable. Crews disperse, qualified suppliers leave a market with no order book and the engineering teams that carried the learning move on. The next project drifts back toward first-of-a-kind cost, the distribution around it widens and the exposure that stopped the project in the first place gets larger. Delay does not preserve the option to build later. It prices it higher.

Demand, meanwhile, runs on a schedule. Firm clean power commitments at gigawatt scale carry dates, and a delivery market that cannot be financed sends that load to gas.

The decision nobody has made

A project cannot credibly promise that it will stay within budget. Very few large industrial projects can. What a project can do is decide, before the first pour, who pays if it does not.

Neither reference project made that decision in advance. Both attempted it in the crudest available form, and in both cases the assignment expired with the counterparty. Vogtle then made the real decision late, under duress, among parties whose alternatives had already collapsed. Summer never made it, and the units sat unfinished for years. Neither decision was made at a moment when anyone had the standing to make it well.

Cost overrun exposure does not yield to better building. It yields only to structures agreed in advance, with each dollar assigned to a party that has examined it, priced it and accepted a defined maximum. No project in the United States has published one.