There is enough water in the Danube. The pumps cannot reach it.

Monday, 3 August, 09:00 CET. Three of the four blocks at Hungary’s Paks nuclear plant have stopped. The fourth is running at half power, and the Danube stands one centimetre above the level at which that one has to be switched off as well. Output is 240 megawatts against a normal 2,000. Paks supplies close to half of Hungary’s electricity, and in forty-four years of operation it has never been fully shut down. By the time you read this, that may no longer be true.

The reason is where this becomes interesting, and it is not the reason most people assume.

On 30 July, the plant’s operator, MVM, published an unusually clear statement. At the Danube’s current extremely low level and flow, it said, there would still be enough water to cool the reactors. The difficulty is that the intake nozzles of the pumps which deliver that water sit higher than the river now runs. They cannot do their job.

The water is present. The pumps cannot reach it, by roughly the height of a kitchen counter.

That is not a drought story. It is a specification story, and it carries a number that deserves attention from anyone who signs off on infrastructure business cases.

The number that expired

The engineers who designed Paks in the 1970s worked to a hundred-year minimum water level: the lowest the Danube could reasonably be expected to fall, derived from the best hydrological record available at the time.

The river is now more than a metre below that line.

Not approaching it, not testing it. More than a metre underneath, and still receding. The previous record low, set in 2018, was itself already beneath the design basis. The current level sits roughly 28 cm below that 2018 record.

The engineers were not wrong. They were handed a number produced by sound method from the available history, and they built correctly against it. What failed was not the calculation. It was the assumption that the number would go on describing the river.

A hundred-year minimum, it turns out, is a statement about the last hundred years. It was never a promise about the next fifty.

It is worth pausing on how ordinary this failure is. Somewhere in almost every asset-heavy business case sits a number of this kind: annual water availability, minimum river level, historical flow, expected rainfall. It is calculated once, usually well, by people who are careful. Then it becomes an input rather than a question, and after a few years nobody can remember who produced it or on what basis. It is the least examined and most load-bearing figure in the file.

It is not only climate, and the difference matters

MVM’s own account of the causes is more honest than the headline version, and repeating it accurately changes the commercial lesson.

The operator names two drivers. The first is riverbed deepening. Dams and hydropower structures on the upper Danube trap sediment; the Hungarian stretch has no such structures and therefore continues transporting sediment downstream; and decades of dredging compounded the effect. The bed fell, and the water surface fell with it. The second is climate: this year the usual seasonal high-water events, including the spring melt flood, essentially failed to materialise, leaving the river low throughout.

So the design basis was invalidated by a slow, cumulative, cross-border engineering effect and by a fast climatic one, arriving at the same moment. This distinction matters commercially, because the two have different owners and different remedies. Nobody upstream built a dam intending to shut down a Hungarian reactor. Nobody dredging a navigation channel was thinking about pump intake elevations forty years later. Each decision was locally rational. The system that emerged from them is nobody’s design, and everyone downstream inherits it.

On Friday evening a government decree came into force in Hungary allowing large industrial consumers to be disconnected from the grid. That is the point at which water risk stops being an environmental topic and becomes a line in an industrial operations plan.

Across Europe the pattern is now visible at scale. Around six nuclear plants and roughly fifteen reactors are shut down or curtailed. Romania has taken a reactor at Cernavődă offline. In France, EDF shut a unit at Golfech on 30 July because Garonne temperatures had risen to the point where returning cooling water would breach environmental limits. Switzerland’s Beznau, on the Aare, has been alternating between shutdowns and half power.

Two distinct failure modes sit inside that list. Paks cannot physically reach its water. Golfech and Beznau can reach theirs, but the regulator will not permit them to discharge it that warm. One constraint is mechanical, the other legal, and neither is visible from a drought map. You cannot know which one binds at a given site without examining that site’s actual system.

What this does to the water business

Here the story stops being about energy.

For roughly thirty years, water has been sold as a cost line. The proposition was efficiency: use less, pay less, here is your payback period. It is a sound proposition and it built a substantial industry.

It also contains an assumption that has now become visible. Efficiency arguments quietly presume the water shows up — in the quantity, at the quality, and at the elevation your equipment expects. They are optimisation arguments, and optimisation arguments get evaluated by whoever owns the operating budget, on whatever timetable suits them. This is why so much water technology has such a pleasant, unhurried and non-converting sales cycle.

When supply becomes unreliable, water stops being a cost line and becomes a continuity risk. The question shifts from “what does this cost us per cubic metre” to “what happens to output when the intake is above the waterline.” Those are different questions. More importantly, they are not answered by the same person, from the same budget, on the same evidence, or on the same timescale.

Two distinctions are worth making precise here, because they get blurred constantly.

The first is between water scarcity and water reliability. Scarcity is a physical condition. Reliability is a commercial one. Paks is not short of water in any meaningful physical sense; the Danube is still flowing past it. What the plant lacks is reliable access to that water under the conditions its equipment was specified for. Most industrial buyers do not, in practice, care much about regional scarcity statistics. They care intensely about whether their own intake works in August.

The second is between an engineering failure and a specification failure. Nothing at Paks broke. Every component performs as designed. The specification was written against a world that has since changed. These require completely different responses, and confusing them wastes years. You cannot maintain your way out of a specification problem.

That reclassification changes all four things that determine whether a deal closes.

