Resource profile
What a Leading-Edge Fab Consumes
A fab turns power, ultrapure water, gases and chemicals into patterned wafers at rates matched only by large petrochemical complexes and aluminum smelters.
The resources interact: a site with power but not water cannot host a leading-edge fab, a site with both but thin gas supply faces ramp delays, and a fab short of heat rejection capacity runs into limits on tool density and operating hours.
- Power
- Water and heat
- Gases and chemicals
- On-site utilities
| Resource | Leading-edge fab | Mature-node fab | Notes | Siting constraint |
|---|---|---|---|---|
| Electrical power | 200-600 MW peak; 1-3 TWh a year | 40-150 MW; 0.2-0.8 TWh | About 1 MW per EUV scanner; HVAC 30-50% of load | Hard: dedicated transmission; 3-5 year interconnection |
| Ultrapure water (UPW) | 10-20 million gallons a day gross; 2-10 net | 2-8 gross; 1-4 net | 18.2 MOhm-cm resistivity; cannot be bought or stored | Hard: needs a reliable year-round source |
| Bulk gases (N2, Ar, O2, H2, He) | N2 1-3 million SCFD; Ar 200,000-500,000 SCFD | N2 200,000-800,000 SCFD | N2 and Ar from an on-site air separation unit; helium concentrated in Qatar, the US and Russia | Moderate: 10-20 year supply contract with Linde, Air Liquide or Air Products |
| Specialty gases | NF3 hundreds to thousands of metric tons a year | Lower; tracks CVD tool count | NF3 and SF6 drive Scope 1 emissions; SiH4 pyrophoric; AsH3 and PH3 toxic | Supply chain: few producers; no stockpiling |
| Process chemicals | Sulfuric acid hundreds of thousands of liters a year; CMP slurry millions | Lower; mature supply chain | Resist qualification 12-18 months; JSR, Fujifilm and TOK in Japan | Procurement: qualification lock-in |
| Heat rejection | 300-800 MW thermal; 1-5 million gallons a day evaporated | 60-300 MW thermal | About the cooling load of a mid-sized power plant | Climate: hot sites need more energy and water |
| Compressed air | 100,000-500,000 SCFD | Lower | Generated on-site | Not a siting constraint |
Site comparison
Each US Fab Site Is Bound by a Different Resource
Power is a transmission problem, solvable anywhere given three to five years. Water is a watershed problem that infrastructure cannot fix if the basin cannot sustain the withdrawal. Gas supply is a logistics problem. Heat rejection is a climate problem: hot, dry sites trade cooling efficiency against water.
| Site | Power | Water | Heat rejection climate | Binding constraint |
|---|---|---|---|---|
| TSMC Fab 21, Phoenix, Arizona | Moderate: APS built new transmission | Critical: Colorado River basin; 100% recycle commitment | Challenging: extreme summer heat | Water |
| Intel Ohio One, New Albany, Ohio | Challenging: PJM queue; 400-800 MW needs new lines | Adequate: Ohio River and Columbus supply | Favorable: temperate | Power interconnection; first fab now 2030-2031 |
| Samsung, Taylor, Texas | Moderate: ERCOT islanded grid; Winter Storm Uri precedent | Moderate: growing county; Granger Lake; reclaimed water for cooling | Challenging: hot summers | ERCOT resilience; existing Austin S2 supply chain helps |
| Micron, Clay, New York | Strong: National Grid and NYPA hydropower | Strong: Oneida Lake and Oswego River | Favorable: free cooling most of the year | None binding; gas logistics need buildout |
| GlobalFoundries, Malta, New York | Adequate: National Grid; NYPA access | Strong: Hudson River watershed | Favorable | None binding; mature-node technology limits expansion |
Korea faces the same compound problem at larger scale: the Samsung and SK hynix clusters at Yongin need more than 14 GW by 2041, with three 1-GW LNG plants for the first phase, and water supply lags the fab schedule. Samsung Taylor details are on AustinIO.
