in short
S&P Global estimated TSMC's 2022 water bill at roughly NT$1 to 2 billion, a small line item for the company that builds the world's most advanced chips. In Arizona, three planned fabs will draw about 16 million gallons of water a day gross, yet Phoenix expects to supply only about 4 million of it. Water is a small cost for a fab and a large constraint on where one can be built. That mismatch, a low price attached to a high-stakes resource, explains most of what is happening.
The cornerstone piece in this series described the volumes: how many millions of gallons a fab uses, how pure the water has to be, and why that collides with drought. This piece asks a different set of questions. Who actually pays for the water, why doesn't its price reflect how scarce it is, what changes when a fab recycles most of what it uses, and what happens when the allocation decision ends up in politics rather than in a market?
The answers matter because water is the one bottleneck in this series that is both cheap and capable of stopping a factory. That combination is unusual, and it is the reason the standard tools, more capital and more patience, don't resolve it cleanly.
Gross Versus Net: The Number That Changes the Story
The headline figures for Arizona are large. TSMC's first Phoenix fab consumes about 4.75 million gallons of water a day, and the three planned fabs are expected to need around 17 million gallons a day when all are running (Data Center Dynamics). TSMC's development agreement with the City of Phoenix gives it access to 11.4 million gallons a day, which could expand to 17.3 million to support the additional fabs (Real Estate Daily News).
Those are gross numbers, meaning total water flowing through the plant. The net figure, what the city actually has to supply after reuse, is a different story. Axios reported that when all three fabs are complete they are projected to use about 16.4 million gallons a day, but with recycling the city will have to provide only about 4.2 million (Axios). For the first fab alone, the new reclamation plant is expected to cut its net draw from 4.75 million gallons a day to under 1.2 million (Real Estate Daily News).
One caution on the net figure: these projections date from 2024 and 2025, before TSMC's latest expansion plans, so they are best read as a snapshot of the first three fabs rather than a forecast for the whole campus.
Think of a bathtub with a recirculating pump. The tub holds a lot of water and the water flows through it constantly, but if a pump returns most of it, you only have to top up what evaporates or leaks. The gross flow can be enormous while the net demand on the tap stays modest. The catch is that the pump has to be built, powered and maintained, and it can't recover everything.
The Gap Between Now and 2028
The bathtub's pump isn't fully installed yet. TSMC's industrial water reclamation plant, a 15-acre facility designed for near-zero liquid discharge, is expected to become operational in 2028 (DCD). Until then, the plant runs on its existing in-house recycling, which TSMC says covers about 65 percent of water use at startup, with the new facility designed to lift that to 90 percent or more (TSMC Arizona). Axios reports the plant is expected to reach 85 percent when it begins operating, with a plan to expand to 90 percent or higher (Axios).
Two details are worth noticing. First, the best case is 90 percent, not 100. A fab that recycles nine-tenths of its water still has to find the rest somewhere. Second, the engineering isn't copy-and-paste. The Phoenix plant is partly modeled on TSMC's facilities in Taiwan but had to be adapted because Phoenix water has higher levels of magnesium and calcium (Axios). Water chemistry is local, and recycling systems have to be tuned to it.
Why Water Is Cheap and Scarce at the Same Time
Here is the economic puzzle at the center of this piece. If water were truly scarce in the way a market understands scarcity, its price would climb until users economized or substituted. For fabs, that mechanism barely operates, because water is a small cost next to the factories that depend on it. S&P Global estimated TSMC's water cost at between NT$1.0 and NT$2.0 billion in 2022, based on its municipal water consumption and local tariffs (S&P Global Ratings). That is on the order of tens of millions of US dollars, for a company that has invested more than $65 billion in its three Phoenix fabs so far (DCD), with a $265 billion Arizona expansion now under way (BigGo Finance; AZFamily).
A low price means a weak signal. A fab will pay for water almost regardless of the rate, because the alternative, a halted production line, is far more expensive. And the public system supplying the water often has prices set by policy rather than scarcity. Commentators on Taiwan's water problem have made exactly this point: more desalination plants will be needed, especially if the island doesn't raise its current water prices (Earth.org). When price can't allocate the resource, something else has to. That something is usually politics.
When the Allocation Is Political: Taiwan in 2021
The clearest demonstration came during Taiwan's 2021 drought, the island's most severe in 56 years, which also disrupted hydropower and contributed to power outages (S&P Global Ratings). Chipmakers in Hsinchu Science Park were asked to cut their use of the municipal supply by 13 percent and began relying on water trucks, while a new pipeline began bringing more than 20,000 metric tons a day from wetter Taoyuan (Taiwan Business TOPICS). TSMC's spending on water trucks alone was expected to exceed its original budget (Forbes).
On the other side of the ledger, irrigation rules that favored chipmakers over rice farmers produced real tension (Tom's Hardware). Water rationing left more than 74,000 hectares of farmland fallow, with farmers compensated for lost harvests (Earth.org). By 2023, rice farmers in southern Taiwan had been barred from planting for a third consecutive year, and one farmer in a county where TSMC was building a new factory complained that the government was diverting water for others while his region barely had enough (NPR). NPR reported that the priority given to semiconductor development in Taiwan is not purely economic (NPR). When a resource is cheap, scarce and essential, the decision about who gets it becomes a public argument rather than a transaction.
