Every other cluster in this series has a moment — a valve, a wave, a missile, a relay — a point at which working infrastructure became broken infrastructure. This one has no such moment, because the infrastructure often never worked, or never existed. The disaster is the project: a plan so badly estimated, and a financing so badly structured, that its mere execution destroys billions and saddles the public for decades. The failure happened before the failure — in the estimate.

This is the cluster with the fewest bodies and, arguably, the widest damage, because it re-prices something abstract and enormous: society’s ability to build large things at all. When a nuclear project is abandoned half-built, when a rail line eats sixteen years and lays no track, the loss is not just the money. It is the credibility of the estimate itself — the reason capital now demands a punishing risk premium on exactly the long-lived infrastructure a decarbonising, growing world most needs. The failure before the failure is, in the end, a failure of our capacity to keep a promise about the future.

Over budget, over time, under benefits — over and over

The Oxford economist Bent Flyvbjerg, who has assembled the largest database of megaprojects in the world, reduced the pattern to what he calls the iron law: megaprojects come in over budget, over time, under benefits, over and over again. His explanation has two halves that recur in every case below. The first is optimism bias — planners genuinely underestimate cost and duration, because humans systematically do (the planning fallacy). The second, less forgivable, is strategic misrepresentation — promoters deliberately lowball the number and inflate the benefit, because an honest estimate would never win approval, and once ground is broken the sunk cost traps everyone into finishing.

Both halves point to the same structural truth: the estimate is not a forecast of what the project will cost. It is a price offered to get the project approved. Everything in this cluster follows from that one substitution — a sales figure wearing an engineer’s spreadsheet — and from the mechanisms that then make someone else, usually a ratepayer or a taxpayer, absorb the difference.

Nine billion dollars for an empty field

The purest case in the series is a power plant that produced no power. In 2013, South Carolina’s SCANA and Santee Cooper began building two Westinghouse AP1000 reactors at V.C. Summer — the first new US reactors in three decades — at an estimated $9.8 billion. In July 2017, after the owners had sunk roughly $9 billion (over $10 billion with interest) and analysts projected completion would cost more than $23 billion, they abandoned the project. Westinghouse had already filed for bankruptcy. No reactor was ever finished; no electricity was ever generated. What remained was, in a prosecutor’s phrase, an eleven-billion-dollar nuclear ghost town.

The twin that finished — and proved the point anyway

Georgia made the opposite choice from South Carolina. Facing the same Westinghouse bankruptcy in 2017, its regulators let Plant Vogtle’s twin AP1000 reactors continue — and they became the only new reactors completed in the United States in a generation. That is the good news and the whole cautionary tale at once. Vogtle Units 3 and 4 came online in 2023 and 2024, roughly seven years late, at a total cost near $35 billion against an original estimate of about $14 billion — the most expensive power project in US history. Georgia Power customers paid a construction surcharge for fifteen years before the reactors made a watt, adding up to more than $1,000 for some households, and face base-rate increases to recover the rest over decades.

The verdict that matters is the regulator’s own: the Georgia Public Service Commission’s staff testified that the overruns had eliminated any economic benefit to ratepayers, who would have been better served by natural gas. Together, V.C. Summer (abandoned) and Vogtle (completed at 2.5× cost) are not two stories but one: the practical un-financeability of Western nuclear, the same re-pricing that Three Mile Island first forced in 1979 and that no new build has yet reversed. The reactor works; the business case is the ruin.

The overrun that (eventually) delivered

Boston’s Central Artery/Tunnel Project — the Big Dig — is the transport archetype, and a useful complication. Estimated at about $2.8 billion when it was sold in the early 1980s and scheduled to finish in 1998, it was completed at the end of 2007 at roughly $14.8 billion — and about $24 billion once the borrowing to pay for it is included, with the final bonds due in 2038. It was plagued by leaks, substandard-materials scandals, criminal charges, and, in 2006, the death of a motorist crushed by a falling ceiling panel.

And yet: it works. The Big Dig buried an elevated highway, cut cross-city travel times dramatically, reclaimed a waterfront and created dozens of acres of parks and a new district. That is the complication this cluster must hold honestly — a megaproject can be a governance disaster in its estimate and its execution, and still deliver real, lasting benefit. The failure was not the tunnel; it was the promise that the tunnel would cost a fifth of what it did, made to a public that would not have approved the true number. The reform question is not “should we have built it” but “why can we only get these things approved by lying about the price.”

