THE RICE AND THE WAFER

What happens to the one substrate that cannot be moved

Article 4 of 8  ·  Series II of III  ·  Published 4 November 2026  ·  Analysis → Forecast → Recommendations

Series II, fourth row. Each article measures one physical layer with the same three questions — concentration, criticality, substitution time — and adds a row to one comparison table. The first two rows found dependencies with no practical alternative. The third found real alternatives and a permanent price. This one breaks the instrument, because the substitution question has no engineering answer at all.

 1. The Year the Fabs Drank the Rice Fields

In 2020, for the first time in fifty-six years, no typhoon made landfall on Taiwan.

That reads like good fortune. It was the opening of the island's worst drought since 1964, because roughly seventy percent of Taiwan's rainfall arrives with typhoons. Taiwan is not short of water in any ordinary sense. It is short of storage, and it depends on a handful of very large weather events to fill what storage it has. When those do not arrive, the reservoirs do not refill, and there is no second mechanism.

By the spring of 2021, several reservoirs serving central and southern Taiwan were below five percent of capacity. The government did three things, in this order.

—  It reduced supply to households and commercial users in the affected regions, cutting to five days a week in some areas.

—  It paid farmers across roughly 74,000 hectares not to plant — a compensated suspension of the rice crop in the middle of a growing season.

—  It trucked water to the semiconductor fabrication plants in the Hsinchu, Taichung and Tainan science parks, and kept them running.

The chips kept shipping. The fabs took an efficiency hit and spent money on tankers and on recycling capacity they had not planned to build that year, and global supply of advanced semiconductors was not meaningfully interrupted.

It is worth being precise. Taiwan did not solve a water shortage in 2021. It decided who would go without.

The decision was defensible on almost any economic reading. Those fabs produce a large share of the world's leading-edge logic; the rice was worth a fraction of that and was compensated. But the mechanism is the point, and the mechanism is not an engineering one. There was no additional water. There was a reallocation, arrived at politically, with a named party on the losing side of it.

That is a different kind of constraint from the three already on the table. A grid connection queue is slow but impersonal; nobody is deprived so that you can connect. A reallocation of water has a counterparty who knows their name is on it.

Five years on, Taiwan has not returned to comfort. Hsinchu spent the early part of 2026 under a water alert, with the economics ministry stating an objective of avoiding industrial cuts before June; the alert was downgraded from the pressure-reducing level to conservation at the end of April, after the working group's third meeting on the matter. Two things follow. The management is now continuous rather than episodic, and the goal has been reframed from adequacy to the avoidance of industrial curtailment specifically.

2. What a Water Constraint Actually Does

The intuitive model of water scarcity is a drought: rainfall falls, supply falls, everyone gets less. That model is almost useless here, because industrial water constraints rarely present as a shortage of water. They present as a competition for an allocation.

The volumes, so the scale is not in dispute

A leading-edge fabrication plant consumes water on a municipal scale. TSMC used approximately 101 million cubic metres in 2023; Samsung's semiconductor operations have been reported at around 344,000 tonnes per day, and microchips are South Korea's largest export by value — the relevant political fact rather than an incidental one. Industry forecasts put semiconductor water demand on course to roughly double by 2035, near eight percent annually, against best-in-class recycling rates already above seventy percent that will have to climb further between 2030 and 2035 simply to stay level.

Running the instrument

That third cell is why this row is worth its own article. In the first three rows the substitution time was a number — four to seven years for a grid connection, fifty-two to seventy-eight weeks for advanced packaging, days to weeks for a maritime reroute. Here it depends on who is asked to give something up, which is not forecastable from the physical data.

Two mechanisms, often conflated

There are two ways out of a water constraint for a large industrial user, and they are usually discussed as though they were one.

The first is intensity reduction: recycle more, evaporate less, switch to closed-loop or dry cooling. This is real, happening and measurable. SK Hynix reported cutting daily consumption by 170,000 tonnes in 2025 and reusing 73.59 million tonnes across the year, up fifty-four percent from 47.88 million in 2022.