Heat rejection
Every Watt In Leaves as Heat
A fab drawing 400 MW rejects about 350 to 380 MW of heat to the environment, and the cooling plant shapes the site as much as the substation does.
| Technology | Principle | Water use | Fab application |
|---|---|---|---|
| Evaporative cooling towers | Evaporates condenser water | 1-5 million gallons a day plus blowdown | Primary heat rejection at most leading-edge fabs; second-largest water use after UPW |
| Dry coolers | Fans across finned coils; no evaporation | Zero | Tool cooling circuits; cannot cool below ambient plus about 5 C; adiabatic pre-cooling cuts water 80-90% against full evaporative |
| Free cooling (economizer) | Bypasses chiller compressors when it is cold enough outside | Same as underlying system | Cuts chiller power 70-80% during free cooling hours |
| Waste heat recovery | Captures 80-160 C heat from compressors and abatement | Indirect reduction | Strongest case in cold climates; TSMC and Intel European sites studying community heat |
Free cooling hours a year at four fab locations. Temperate sites run chillers far less, an operating cost advantage desert and subtropical sites cannot match.
Node scaling
Resource Use per Wafer Rises With Every Node
Each new node adds process steps, gases, chemicals and power per tool. Throughput gains do not offset them, so resource intensity rises with both volume and node.
| Resource | Driver | Per wafer start | Per fab |
|---|---|---|---|
| Electrical power | EUV at about 1 MW a scanner; High-NA more; more plasma tools | Up; EUV was the largest step | About 100 MW in the 180nm era to 400-600 MW at N3-N5 |
| Ultrapure water | More cleaning, CMP and EUV develop and rinse steps | Up: 5nm wafers use about 2-3 times the UPW of 28nm | Up; recycling partly offsets |
| NF3 | More CVD and ALD steps; gate-all-around adds ALD cycles | Up; tracks CVD step count | Up; main driver of Scope 1 growth |
| Photoresist and slurry | More EUV coat-expose-develop cycles; more CMP steps | Mixed: thinner coats, more of them | Cost up; EUV resist 5-10 times ArF per liter |
| Heat rejection | Proportional to fab power | Up with power | Up; chiller capacity and tower water grow with power |
Clusters as a moat
Established Clusters Carry Decades of Infrastructure
Building and qualifying power, ultrapure water, gas and chemical supply at a greenfield site takes three to five years and billions of dollars beyond the fab and its tools, and the air separation plant, water plant and chemical distribution have little alternative use. Established clusters, Hsinchu Science Park in Taiwan, the Hwaseong-Pyeongtaek corridor in Korea, Kumamoto in Japan, and Shanghai, Wuxi, Wuhan and Hefei in China, already have competing gas suppliers, local chemical inventory and grids built for fab loads. Greenfield fabs in central Ohio, the Sonoran Desert or upstate New York build all of it in parallel with the fab, and it often sets the critical path.
Renewable targets
Renewable Supply Lags Fab Growth
TSMC used 28.8 TWh of electricity in 2025, about 20% of it renewable, against an RE100 target of 60% by 2030 and 100% by 2040; its overseas sites already run on renewable power. Renewable electricity cuts Scope 2 emissions only: the specialty gases above remain a Scope 1 problem, covered on Emissions and Abatement and Decarbonization.
Resource pages
Fab Resource Detail Pages
| Resource | Page | Coverage |
|---|---|---|
| Electrical power | Power for Fabs | Demand by tool, power quality, grid to tool, UPS, grid interconnection |
| Power resilience | Microgrids | Battery storage, islanding, blackstart, load shedding |
| Ultrapure water | Ultrapure Water | Purity spec, production train, consumption, water-stressed sites |
| Bulk and specialty gases | Gas Delivery | Gas hazards, delivery tiers, air separation, purification |
| Process chemicals | Chemical Delivery | Delivery tiers, photoresist dispense, CMP slurry, safety |
| Emissions | Emissions and Abatement | GWP table, abatement methods, regulation |
| Decarbonization | Decarbonization | Scope 1 and 2, RE100, site comparison |
| Vacuum | Vacuum Systems | Pressure regimes, dry pumps, abatement integration |
Record
Page Record
- Covers
- Fab resource consumption, US site comparison, heat rejection, node scaling, cluster infrastructure and renewable targets
- Sources
- TSMC sustainability report (2025 electricity and RE100 targets); Korea Herald (Yongin power plan, August 2026); utility and watershed data (APS, AEP Ohio, Oncor, National Grid, NYPA); company filings (TSMC, Intel, Samsung, Micron, GlobalFoundries); ASHRAE climate data for free cooling hours
- Last verified
- October 2026