Sixteen years, fifteen billion dollars, no track

If the Big Dig is the overrun that delivered, California’s high-speed rail is the overrun that may never deliver at all. Approved by voters in 2008 with a $33 billion price tag and a 2020 completion date for a San Francisco–Los Angeles line, it had, by 2025, consumed some $15–18 billion over sixteen years without a single mile of track laid. The full-system estimate has climbed to roughly $89–128 billion (critics cite figures past $200 billion), the scope has been cut back to a 171-mile Central Valley segment not due until 2033, and in July 2025 the federal government terminated about $4.2 billion in funding, citing “no viable path.”

This is optimism bias and strategic misrepresentation in their textbook form — the 2008 estimate was widely judged politically motivated from the start — compounded by the specific American disease of building megaprojects through litigation, fragmented approvals and year-to-year funding rather than a committed capital plan. Whether or not a train ever runs, the project has already done its most durable damage: it has become the reference case every opponent of every future rail project will cite, raising the political and financial hurdle for the next one.

Not just an Anglo-American disease

Lest the pattern look like a failure of American governance specifically, Berlin Brandenburg Airport is the correction. Germany — the byword for engineering competence — planned to open its new capital airport in 2011 at a budget near €2 billion. It opened on 31 October 2020, exactly nine years late, at roughly €7 billion, into a pandemic that left its gleaming halls empty. The proximate cause was a bespoke smoke-extraction system, designed to push smoke down through the floor rather than up, that never passed fire-safety certification; the structural cause was governance — a public operator acting as its own general contractor, overseen by a supervisory board of politicians without the technical expertise to judge the problems in front of them. A manager was jailed for bribery along the way.

BER completes the geography of the cluster and sharpens its lesson. The iron law is not cultural; it is institutional. Wherever a project’s estimate is a political instrument, its oversight is amateur, and its risk falls on someone who cannot refuse, the same failure recurs — in Columbia, in Boston, in the Central Valley, in Brandenburg.

The estimate was the failure

Across a nuclear ghost town, a completed reactor, a delivered tunnel, an unbuilt railway and an empty airport, the common fault is not in the concrete. It is in the number that launched each one — an estimate that functioned as a sales pitch, protected by optimism bias, sharpened by strategic misrepresentation, and made survivable for its authors only because the overrun risk was structurally transferred to a captive public. The Base Load Review Act and Vogtle’s construction surcharge did this explicitly for nuclear ratepayers; general-obligation bonds and federal grants do it more diffusely for taxpayers; but the mechanism is always the same: decouple the party that makes the estimate from the party that pays for its being wrong.

What this cluster permanently re-prices, therefore, is the estimate itself — and, through it, the cost of capital for long-lived infrastructure. Once the market learns that a category’s headline numbers are systematically fictional, it prices that fiction in, which is precisely why Western nuclear now struggles to be financed at all and why large public works carry ruinous contingency. The reforms are known and mostly unadopted: reference-class forecasting (estimate against the actual distribution of comparable projects, not the promoter’s model), independent and adversarial cost review, staged financing that puts capital at risk on milestones, and a hard rule that the party which estimates should bear the consequence of estimating wrong. Until the estimator has skin in the overrun, the estimate will keep being a bid, and the failure will keep happening before the failure.

Recommended → codified → corrective → still live?

Every failure before this one was, in some sense, an accident — a gap between what an operator intended and what the world delivered. This cluster is different. Here the outcome was the intent, or an accepted price of it. One kind of failure is inflicted by an enemy who wants the infrastructure down; the other is subsidised by a state draining a resource it has decided is cheaper spent than saved. Neither is an engineering error. Both are decisions.

That changes the entire character of the reform ledger. You cannot instrument, retrain or re-standardise your way out of an adversary’s targeting or a subsidy’s incentives, because nothing malfunctioned. The grid did exactly what a missile does to a grid; the aquifer did exactly what an aquifer does when you pump it faster than it fills. The only reform that touches a chosen failure is a change in the politics that chose it — a treaty, an alliance, a subsidy repriced, a governance overhauled. This is the cluster where the analyst has to stop asking “what broke” and start asking “who benefited, and what would it cost them to stop.”