The second is reallocation: someone else uses less so you can use more. Taiwan did this in 2021. It is faster than the first mechanism and considerably cheaper for the party gaining the water, which is why it keeps being chosen under time pressure.

The confusion between them matters because they carry opposite risk profiles. Intensity reduction is capital expenditure with a known payback and no counterparty. Reallocation is a political transaction whose durability depends on the loser continuing to accept the outcome. A company that has solved its water problem by the first route has a solved problem. One that has solved it by the second has a standing arrangement.

3. Subtheme One — The Number Everyone Measures Is the Small One

Almost every public discussion of data-centre water measures the wrong quantity, and the error is not marginal. What gets measured is on-site consumption: water evaporated by cooling towers inside the facility fence. United States data centres consumed approximately 17.4 billion gallons that way in 2023, and it is the figure that appears in sustainability reports, permitting hearings and most journalism, because it is the one the operator can count.

Water consumed to generate the electricity those same facilities used in that year was approximately 211 billion gallons.

The indirect footprint is roughly twelve times the direct one. The figure the operator publishes, the regulator reviews and the community debates is about eight percent of the total.

This is not an accounting quibble. Thermoelectric generation — coal, gas, nuclear — consumes water for cooling, and far more of it per unit of output than a server hall does. That water is consumed at the power plant, in whatever basin the plant occupies, under whatever permit the utility holds, and nobody in the data centre's own permitting process is required to look at it.

Why this connects row four to row one

Series II opened with the electricity grid: a substrate with no alternative path, connection queues of four to seven years, and waiting times roughly doubled in fifteen years. This row attaches to it from an unexpected direction.

If eleven-twelfths of a facility's water footprint sits in the generation mix, its water intensity is mostly a property of its grid rather than of its cooling design. And grid water intensity varies enormously — estimates put consumption at roughly 2.1 gallons per kilowatt-hour in the hydro-heavy Pacific Northwest against about 0.13 in solar-heavy California, a spread of more than fifteen times driven by generation mix rather than by anything the facility does.

The consequence is uncomfortable for the way these decisions are currently made. A facility's water footprint is settled mainly at the siting stage, by which grid it joins. Yet siting is optimised for power, land, latency and tax treatment, while water is argued about afterwards, in terms of the eight percent.

The figure that makes the category look smaller than the argument

One more number belongs here, because leaving it out would be dishonest. Data centres account for less than one percent of total United States water consumption; agriculture and thermoelectric generation dominate the national picture by an enormous margin, and always have. That is true and it settles nothing, because water is a basin-level resource and a national average describes no basin. Nearly two thirds of new United States data centres built since 2022 have gone into regions of high water stress, and a facility drawing an unremarkable national share can be the largest new industrial withdrawal in a county whose aquifer is already in decline. Both the industry statement and the local objection are accurate at their own scale — the same structure as the Hormuz vocabulary problem in the previous article, and the same resolution. Ask which basin, not which country.

4. Subtheme Two — Building Where the Water Is Least Certain

The most interesting fact about industrial water risk is that the industries most exposed to it keep choosing to increase their exposure, with full knowledge of the data.

Roughly forty percent of existing semiconductor fabrication plants sit in basins projected to face high or extremely high water stress by 2030. Among plants announced since 2021 — after the Taiwan drought, after the supply-chain disruptions of 2020 and 2021, during a period of intense public attention to semiconductor resilience — the share is slightly higher. The industry did not fail to notice. It optimised for other things, and water lost.

Why water loses the siting decision

A fab or a large data centre is sited against power availability, skilled labour, supplier clusters, land, political stability, subsidy and customer proximity. Water is on that list, and it differs from every other item on it in one specific respect.

