What happens when the infrastructure of the future depends on a resource that cannot simply be moved somewhere else?

In 2021, Taiwan faced an unusually difficult choice.

After a year without a single typhoon making landfall, the island entered its worst drought since 1964. Reservoirs serving important industrial regions fell to critically low levels, threatening households, agriculture and some of the world's most strategically important semiconductor factories.

The government compensated farmers for leaving approximately 74,000 hectares unplanted. Water deliveries to semiconductor manufacturing facilities were protected, and factories continued operating.

The global chip supply chain avoided another major disruption.

From the perspective of international technology markets, the response was successful. But that success concealed an important distinction.

Taiwan had not eliminated its water shortage. It had decided which users would absorb its consequences.

That decision introduces a different kind of infrastructure constraint into our Series II analysis.

Our previous articles examined electricity, frontier computing and maritime chokepoints.

Each revealed a different relationship between concentration, criticality and substitution time.

Electricity is difficult to replace because new grid connections and essential equipment can take years to deliver.

Frontier computing is constrained by highly concentrated manufacturing processes, including advanced packaging and high-bandwidth memory.

Maritime shipping is comparatively flexible. Ships can change routes relatively quickly, although doing so increases costs and reduces effective transport capacity.

Water does not fit neatly into this framework.

It can be transported over limited distances, recycled, stored and, in suitable locations, produced through desalination. But these options cannot always provide additional supply quickly or economically to an inland industrial region.

When a watershed experiences an immediate shortage, the most accessible alternative is frequently a change in how existing water is distributed.

That makes substitution a question of competing claims on a shared resource.

And it means that engineering capacity alone cannot determine when an alternative becomes available.

The semiconductor industry is making considerable progress in reducing water consumption per unit of output.

SK Hynix reported cumulative water savings of approximately 303 million tonnes between 2018 and 2025, alongside substantial increases in reuse.

These improvements are real and important.

But they need to be compared with the scale of new industrial demand.

South Korea's planned Yongin semiconductor cluster could require approximately 390 million cubic metres of water annually at full development.

One year of consumption from the completed cluster would therefore exceed eight years of reported company-wide savings.

The proposed supply plan includes recycled wastewater, existing reservoir surplus and the potential conversion of a dam to support industrial demand.

These measures address different parts of the problem.

Recycling improves efficiency. New infrastructure can expand usable supply. Changing the purpose of an existing reservoir reallocates water between competing users.

A company relying on the first approach faces primarily engineering and capital-expenditure risks.

A company relying on the third also faces the possibility that its allocation arrangements will be reconsidered in the future.

The distinction matters because both approaches can appear identical in a sustainability report.

Data centres introduce another problem: much of their water footprint occurs outside the facilities themselves.

Public debate usually concentrates on cooling systems because their consumption is relatively easy to measure and report.

But the electricity required to operate those facilities also has a water footprint.

Estimates examined in our latest article suggest that US data centres consumed approximately 17.4 billion gallons of water directly in 2023.

Generating their electricity required an estimated 211 billion gallons.

The indirect footprint was therefore roughly twelve times larger than direct consumption.

These figures are estimates, and different methodologies produce materially different results. They should not be treated as a universal ratio for individual facilities.

Nevertheless, the underlying mechanism is important.

A data centre's total water footprint depends substantially on the electricity system supplying it.

Electricity generated using water-intensive cooling technologies can produce a much larger indirect footprint than electricity generated using low-water technologies such as wind and solar photovoltaics.

This creates a connection between our first and fourth Series II articles.

Electricity availability determines whether a data centre can operate.

Electricity generation also influences how much water is required to support that operation.

The two constraints are not independent.

Data centres account for a relatively small share of total US water consumption.

At the national level, agriculture and electricity generation dominate the picture.

But water is fundamentally a local resource.

A facility representing a negligible share of national consumption can still be a major new industrial user in a watershed experiencing declining groundwater levels or repeated drought.

Both observations can be true simultaneously.

This is why national water-consumption statistics are a poor substitute for watershed-level analysis.

The relevant questions concern local supply, competing users, seasonal availability, existing allocation rights and long-term changes in demand.