A power system on the front line

Since Russia’s full-scale invasion in 2022, Ukraine’s energy system has been not a casualty of war but an objective of it. Over 2022–24, roughly half of the country’s energy infrastructure was hit; about half of its large network substations were damaged; and the Zaporizhzhia nuclear plant — Europe’s largest, some 6 GW — has sat occupied and offline since March 2022. In the strike waves of 2024 alone, Ukraine lost around 9 GW of generating capacity, roughly a third of pre-war consumption; every one of its fifteen thermal power plants has been damaged or destroyed, collapsing thermal power from about 23.5% of the mix to around 5%. In June 2023 the Kakhovka dam was destroyed, draining a reservoir that cooled the nuclear plant and irrigated a region — an attack Ukraine attributes to Russian forces and Russia denies.

This is the defining feature of inflicted failure: nothing was neglected and nothing malfunctioned. The system was engineered and maintained to a normal standard and then deliberately dismantled by an adversary aiming to break civilian morale through cold and dark. The “reform,” accordingly, is not engineering but statecraft and defence: Ukraine’s emergency synchronisation with the European grid (ENTSO-E) days into the invasion, decentralisation toward distributed generation that is harder to target, air defence over substations, and an international scramble for the one component that decides recovery speed — large replacement transformers. The lesson the rest of Europe took is that a modern power grid is now a war target, and its protection is a security question before it is a technical one.

The pipeline that was blown up, and the attribution that wasn’t settled

On 26 September 2022, a series of underwater explosions ruptured the Nord Stream 1 and Nord Stream 2 pipelines beneath the Baltic near the Danish island of Bornholm, releasing what is regarded as one of the largest single discharges of methane ever recorded and rendering the pipelines inoperable. That it was deliberate sabotage was established quickly; who did it is, years later, still a contested and partly sub judice question — and this cluster’s evidentiary discipline demands it be left exactly there.

What the record supports: Denmark and Sweden, in whose waters the blasts occurred, closed their investigations in 2024 without charges, citing insufficient grounds. Germany’s investigation continued and has focused on a group of Ukrainian nationals allegedly operating from a chartered yacht — issuing an arrest warrant in 2024 for one suspect (who reached Ukraine), arresting another, Serhii K., in Italy in 2025, and filing formal charges against him in 2026, which he denies. What the record does not support is any conclusion about state direction: German prosecutors have not alleged the group acted on the orders of the Ukrainian government, Kyiv denies any involvement, and Russia has alleged Western responsibility without substantiation. The honest statement is that individuals have been charged, culpability has not been adjudicated, and the geopolitics of the blast remain a battlefield of their own.

A sea spent on cotton

The Aral Sea is the purest case of a chosen failure that no one attacked. In 1960 it was the world’s fourth-largest lake, about 68,000 km². Then the Soviet Union diverted its two feeder rivers, the Amu Darya and Syr Darya, to irrigate cotton across the Central Asian desert — through unlined canals that lost much of the water before it reached a field. The sea was not neglected; it was spent, as a deliberate and acknowledged trade-off for cotton. It has since shrunk to roughly a tenth of its former size, split into fragments, poisoned its own former seabed into the toxic Aralkum desert, and stranded fishing towns like Moynaq dozens of kilometres from any water.

The reform proves the thesis. There is no engineering fix for a sea drained by policy; there is only a political choice about who gets the water. Kazakhstan made one: with the World Bank it built the Kok-Aral Dam (2005), which revived the North Aral — water levels and fisheries returned — precisely by cutting off the South Aral, which Uzbekistan still drains for cotton. Half the sea was saved by deciding to sacrifice the other half. The underlying tension — too many claims on too little river — was not solved; it was allocated.

A slow-motion Aral, one tube-well at a time

India is the world’s largest user of groundwater, extracting around 247 billion cubic metres a year — rivalling the United States and China combined. In the northwest, NASA’s GRACE satellites measured a loss of roughly 109 km³ of groundwater across Punjab, Haryana and Rajasthan between 2002 and 2008 (later, tighter estimates put the annual rate somewhat lower, but the depletion itself is not in doubt). India’s Central Ground Water Board now classifies about a quarter of the country’s assessment blocks as over-exploited, critical or semi-critical; in Punjab, extraction runs far ahead of recharge.