Every other constraint is adequate or inadequate at the moment of the decision, and the answer is knowable. Water is almost always adequate at the moment of the decision, and uncertain across the twenty-to-thirty-year life of the asset. Present adequacy is verifiable; future scarcity is a projection with error bars. A committee comparing a verifiable yes against a probabilistic maybe will discount the maybe every time, without anyone in the room behaving unreasonably.

There is a second reason, less often stated. Water is the constraint most likely to be resolved in the builder's favour after the fact, because reallocation exists. No company can obtain a grid connection five years early or conjure a second packaging supplier. Many can reasonably expect that if water tightens, a large employer with a nationally strategic product will not be the user asked to stop. Taiwan 2021 established that expectation.

The Yongin arithmetic

South Korea's Yongin semiconductor cluster is the clearest available illustration, because the numbers on both sides have been published. SK Hynix's conservation record is genuinely strong: 303 million tonnes of cumulative savings across the eight years from 2018 to 2025. The Yongin cluster at full build-out is projected to require approximately 390 million cubic metres per year.

One year of the new cluster would consume more water than the company saved in eight years of company-wide conservation.

The supply plan shows which mechanism is actually being relied on. Near term, around 200,000 tonnes per day from recycled wastewater together with surplus from the Paldang Dam, on a timeline pointing at approximately 2031. Longer term, up to 600,000 tonnes per day from a converted Hwacheon Dam.

Converting a dam is a reallocation. Whatever that reservoir currently supports — downstream agriculture, municipal supply, an ecological flow requirement, hydroelectric output — is repurposed toward industrial use. This may be entirely sound on the merits. It is not an efficiency gain, it depends on a political settlement holding for decades, and it sits on the critical path of a cluster central to the country's export economy.

The pattern generalises

Efficiency improves, per-unit intensity falls, aggregate withdrawal climbs anyway, and the gap is closed by reallocation. Taiwan closed it with rice fields in 2021. Korea plans to close it with a dam by the mid-2030s. In the United States it is being closed, or refused, one county at a time: as of late September 2026, twenty-seven states were restricting or considering restrictions on data-centre development, twenty had active bans or moratoriums, and more than three hundred local moratoriums were in place. Water is one driver among several — electricity prices, land use and noise all appear — but it recurs consistently, and in the arid Southwest and the Central High Plains it is frequently the leading one.

That is what reallocation looks like when the losing party has a vote: slower, noisier, less predictable than a ministerial decision, and it produces a different outcome — not curtailment of the industrial user, but refusal at the siting stage.

5. What Most Analysis Gets Wrong

Three errors recur, and each changes a decision.

—  Treating scarcity as the trigger. Physical scarcity is not what stops an industrial user; allocation decisions are. Taiwan in 2021 had very little water and its fabs ran, while a temperate basin with adequate rainfall can halt a project through permitting refusal. The operative variable is the allocation process, not the hydrology.

—  Reading efficiency gains as solutions. Recycling and closed-loop cooling are real, deployed at scale, and create no new source. Where demand growth exceeds efficiency growth they postpone the allocation question rather than removing it. A seventy-percent recycling rate alongside a threefold expansion improves a company's position in the argument; it does not reduce its exposure to one.

—  Averaging across basins. Water does not aggregate meaningfully above the basin level, so any national figure is context rather than evidence about a site.

6. Base, Stress and Extreme

Three ways the next decade runs, so the forecasts below have a frame.

Base — continuous management, local refusals

Efficiency improves steadily and is outpaced by volume. Reallocations happen case by case: compensated agricultural curtailment where the state can direct it, dam and wastewater reallocation where the infrastructure exists, project refusal where local government controls permitting. No leading-edge fab is curtailed in a way that reaches global chip supply. Water becomes a routine siting constraint priced into development, adding cost and schedule rather than blocking the category. The most probable path, and it describes 2026 reasonably well already.