What Residents Are Asking in Arizona
Arizona hasn't had Taiwan's drought-driven crisis, but the same conversation is starting. Residents are asking how much of the desert's water the TSMC campus will consume and whether it will affect their taps (AZFamily). A water policy expert at Arizona State University has said TSMC's expansion could make it Phoenix's largest tap-water customer (AZFamily).
The city's answer reveals the logic in play. Phoenix, according to the same coverage, made a deliberate calculation: it is a big employer, and the city looked at the water use and decided it had enough to serve it (AZFamily). The expert quoted said the city had effectively answered the question already by agreeing to serve TSMC, citing the economic return (AZFamily). That is a rational public decision, and it is exactly the kind of decision a price signal would otherwise make. It's being made by a city council weighing jobs against long-term supply, with the water bill playing a supporting role.
TSMC also operates under a strict industrial pretreatment discharge permit from the city, which requires it to clean its wastewater on-site before anything enters the public sewer (AZFamily). The public interest is being managed through permits, agreements and planning rather than through the market.
The Public Side of the Ledger
Fabs don't carry the whole cost of the regional water system. Arizona has invested more than $1.2 billion in water-related initiatives, and, as of 2023, the City of Phoenix had begun designing a water recycling facility expected to produce nearly 70,000 acre-feet of water a year by 2030 (GPEC). Intel says it returned and restored more than 100 percent of its Arizona freshwater use in 2021, and its Chandler campus treats up to 9.1 million gallons a day in an on-site reclamation facility (GPEC). Those figures come from a regional economic-development group repeating company claims, so they're best read as the industry's own account. Newer projects are building the water question into the plan from the start: ENR reports that Terafab's Grimes County development includes on-site power generation and water-treatment facilities (ENR).
For places where reuse and imports aren't enough, desalination is the backstop, and it also sets a ceiling on what water certainty costs. One Harvard-published estimate put a modest four-million-gallon-per-day desalination plant at under $100 million (Harvard Digital Innovation). That is affordable next to the cost of a fab, but it is a permanent new line item for every facility built where water can't otherwise be guaranteed.
The Challenge: Water May Be the Most Manageable Bottleneck of the Four
The honest counter-argument is strong. Unlike lithography machines, which one company makes, or photoresist, which only a few firms can supply, water has engineering answers that already work. Process-water recycling rates above 90 percent have been reported in Taiwan (UltraFacility), TSMC's Arizona plant is designed for near-zero liquid discharge, and Intel reports net-positive water use in Arizona. The cost is small relative to the factory. A resource that is cheap, recyclable and replaceable with desalination at tolerable cost doesn't look like a hard ceiling.
Three things keep it from being a solved problem. The first is timing: the Arizona reclamation plant doesn't open until 2028, and the fabs are operating on partial recycling in the meantime. The second is extremes. Taiwan's chipmakers had invested heavily in water reuse by 2021 and still needed trucked water during the drought, and a projection cited by The Diplomat has Taiwan's overall water consumption rising 7.3 percent above 2021 levels by 2036, leaving a daily supply deficit of 680,000 cubic meters (The Diplomat). The third is legitimacy. Even where the engineering works, the right to use the water has to be granted, and public patience for that grant isn't unlimited. Water is manageable. It is not manageable by technology alone.
Conclusion
Water is the bottleneck where the usual logic inverts. It is cheap, so price can't allocate it. It is recyclable, so the net demand is much smaller than the headline. And it is essential, so when supply tightens, the decision lands in politics. The fabs that will thrive are the ones that treat water not as a utility bill but as a permit, a relationship and an engineering system to be built before it is needed. The headline number, 17 million gallons a day, tells you how thirsty silicon is. The net number, 4 million, tells you what the city actually has to supply. The gap between them is the work.
This piece expands the fourth bottleneck in The Four Things Standing Between Us and Infinite AI Compute. Its companions cover the materials chokepoint and the talent gap.
Sources
source
covers
Phoenix fab water use; reclamation plant timing; development agreement
Gross versus net draw; recycling rates; Phoenix water chemistry
Net draw reduction for the first fab; agreement expansion to 17.3M gallons
65% startup recycling; 90% goal; near-zero discharge design
Resident concerns; city's reasoning; discharge permit
ASU expert on TSMC as largest tap-water customer; economic-return logic
TSMC's estimated water cost; 2021 drought severity
13% municipal-supply cut; water trucks; Taoyuan pipeline
TSMC water-truck spending in 2021
Irrigation policy favoring chipmakers
Fallow farmland; desalination and water-pricing argument
Farmers barred from planting; priority "not purely economic"
Projected 2036 Taiwan supply deficit
Arizona water investment; Intel net-positive claim; Phoenix recycling facility
Desalination plant cost estimate
Taiwan process-water recycling rates
TSMC's $265B Arizona expansion
Terafab's planned on-site water treatment and power
