The Invisible Ceiling:
People, Power & Water

The first map showed where the money landed — roughly $60.8 billion in chip FDI since 2020, nearly 80% of it in two countries. This is the part the money can't buy. Three constraints decide whether a site can actually run what it attracted, and none of them appears at a ribbon-cutting.

The headline arrives. The capacity has to catch up.

The first piece in this series mapped where the investment went: five very different national bets, with nearly 80% of the money concentrated in Singapore and Malaysia. The optimistic read is that the capital has chosen Southeast Asia. The honest read is that announcing an investment and running it are two different problems — and the second one is far harder to fix with a press release.

At every industry forum in Penang this year, the same worry surfaces underneath the optimism. Not "will the money come?" — it's coming. The question is whether these countries can absorb it. And the constraint isn't capital. It's three unglamorous things that no incentive package can conjure on schedule: skilled people, electricity, and water.

Ceiling one: people

Start with the one that can't be imported overnight. Malaysia — the most developed semiconductor base in the region outside Singapore — is estimated to need around 50,000 skilled engineers to meet current demand. Its universities produce roughly 5,000 engineering graduates a year. That is a tenfold gap, and it is not one any policy bridges quickly. The National Semiconductor Strategy targets training 60,000 highly skilled engineers by 2030; industry insiders, asked directly, are candid that the pipeline is nowhere near that pace today.

Worse, the bucket leaks. The Malaysia Semiconductor Industry Association has put the country's annual talent loss to brain drain at roughly 15% a year — engineers poached by Singapore next door and by multinationals competing for the same shallow pool. Industry leaders describe it plainly as a "war for talent," where firms bid up salaries and raid each other rather than grow the pipeline. This is not a Malaysia problem alone: SEMI projects Asia-Pacific will be short more than 200,000 engineers by 2030.

Vietnam sits one rung earlier on the same ladder. Its ambition is real — the government has set out a national semiconductor strategy and a dedicated workforce program — but when SEMI's leaders met Vietnam's prime minister, their advice included something telling: promote the use of English in professional and regulatory settings. When the guidance is still at the level of working language, the talent base is being built, not drawn down. That is not a criticism. It's a phase. But phases matter when you're deciding where to put a plant this year versus in 2030.

"A war for talent."

Ceiling two: water

Chipmaking runs on water most people never think about — ultrapure water, or UPW, used to wash wafers between nearly every process step. It is not "clean" water in any ordinary sense. It must reach a resistivity of about 18.2 megohm-centimetre, with organic carbon below one part per billion. And making it is wasteful: roughly 1,400 to 1,600 gallons of raw water yield 1,000 gallons of UPW — about a third is lost in the process.

The scale is the part that reorganises a map. A single large advanced fab can consume on the order of 38 million litres of UPW a day. UPW technology itself can be imported — the membranes and systems are available off the shelf from global vendors. What cannot be imported is a stable, clean source of raw water to feed it. That single fact sorts the region more sharply than any incentive table.

Read those three together and a pattern appears. Water-rich does not mean fab-ready: Vietnam has the volume but pays for clean-up; Penang has the ecosystem but a structural supply question; Batam has the location but a hard physical ceiling. The same word — "water" — hides three different investment realities.

Ceiling three: power

The third constraint is the one that has changed fastest, because the fabs are no longer the only heavy users on the grid. Of the 89 new wafer fabs expected to come online worldwide by 2029, Southeast Asia is projected to host just six. That scarcity is partly capital and partly something more physical: front-end fabs need vast, uninterrupted, stable power, and they now compete for it with the AI build-out happening in the same countries.

The numbers behind that competition are moving quickly. AI server racks are climbing from roughly 120–150 kilowatts toward 600–1,000 kilowatts each as they densify. Every data centre that lands near a fab corridor is bidding for the same megawatts, the same grid upgrades, the same priority connections. A government can offer a tax holiday in an afternoon. It cannot conjure firm, stable baseload power on the same timeline — and increasingly it must choose who gets it.

Why the ceiling sorts front-end from back-end

Put the three constraints together and they don't weigh on every kind of chip plant equally. Assembly, testing and packaging — the back end, where ASEAN already lives — are comparatively light on water and power and shallower on process-engineering depth. Wafer fabrication — the front end, the ambition everyone states — is the opposite: it is the part that drinks 38 million litres a day, demands the deepest engineering bench, and needs the most uninterrupted power.

That is the quiet logic underneath the headlines. When Malaysia, Vietnam and Indonesia concentrate on back-end and materials despite stated front-end goals, it is not a lack of nerve. It is the ceiling expressing itself. The constraints decide what's realistic in three years versus what stays a 2030-plus aspiration — and they decide it differently in each country.