Behind the Gigafab Deal: The Tropical Grid and Ultra-Pure Water Ledger Nobody Puts in the Deck

1. 581.8 GWh: one plant, 0.98% of national electricity

At 40,000 wafer starts a month the plant runs 480,000 wafers a year. The process line draws about 1,200 kWh per wafer, which is 576.0 GWh; ultra-pure water adds roughly 12 kWh a wafer, or 5.8 GWh. The two together are 581.8 GWh.

That number looks unremarkable at first: 0.98% of national electricity. The trouble is that it barely compresses. EUV tools do not stop, cleanrooms cannot be switched off, and the UPW loop runs around the clock, so the load curve is flat and demand response has nothing to work with.

2. Three cubic metres of UPW per wafer: 0.206% of the water supply and a hidden second power bill

Ultra-pure water is the most overlooked utility in a fab. Each 300mm wafer is rinsed dozens of times between clean, etch and CMP, adding up to 3 cubic metres. Across a year that is 1,440,000 cubic metres, or 0.206% of national water sales.

The real bill is not the water tariff but the power used to make it. Turning tap water into 18.2 megohm water takes reverse osmosis, EDI and polishing beds, about 4 kWh per cubic metre. So 5.8 GWh of electricity hides inside the water line: booked as power, driven by volume.

3. S$320.25 per wafer: inside the envelope that number runs 148 to 620

Plug in power at S$0.26/kWh and industrial water at S$1.71 per cubic metre: the year costs S$153,720,000, or S$320.25 per wafer across 480,000 units.

Take every parameter to its lower bound and the number is S$147.96 a wafer. Take every one to its upper bound and it is S$620.00. Plant power spans 145.0 to 1,461.6 GWh. This is not model imprecision, it is the defining feature of the asset: the interval matters more than the point.

So this piece will not hand you a single number. It publishes the range and the red lines, and any model that quotes one figure should be asked which point of the envelope it came from.

4. Power price is the only lever that moves on its own

Power is exogenous. Yield cannot hedge a tariff. Multiply the base tariff by 1.3 and utilities climb from 5.34% to 6.91% of revenue; by 1.507 and the line breaks 8%; by 1.888 and it breaks 10%.

Run it the other way: 6% needs only a 12.6% increase, S$0.2928/kWh. That is not a stress-test extreme, it is what the next contract cycle can deliver.

5. 1,057.7 GWh(th) of gas and 290,124 tonnes of CO2, all of it sitting in the permitting file

581.8 GWh does not appear from nowhere. Served by combined-cycle gas at 55% efficiency it needs 1,057.7 GWh of thermal input, which is 290,124 tonnes of CO2, or 604.4 kilograms per wafer.

In a tropical city those three numbers decide three things: where the grid connection sits in the approval queue, how long the environmental disclosure runs, and who carries the future carbon cost. The ribbon-cutting will not mention them. The ESG covenant in the loan agreement will.

6. Reverse test: cut the power price 20% and the ledger becomes 4.3%

A model that only moves one way is not worth trusting. Cut the tariff by 20% and utilities fall to about S$123.5M, or 4.29% of revenue. The elasticity is roughly 1:0.2, so every 10% move in tariff shifts the share by about 0.53 percentage points.

That elasticity tells you what to watch. Not the water tariff, not headcount, but the power curve. It decides whether this plant carries a 5% or a 10% cost structure over a ten-year cycle.