The distinction becomes particularly important when deciding where to build infrastructure intended to operate for several decades.

The semiconductor industry illustrates this problem particularly clearly.

Research examined in the article suggests that approximately 40% of existing semiconductor fabrication facilities are located in watersheds projected to experience high or extremely high water stress by 2030.

The share among facilities announced since 2021 is slightly higher.

These are modelled projections rather than observations of future shortages. But they expose a recurring difficulty in industrial planning.

A new factory needs electricity, specialised suppliers, skilled workers, suitable land and access to customers.

Most of these conditions can be assessed relatively clearly when the investment decision is made.

Water may also be adequate at that moment.

The difference is that water availability across the next twenty or thirty years is much less certain.

Industrial planners therefore face a choice between measurable advantages today and uncertain resource constraints decades into the future.

The danger is not necessarily that a project will run out of water.

It is that maintaining its supply may eventually require expensive infrastructure, restrictions on other users or a renegotiation of existing allocation arrangements.

Our article establishes four forecasts with defined dates and measurable resolution criteria.

One year · 31 December 2027

US project disruption

At least one announced US semiconductor fabrication plant or data-centre project valued at $1 billion or more is delayed, relocated or cancelled, with water availability or permitting officially identified as a reason.

Three years · 31 December 2029

Yongin water-supply revision

South Korea's Yongin water-supply plan is formally revised through approval of the Hwacheon Dam conversion or a documented delay to the short-term supply schedule beyond 2031.

Five years · 31 December 2031

US regulatory requirements

At least three US states introduce enforceable requirements for large new data centres to disclose or limit water withdrawal as a condition of permitting.

Ten years · 31 December 2036

Repeated Taiwanese restrictions

Taiwan imposes industrial water curtailment affecting at least one of its three major science-park regions in three separate calendar years between 2027 and 2036.

These probabilities are the article's analytical estimates, with medium to medium-high confidence, not observed outcomes.

The forecasts are designed to test different aspects of the same structural constraint: industrial development, infrastructure planning, regulation and repeated resource stress.

For individuals, the most useful water information is local.

Anyone considering a long-term property purchase or relocation should examine reservoir trends, groundwater conditions and existing allocation pressures in the relevant watershed.

For businesses, water should be assessed during site selection, alongside electricity, workforce and logistics.

A water strategy should also distinguish between efficiency improvements and dependence on an allocation that could be reconsidered.

For capital, the key is to evaluate absolute consumption and watershed exposure rather than relying exclusively on corporate recycling rates.

A company can report excellent efficiency improvements while becoming increasingly dependent on a limited local resource.

The same applies to indirect consumption. A data centre with efficient cooling may still have a substantial water footprint if its electricity comes from water-intensive generation.

The central argument would weaken if improvements in recycling and cooling consistently reduced absolute water withdrawals even as industrial production expanded.

It would also require reconsideration if new supply infrastructure substantially reduced dependence on reallocating existing water resources.

The indirect water-footprint estimates are another important uncertainty. Their scale varies between methodologies, so better measurement could materially change how the burden is distributed between electricity generation and on-site cooling.

These are the developments to monitor, rather than treating every drought or new industrial project as evidence of an inevitable shortage.

With four articles completed, Series II now reveals a limitation in its original analytical instrument.

Concentration and criticality remain useful physical measures.

But substitution time is not always an engineering variable.

For electricity, advanced computing and maritime transport, alternatives can be assessed through construction schedules, manufacturing capacity or transport routes.

Water adds another dimension.

When the alternative depends on a resource currently allocated to someone else, technical feasibility does not guarantee practical availability.

That is why the lesson from Taiwan extends beyond its semiconductor industry.

The physical resource may be available. The question is whether the system can agree on who gets to use it.

And that agreement may ultimately determine whether an industrial project proceeds, how much it costs and how resilient it remains over the next several decades.

THRIVE IN CHAOS Decision Intelligence for an Uncertain World

Signal → Meaning → Action → Stability

Read the full Series II analysis at thriveinchaos.ai.

The forecasts above are analytical estimates, not certainties. This publication supports independent decision-making and is not investment, legal or tax advice.