The driver is a subsidy, which is why it belongs here rather than in the deferred-maintenance cluster. Free or heavily subsidised electricity for agricultural pumping, coupled with assured state procurement of water-hungry paddy and sugarcane, pays farmers to mine the aquifer — the number of tube-wells in Punjab alone roughly doubled between 1990 and 2018. No one is neglecting the groundwater; the policy is actively, expensively incentivising its depletion. And so, once again, the technical fixes (drip irrigation, artificial recharge) nibble at the edges while the subsidy drives the drain, because the real reform — repricing farm power and rebalancing crop procurement — is electorally radioactive. It is the Aral Sea in slow motion, run not by a central planner’s cotton quota but by a democracy’s inability to withdraw a subsidy.

The slow drain that (for now) reversed

South Africa’s Eskom is the slow drain applied to a power system, and the one case in this cluster with a recovery to study. Decades of under-investment, two flagship coal plants (Medupi and Kusile) delivered late and defective, an ageing fleet whose energy-availability factor sank toward the mid-50s in percentage terms, and the documented hollowing-out of the utility during the “state capture” era combined into chronic load-shedding. 2023 was the nadir: about 335 days of rotational blackouts and some 16.6 million MWh shed — a national economy throttled by its own power company.

The turnaround was governance, not gigawatts. A Generation Operational Recovery Plan from March 2023, the beginnings of Eskom’s unbundling into separate generation, transmission and distribution entities, a liberalisation that let private and rooftop solar surge (embedded solar rose from about 1.2 GW in 2021 to 6.1 GW by 2024), and a hard managerial focus on returning units to service produced a striking result: load-shedding was suspended for long stretches from March 2024, with roughly 352 load-shedding-free days in the 2025 financial year against 36 the year before. It is real, and it is fragile — the coal fleet is still old, the debt still heavy, and a chosen decline is only ever one set of political choices away from resuming. But it demonstrates the cluster’s converse: if a slow drain is a political failure, then political will can reverse it.

Chosen failures need political reforms

Line up a war-ruined grid, a sabotaged pipeline, a drained sea, a mined aquifer and a captured utility, and the common thread is not a mechanism but an intention. In each, the failure was selected — by an enemy who wanted the lights out, or by a state that decided the resource was better spent than conserved. That is why the reforms in this cluster look nothing like the instrumented, standardised fixes of the earlier posts. They are treaties and alliances (grid synchronisation, subsea-infrastructure patrols), allocations (a dam that saves one basin by sacrificing another), repricings (the farm-power subsidy no one will touch), and governance overhauls (Eskom’s unbundling). Where a technical failure is answered by an engineer, a chosen failure is answered by a politician — or not at all.

For the analyst, this reframes the entire risk. In the acute clusters, the question was whether an operator would maintain, widen or coordinate. Here the questions are geopolitical and institutional: is this asset a plausible target in someone’s conflict; does a subsidy or a quota reward its depletion; is the resource shared across a border where an upstream power can choose to withhold it; and is there a political constituency strong enough to reverse the drain, as South Africa found and India has not. These are not questions an engineering due-diligence answers. They are the questions that decide whether the infrastructure survives the politics that owns it.

Recommended → codified → corrective → still live?

A cascade begins with something small — a tree touching a line, a relay tripping, a voltage drifting high — and ends with tens or hundreds of millions in the dark. The distance between those two facts is the subject of this post. It is never bridged by the physics alone. The physics only propagates the fault; what decides whether it propagates or is contained is whether an institution caught it in time.

That is why the autopsy of every great blackout reads the same. The engineers find the trigger — and then the report spends its remaining hundred pages on situational awareness, coordination, standards, enforcement and who was supposed to be watching the whole system rather than their own piece of it. Tight coupling is the precondition; coordination failure is the cause. The reform, accordingly, is never a bigger wire. It is an institution.

The archetype: trees, training and tools

On 14 August 2003, roughly 55 million people across eight U.S. states and Ontario lost power in the largest blackout in North American history to that point. The trigger was almost comically small: high-demand lines in northern Ohio sagged into untrimmed trees and tripped. The reason it cascaded was institutional. In FirstEnergy’s control room, a software flaw had frozen the alarm system, so operators did not know their lines were failing; lacking situational awareness, and without effective coordination with neighbouring operators or the reliability coordinator, they took no corrective action while there was still time to island the fault. Over the next hours the loss propagated until, in its final phase, the Northeast collapsed in minutes.