Stress — a named industrial curtailment

A drought coincides with high industrial demand in a basin where the political balance does not favour the industrial user — a large agricultural constituency, a badly timed election, or a previous reallocation still resented. An operating fab or a cluster of data centres takes an actual allocation cut with production effects. The consequence is not a chip shortage; it is the removal of the assumption that strategic industry is exempt, which would reprice every siting decision made since 2021.

Extreme — multi-year regional failure

A basin hosting concentrated advanced manufacturing suffers consecutive years of failed replenishment, exhausting both storage and the political room for reallocation. Taiwan is the obvious candidate. Effects reach global semiconductor supply, and the substitution problem becomes the one from this series' second row: no alternative capacity at the leading edge, and years to build it. Low-probability, and the case that would make water the binding constraint on the whole technology stack.

7. Forecast — One Year, to end-2027

Direction. At least one large announced project in the United States is publicly delayed, relocated or cancelled with water named by the company or the regulator as a reason. Probability 0.65. Confidence: Medium.

Three hundred local moratoriums and twenty states with active bans are a mechanism already in operation, and the share of new facilities sited in high-stress regions means the two collide continuously. What makes this resolvable rather than merely likely is the requirement that water be named in a company's or regulator's own statement — a higher bar than commentary attributing a cancellation to water afterwards.

Second-order effect. One clearly attributed cancellation changes diligence practice across the sector faster than any projection has. Water risk currently sits in sustainability reporting; a named cancellation moves it into site selection, where it competes with power and land on equal terms.

What weakens it. Companies have strong incentives to attribute cancellations to market conditions, permitting delay or customer demand rather than to a resource constraint that implies poor planning. The claim could be true in substance and fail on attribution.

8. Forecast — Three Years, to 2029

Direction. The Yongin cluster water supply plan is formally revised — either the Hwacheon Dam conversion receives approval, or the short-term supply schedule slips beyond 2031 in an official planning document. Probability 0.60. Confidence: Medium.

A plan whose first tranche depends on recycled wastewater plus dam surplus by approximately 2031, and whose second on converting a dam currently doing something else, carries two large uncertainties on a critical path. Korean infrastructure planning of this kind is documented publicly, which makes a revision observable either way. The forecast is deliberately two-sided — approval and slippage both resolve it yes — because the claim is that the plan as stated will not survive contact with the engineering or the politics unchanged.

Second-order effect. Yongin is the most-watched greenfield semiconductor cluster outside Taiwan and the United States. If its water plan needs revision, water moves up the list of things scrutinised in every subsequent cluster announcement, including those already committed under industrial policy elsewhere.

What weakens it. Large state-backed industrial projects in Korea have a strong record of delivering infrastructure on the announced schedule when the export economy depends on them. The plan may simply proceed, and the dam conversion may be approved quietly enough that no revision is recorded.

9. Forecast — Five Years, to 2031

Direction. At least three United States states have in force a statutory or regulatory requirement that new large data centres disclose or cap water withdrawal as a condition of permitting. Probability 0.60. Confidence: Medium.

Twenty-seven states are already restricting or considering restrictions, and California has added water and land-use requirements to environmental review. Disclosure is the characteristic first regulatory step here: cheaper politically than a cap, defensible as transparency, and a precondition for any later cap. Five years allows a legislative cycle plus rulemaking in states now at the consideration stage.

Second-order effect. Mandatory disclosure would surface the measurement problem in subtheme one. If a statute requires reporting of on-site withdrawal only, it will formalise the eight-percent metric as the regulated number and entrench a comparison that is mostly meaningless. Which quantity the first statutes choose to define matters more than the caps that follow.

What weakens it. Data centre investment is actively courted for its tax base and its association with national competitiveness in artificial intelligence. State-level pressure could run the other way, pre-empting local moratoriums rather than adding requirements, in which case restriction concentrates at county level and never reaches statute.

