TSMC Made Kumamoto a Chip Hub. Water Is the Constraint.

TSMC opened its first Kumamoto fab on 24 February 2024, with mass production targeted by the end of that year. The plant belongs to Japan Advanced Semiconductor Manufacturing, or JASM, in which TSMC holds 86.5%, Sony Semiconductor Solutions 6.0%, DENSO 5.5% and Toyota 2.0% (TSMC). A second fab was announced in the same month. Together the two were to take the site above 100,000 twelve-inch wafers a month, on total investment above US$20 billion, with the second building starting construction by the end of 2024 and operating by the end of 2027 (TSMC).

Almost everything about the chips has since changed. The second fab was announced for 40, 22/28, 12/16 and 6/7 nanometre processes. On 5 February 2026, TSMC chairman C.C. Wei told Japan’s prime minister that it would instead produce 3-nanometre chips (KAB). The schedule has moved too. The Kumamoto Regional Economic Research Institute records the chairman acknowledging a delay in October 2025, against a prefectural position that groundbreaking within that year was still the plan, and puts the second fab’s investment at roughly ¥1.69 trillion, or about ¥3 trillion for the site, against government subsidies of up to ¥476 billion for the first fab and ¥732 billion for the second (熊本地方経済総合研究所).

The node, the money and the timing are still moving. The water under Kikuyo and the wires into Kyushu are not, and they set the outer limit on whatever gets built.

Kumamoto regulates groundwater in a way most places do not

Kumamoto City and the surrounding basin take essentially all of their tap water from groundwater. The prefecture has had a groundwater conservation ordinance since 1978, revised in March 2012. It works through pump size: in designated priority areas, a pump with an outlet cross-section above 19 square centimetres needs a permit, and elsewhere the threshold is 125 square centimetres. Permit holders are pushed toward rational use and toward recharge, under separate prefectural guidelines for each (熊本県).

That is the frame JASM entered. On 16 May 2023 it signed a five-party recharge agreement with Kikuyo Town, the Water-Circulating Farming Council, the Kumamoto Groundwater Foundation and the prefecture, committing to recharge more groundwater than the business takes (熊本県). TSMC’s own opening announcement put the same commitment in one line: the site will replenish more than 100% of the groundwater it uses, and will run on 100% renewable energy from the start of operation (TSMC).

The mechanism behind the recharge promise is agricultural. Paddy fields flooded outside the normal growing season let water percolate down into the aquifer. So a chip fab’s water balance in Kumamoto is settled by how many hectares of rice paddy stay under water, and by who pays the farmers to keep them that way.

The second fab takes the draw up two and a half times

The numbers are large but not mysterious. The research institute puts the first fab at 3.1 million tonnes of groundwater a year, an average of 8,500 tonnes a day, with a 75% recycling rate, rising to about 8 million tonnes a year once the second fab operates. In Kikuyo Town, industrial water use was 4.44 million tonnes in 2023 and is projected at 12.44 million tonnes in 2027 with the second fab included, which would be roughly 67% of everything the town extracts (熊本地方経済総合研究所).

Extraction across the Kumamoto area fell by about 12 million cubic metres between 2009 and 2023, so the basin has been drawing down its total even as the fab arrived. The revised ordinance sets a recharge target of, in principle, ten-tenths of what is taken, and the 2024 planning figures were 38 million cubic metres of recharge against 165.5 million cubic metres of extraction (same source).

A ten-tenths recharge target alongside a basin-wide ratio near a quarter is not a contradiction. The target binds permit holders, and the extraction total counts every well in the basin, including the farms and the households. What it does mean is that JASM keeping its promise settles JASM’s balance and not the aquifer’s. The harder question is whether a stack of company-level promises adds up to a basin that balances.

Kyushu’s grid is the opposite problem

Kumamoto regulates its groundwater tightly. Kyushu’s electricity runs into a different problem: on certain days, at certain hours, power is thrown away.

Kyushu Electric operates four reactors, Sendai 1 and 2 and Genkai 3 and 4 (九州電力), alongside the densest solar fleet in the country. The result is curtailment: in the grid working group’s FY2024 outlook, Kyushu had the highest renewable curtailment rate of any area in Japan at 6.1%, ahead of Chugoku at 5.8% and Shikoku at 4.5% (RTS Corporation).

A fab is a flat load. It runs at nearly constant power, day and night, all year. For a grid that throws away spring sunshine because demand is too low at midday, a very large constant industrial load in the same region is a genuine asset, not only a burden. This is the same structural gap we described in Japan’s battery storage interconnection queue: the country is short of things that can absorb power where and when it is produced, and long on applications to connect things that produce more.

The constraint is the local network between a substation and a plant drawing hundreds of megawatts. What decides the project is whether that reinforcement lands before the building does.

Two ledgers, and they disagree on purpose

Kumamoto Prefecture is working toward net zero greenhouse gas emissions by 2050, and reports FY2023 emissions 28.1% below its FY2013 baseline (熊本県).

Now add a cluster. JASM says it runs on 100% renewable energy. That claim is a market-based Scope 2 claim: contracts and certificates matched to consumption. It is real, it is auditable, and it is how corporate reporting works.

A prefecture’s inventory is territorial. It counts electricity consumed inside the boundary at the grid’s average emission factor. Certificates bought by one company do not change the factor applied to everyone else. So a prefecture can hold its 2013-baseline reduction, host a plant that reports zero, and still see its territorial industrial electricity line rise.

Each number is computed correctly, and they answer different questions. A company reports what it bought under contract, while a prefecture counts what was burned inside its borders. Most regional decarbonisation arguments in Japan will now be fought in that gap, because the GX-ETS entered its mandatory phase on 1 April 2026 and put verified corporate numbers into circulation without changing how a prefecture counts.

What this means if you build or fund here

Semiconductor and data-centre developers in Japan have spent a decade optimising for land, subsidy and labour. Kumamoto shows the binding constraints are a permit under a 1978 ordinance and a substation upgrade. Both have lead times measured in years, and neither is fixed by capital.

Recharge behaves like a supply chain. Flooding paddies is cheap and it can be verified, and it depends on farmers who are ageing out. A recharge obligation that outlives the farms delivering it carries a counterparty risk nobody has priced. Measurement, aggregation and long-term contracting for aquifer recharge is a buildable business.

Regional governments need territorial numbers that reconcile against corporate claims, and nobody is selling that reconciliation well. It is dull work, and it decides whether a prefecture’s 2050 target survives contact with an industrial cluster.

Japan’s semiconductor policy has been framed as economic security, most explicitly in the Rapidus 2nm programme in Hokkaido. Kumamoto is that policy landing on a watershed with farmers in it. The permit and the substation upgrade are decided in Kikuyo.

Tech for Impact Summit 2027 is in Tokyo on 18–19 May 2027, co-hosted with the United Nations University. The people who decide whether a fab gets its water and its power are rarely in the same room as the people financing the chips, and that room is what we build. It is invitation-only, and if this is your work we would like to invite you. Join the waitlist at tech4impactsummit.com/apply.