The U.S.–Canada Power System Outage Task Force distilled the fix into a phrase — “trees, training and tools” — but the durable reform was constitutional. Before 2003, the reliability standards set by the North American Electric Reliability Council were voluntary. The Energy Policy Act of 2005 changed that: it added Section 215 to the Federal Power Act, empowered FERC to certify an Electric Reliability Organization, and made NERC’s reliability standards mandatory and enforceable, with penalties that can run past a million dollars a day. This is the cleanest “voluntary → mandatory” reform in the whole series — and the fact that the 1965 Northeast blackout had taught the same region the same lesson, without producing mandatory standards, is exactly why 2003 was needed to force the institution into being.

The largest blackout in history — a discipline failure

On 30 and 31 July 2012, India’s grid collapsed twice. The 30 July event isolated the Northern grid; the 31 July event brought down the Northern, Eastern and North-Eastern grids together, shedding some 48 GW and leaving about 620 million people — roughly half the country — without power. It remains the largest blackout in human history by population. Trains stopped, coal miners were trapped underground, and traffic in Delhi and Kolkata seized.

The Central Electricity Regulatory Commission’s enquiry committee found a governance failure with a technical trigger. A weak monsoon had driven up agricultural pumping demand in the north while the west ran a surplus, forcing huge inter-regional transfers across a corridor already thinned by line outages — effectively a single 400 kV Bina–Gwalior–Agra circuit doing the work of many. Northern states, chiefly Uttar Pradesh, Punjab and Haryana, drew power well beyond their schedules, and the state load-despatch centres did not act on the regional centres’ instructions to cut back. When the overloaded Bina–Gwalior line tripped on a protection mis-operation, there was no margin left, and the grids fell. The committee was blunt that the enforcement tools were toothless: the maximum penalties under the Electricity Act were “meagre,” and even those often went unpaid.

A modern cascade — and a modern alibi

At 12:33 on 28 April 2025, the entire electrical system of Spain and Portugal — plus parts of southern France — collapsed, leaving about 47 million people (some estimates run to 55–60 million) in Europe’s largest blackout on record. Around eight deaths were attributed to the outage and its aftermath, with economic losses estimated in the low billions of euros. The immediate physics was not frequency, as in 2003 and 2012, but voltage: a cascade of generator disconnections driven by rapidly rising voltage across the Spanish system.

The politically convenient story was that renewables did it — Spain was running a very high share of solar at the time. The July 2025 ENTSO-E expert-panel final report tells a more precise, and more familiar, story. The blackout came from a combination of voltage oscillations, gaps in voltage and reactive-power control, differing regulation practices, and generators disconnecting instead of riding through — including plants under contract to provide voltage control that failed to deliver it, and in one case supplied reactive power the wrong way. The academic post-mortems go further and name the root cause institutional: regulatory barriers that kept renewable capacity with certified voltage-control capability from being used, market-design incentives misaligned with real-time stability, and fragmented governance that slowed the crisis response. Spain’s key voltage-control operating procedure was roughly a quarter-century old, and a drafted update had reportedly sat with the competition regulator for five years.

In other words: the grid had the physical means to control the voltage; the institutions had not procured, priced or deployed it. That is not a renewables failure. It is the 2003 and 2012 failure — coordination and governance lagging a tightly-coupled system — wearing the newest available costume. The genuinely new element, and the one that belongs in the flashpoint rather than the blame, is that a low-inertia, weakly-interconnected, power-electronics-heavy grid gives the institutions far less time and far less margin to get it right.

A governance failure in an engineer’s costume

Across three decades, two continents and two different kinds of physics (frequency in 2003 and 2012, voltage in 2025), the pattern is identical. The trigger is always trivial and always technical — a tree, a relay, a voltage excursion. The propagation is always physics doing what tightly-coupled systems do. And the cause — the reason the fault spread instead of being islanded in the seconds available — is always that the coordination layer failed: an operator without situational awareness (2003), states defying dispatch instructions with impunity (2012), a market and regulator that never deployed the voltage control the grid physically had (2025).

This is why the reforms rhyme, and why they are always institutional. You cannot buy your way out of a cascade with copper. You get out of it with mandatory standards that bind every operator (2005), with an integrated system and enforceable discipline (2013), and with market and regulatory frameworks that price and procure stability in real time (the still-unfinished Iberian agenda). The engineer’s instinct — build a bigger interconnection — is double-edged: every integration that buys resilience against small faults also raises the ceiling on how large a coordinated failure can grow. The grid gets safer against the common case and more catastrophic in the rare one, and only governance decides which way the trade nets out.

Recommended → codified → corrective → still live?