10. Forecast — Ten Years, to 2036

Direction. Taiwan imposes industrial water curtailment on the Hsinchu, Taichung or Tainan science parks in at least three separate calendar years between 2027 and 2036. Probability 0.70. Confidence: Medium-High.

The structural facts support this more strongly than any other forecast here. Typhoon dependence for roughly seventy percent of rainfall, limited storage, no alternative filling mechanism, and industrial demand growing near eight percent annually against a supply system already managed continuously. The 2026 alert sequence in Hsinchu, and the explicit ministerial objective of avoiding industrial cuts before June, show that curtailment is now the routine thing being actively averted rather than an unthinkable one. Three years in ten is modest against that background.

Second-order effect. Repeated curtailment does not primarily threaten output; the fabs absorbed 2021. It changes where the next generation of capacity is built. A decade of managed shortage is a stronger argument for geographic diversification of leading-edge manufacturing than any policy programme has yet provided, and it pushes in the same direction as the industrial policies of three separate blocs.

What weakens it. Taiwan is investing in desalination, reclaimed-water plants and inter-basin transfer specifically to protect the science parks, and the fabs continue to raise recycling rates. A decade is long enough for that programme to substantially work. Typhoon frequency is also variable rather than trending cleanly, and a run of ordinary years would postpone the question.

11. Signals to Watch

Five observable items, each published, each moving before the outcomes above.

—  Taiwan Water Resources Agency alert levels for the three science-park regions — published, colour-coded, and the earliest indicator of curtailment pressure. The move to the pressure-reducing level is the one that matters.

—  Reservoir percentage-of-capacity for the basins serving Hsinchu, Taichung and Tainan, read at the end of the dry season. The annual average hides the relevant variable entirely.

—  Korean planning documents on Yongin supply: the schedule for the recycled-wastewater tranche, and any movement on the Hwacheon conversion.

—  The count of United States local moratoriums and its composition. A rising share citing water rather than electricity price or land use would mean the constraint is migrating from cost to resource.

—  Corporate recycling rates read against absolute withdrawal in the same report. A rising rate alongside rising total withdrawal is the signature of the gap in subtheme two, and both figures are usually disclosed.

Read together: the first two measure physical pressure, the next two the political capacity to reallocate. A water constraint binds only when both move at once — pressure rising while the room for reallocation narrows. Either alone has been absorbed repeatedly.

12. Recommendations — Individuals

For most people this substrate arrives as a local political question long before it arrives as anything else, and that is where the useful attention sits.

Immediate — 30 days

If a large facility is proposed near you and water is part of the argument, establish which number is being discussed. On-site withdrawal is roughly a twelfth of the footprint, and a figure quoted without that distinction tells you little either way. Ask what the facility's grid mix is, because that sets the other eleven-twelfths.

Build — 12 months

Learn the state of your own basin rather than your country's. Most regions publish reservoir levels, aquifer trends or groundwater-district status, and what matters is the trend across dry seasons rather than the annual total. It is the one piece of water information with direct bearing on where you live, and it is widely available and widely unread.

Position — 3 years

For a location decision on a multi-decade horizon — property, where to settle, where a business will sit — treat long-run basin trend as a real input. The question is not whether water is available now, which it almost always is, but whether the basin is already allocating under pressure. A basin managing scarcity continuously will keep making allocation decisions, and residential users are rarely first in line for protection.

Avoid. Reading national water statistics as information about your situation. Less than one percent of national consumption and the largest new withdrawal in your county describe the same facility, and only the second affects you.

Why this works. Water is the most local substrate in this series. Everything that matters about it is decided at basin and county level — the level at which an individual has both the best information and the most influence.

13. Recommendations — Business

For water-intensive operations this is where exposure is most often mismeasured, and the correction is cheap.

Immediate — 60 days

Calculate your indirect water footprint alongside your direct one. Take your electricity consumption, apply the water intensity of your grid region, and compare it with your reported on-site figure. For most operations the indirect number is several times larger. This takes an afternoon and changes which lever is worth pulling.