The buyer. From the efficiency or environmental engineer to the operations director, plant manager or CFO who owns production continuity. Larger budget, shorter tolerance for ambiguity, no particular interest in your technology’s elegance.

The value argument. From cost saved to output protected. Payback period is the wrong frame for a continuity purchase; the relevant comparison is the cost of the solution against the cost of stopped production, a breached supply contract, or a site that cannot be permitted to expand.

The proof. Performance data under normal conditions stops being sufficient. What a continuity buyer needs is evidence of behaviour under stress: the dry year, the reduced allocation, the degraded input. That evidence is harder, slower and more expensive to generate, and very few suppliers hold it.

The urgency. From “sensible, add it to the five-year plan” to something with a date attached: a licence renewal, an allocation cut, a permit condition, an expansion that cannot proceed without secured supply.

A company still running an efficiency pitch into a continuity market will keep having excellent technical meetings that do not close. The technology is not the problem. The argument is aimed at a buyer who is no longer the one deciding, and no amount of additional sales activity corrects for that.

Deadlines, not drought maps

Concern does not produce procurement. Deadlines do, and this is where most market maps for water technology go wrong.

The United States is arriving at the same shift from the opposite direction, through regulation rather than physics. The rules governing the Colorado River, the 2007 Interim Guidelines and the 2019 Drought Contingency Plans, expire at the end of 2026. Reporting on the federal replacement plan released on 31 July points to substantial reductions for California, Arizona and Nevada. Those specific figures come from press coverage rather than from the document itself, so they should be treated as indicative rather than final. But the replacement process drew more than 18,000 public submissions during its comment period, which is a reasonable measure of how many parties understand their allocation assumptions are about to be rewritten in a legally binding way. Everyone downstream of that decision now has a date in their calendar.

Europe has the opposite condition. The July attribution study on the European drought notes that drought management plans remain non-mandatory across EU member states, leaving preparedness uneven. The physical problem is at least as severe, since the reactors are visibly stopping, but obligation has not been assigned, so in most jurisdictions there is no date, and without a date there is no procurement.

For anyone selling, financing or building water technology, this is the real market signal, and it is not the one most are watching. The addressable market is not “regions experiencing drought.” Drought maps make poor sales territory. It is the narrower set of places where a legal or regulatory deadline has converted physical stress into an obligation with an accountable party and a date attached. Those two maps overlap, but they are not the same map, and the difference determines whether a pipeline converts this year or stays warm for five.

The part that should reassure you

None of this is a counsel of despair, and the optimism here is specific rather than decorative.

The fix at Paks is already under way. MVM states that engineering work to lower the intake pipes has begun, and that within a few years the plant will be able to generate at levels below today’s. The problem is a pipe at the wrong elevation. That is a solvable problem, and it is being solved. In the interim the plant operates a four-stage low-water protocol and, with the national water directorate, has deployed emergency pumps to keep the shut-down reactors cooled, which requires around 2.5 cubic metres per second against roughly 100 per second when the plant is generating. Continuous upgrading over forty years had already allowed Paks to operate 120 cm below its original design level. The current situation is not a failure of that effort; it is the point at which the effort has to go further.

And at city scale, this transition has already been completed somewhere. Perth has been living in everyone else’s future for about twenty years and never sent a postcard. Its surface storages sat at 36.4 percent of accessible capacity at the end of June, a level that would trigger emergency measures across most of Europe, and the city functions normally, because it stopped depending on rainfall decades ago and rebuilt its supply around desalination, groundwater and recycling. Another desalination plant is under construction. Adelaide, after its lowest rainfall since 2006, increased desalinated production in January 2025 and avoided restrictions entirely.

That is not a scenario or a pilot. It is a completed transition, at city scale, under hydrological conditions worse than most of Europe currently faces. The engineering is proven, the operating economics are understood, and the institutional arrangements exist and function.

Which tells us something precise about the problem in front of us. It is not primarily a technology problem, since the technology largely exists and is in commercial operation somewhere. It is a commercialization problem: sequencing, financing structure, procurement design, risk allocation between parties who each control one piece of a system nobody controls in full, and getting the right evidence in front of the right buyer before the deadline rather than after it.

Those are solvable. They are also exactly what gets skipped when a sector assumes its historical baseline still holds, because if the baseline holds there is no urgency, and without urgency nobody redesigns anything.

The shift

The water business spent a generation selling efficiency to customers who had enough water. The next decade will sell continuity to customers who do not. In many cases the technologies are the same. The buyer, the argument, the proof and the clock are not.

This is a genuine commercial opening, and it is larger than the sector’s current self-image. Continuity buyers have bigger budgets than efficiency buyers, because they are pricing against production loss rather than a utility bill. But that budget is only accessible to suppliers who have made the translation: who can talk about output protection rather than cubic metres saved, who hold stress-condition evidence rather than nameplate performance data, and who have found the person with continuity in their objectives rather than the engineer with efficiency in theirs.

The companies that struggle over the next few years will not be the ones with weaker technology. They will be the ones who noticed late that the buyer changed.

Somewhere in most infrastructure business cases there is a number describing how much water will be there. At Paks it was a hundred-year minimum, and the Danube is now more than a metre below it. It is worth finding out what your equivalent number is, who calculated it, and when anyone last asked whether it still describes the world.

---

Status figures were current at 09:00 CET on 3 August 2026 and the situation was moving daily. The commercial argument is the author's analysis, not a claim made by any source cited.