Then identify which basin each significant site draws from, and what the allocation process is there. Not the country, not the state — the basin, and the body that decides who gets what when it is short.

Build — 12 months

Establish your position in your basin's allocation order, in writing where that is possible. Which users hold senior rights, where does your permit sit, and what happened in that basin the last two times supply was constrained? Firms typically learn their position during the shortage, which is the most expensive moment to learn it.

Treat grid mix as a water decision as well as an emissions one. A power purchase agreement with low-water generation improves your total footprint by more than any plausible cooling retrofit, and the two projects are usually owned by different functions who do not compare notes.

Position — 3 years

Move water from environmental reporting into site selection, with the same standing as power availability. The forty-percent figure for fabs in high-stress basins exists because water was reviewed after the site was chosen, and any process where water appears in compliance review rather than in the shortlist criteria will reproduce it.

And distinguish the two mechanisms in your own plan. If your water strategy depends on an allocation someone else gives up, it is a political arrangement with a renewal risk and should be monitored as one rather than logged as solved.

Avoid. Reporting a recycling rate as a risk position. A rate is an efficiency measure. The risk position is absolute withdrawal against basin availability and your seniority in the allocation order, and a rising rate is fully compatible with a deteriorating position.

Why this works. The two corrections here — measuring the indirect footprint and locating yourself in the basin allocation order — cost very little and address the two errors that produce nearly all the surprises in this substrate.

14. Recommendations — Capital

Water risk is currently disclosed in a form that makes companies difficult to compare, and the gap between disclosure and exposure is where the mispricing sits.

Immediate — this quarter

For water-intensive holdings, establish basin-level exposure rather than reading the sustainability disclosure: which basins, what the allocation process is, and whether the water plan relies on efficiency or on a reallocation from another user. The second carries a political renewal risk that appears nowhere in a water-intensity metric.

Build — 12 months

Treat announced capacity in high-stress basins as carrying schedule and cost risk not usually in the model. The sector is siting above forty percent of announcements in stressed basins, permitting friction is rising in the United States on a documented count, and neither is typically reflected in a timeline drawn at announcement.

Watch the disclosure standard itself. If mandatory reporting arrives defining on-site withdrawal as the regulated quantity, it entrenches a metric covering roughly eight percent of the footprint. Companies whose advantage lies in grid mix rather than cooling design would be systematically undercredited, and a disclosure regime that measures the wrong thing produces comparisons that look rigorous and are not.

Position — 3 years

The durable observation is that this constraint is being resolved by reallocation, and reallocation requires infrastructure: reclaimed-water treatment, desalination, inter-basin transfer, industrial recycling. Those assets get built whichever way the politics resolves, because both routes pass through them. That is an observation about where capital is directed under this constraint, not a recommendation about any instrument.

Avoid. Pricing water risk as drought risk. Drought is weather and it is cyclical. The exposure here is an allocation decision, it is political, and it can bind in a basin with entirely normal rainfall through a permitting refusal.

Why this works. Basin-level allocation processes, permitting dockets and state planning documents are public, slow-moving and read almost exclusively by specialists, which makes them one of the better-documented and least-priced categories of physical risk currently available.

15. What Would Change Our Mind

Each forecast carries its own weakening condition. Three developments would undermine the article as a whole.

—  Efficiency overtakes volume. If recycling and closed-loop cooling improve fast enough that absolute withdrawal falls while capacity grows, the reallocation mechanism becomes unnecessary and this article's central argument stops applying.

—  A water constraint binds through physical scarcity rather than allocation. Our claim is that allocation decisions, not hydrology, determine who stops; a case where the allocation held and the water was simply absent would mean we have the mechanism wrong.

—  The indirect footprint turns out to be the wrong frame. The twelve-to-one ratio rests on estimates that vary considerably between sources; if the better ones converge near parity with direct use, subtheme one loses most of its force.

Jurisdiction and sourcing, recorded on the fourth row. The material divides between Taiwan, South Korea and the United States, which continues the improvement in spread but leaves a real gap: almost nothing on Europe, India or the Gulf, where industrial water politics differ substantially and desalination changes the substitution arithmetic. The sourcing weakness is different again from the first three rows. Water-stress projections are modelled rather than measured, the twelve-to-one indirect ratio comes from estimates that disagree by a factor of two, and the moratorium count comes from an advocacy-adjacent tracker rather than an official register. Each is labelled accordingly below, and the modelled figures should be read as ranges.

The Series II Table — four rows

What four rows show that three could not. The instrument has a blind spot, and this row found it. Concentration and criticality are physical questions with physical answers, and substitution time was a physical question in the first three rows too — years, weeks, days. Here it is not, and for the first time in this series the answer depends on who a government is willing to disappoint.

That matters beyond water, because it suggests the three questions measure engineering substitutability rather than actual substitutability. Any substrate where the alternative exists but belongs to someone else will read as more substitutable than it is. The remaining rows should be tested for that, not only for concentration.

The next row is refined materials.

16. Bottom Line

In 2021 Taiwan paid its farmers not to plant rice across roughly 74,000 hectares and trucked water to its semiconductor plants. The chips shipped. Nothing anywhere was missing, and the episode is remembered, when it is remembered at all, as a supply chain that held.

It held because a government decided who would go without.

That is the substance of this row. Water stays in its basin, so there is no substitution in the sense the previous three articles used the word — only using less, which takes years, or taking it from someone else, which takes as long as the politics takes and leaves a counterparty who knows their name is on the decision.

Efficiency is real and it is losing. SK Hynix cut 170,000 tonnes a day and saved 303 million tonnes across eight years — a serious engineering achievement. One year of the Yongin cluster at full build would consume more than all of it, and the plan to cover the gap involves converting a dam, which is a reallocation from whatever that dam currently supports.

Meanwhile the industry keeps building where the water is least reliable. Over forty percent of fabs announced since 2021 sit in basins projected to be highly stressed by 2030, and nearly two thirds of new United States data centres since 2022 have gone into high-stress regions. This is not oversight. Water is the only constraint reliably adequate on the day the decision is made and uncertain across the life of the asset, and a committee will discount an uncertainty against a verified yes every time.

And the number in every report is the wrong one. On-site cooling is about 17 billion gallons a year for United States data centres; generating their electricity consumes about 211 billion. A twelfth of the footprint is what gets measured, debated and regulated. The rest belongs to the grid mix, varies more than fifteen-fold by region, and is settled at the siting stage by people optimising for power and land.

Which is why the first three rows read as physical problems and this one does not. Grid queues, packaging capacity and shipping lanes have engineering answers with numbers attached. Here the substitution time is the time it takes to decide whose allocation moves — worth noticing, because any substrate whose alternative belongs to someone else will look more substitutable than it is.

For anyone with a decision in front of them the practical content is short. Measure the indirect footprint, which is most of it. Ask which basin, never which country. And know whether your water plan rests on using less or on someone else using less, because only one of those is finished.

Forecast record

Four forecasts, one per horizon, each with a threshold, a named source that will settle it, and a date fixed before the answer is known.

Correlation. The first and third share a parent cause in United States water politics: if local resistance stalls or state policy turns pre-emptive, both weaken together, and they are one family measured at two horizons. The second resolves on Korean infrastructure planning and the fourth on Taiwanese hydrology and allocation, and neither depends on the American pair or on each other. So this set contains three families across four forecasts, which is better independence than the previous row and is recorded either way.

Directional statements elsewhere in this article carry no threshold and are excluded from the record.

Sources

THRIVE IN CHAOS

Signal Over Noise  ·  thriveinchaos.ai

AI intelligence system with human editorial oversight. Analytical material, not investment, legal or tax advice.