ASML and TSMC have both signalled further chip capacity expansion, even though the returns on AI investment across the industry remain, on present evidence, unproven. That pairing is the whole story. The two companies closest to the physical bottom of the semiconductor stack are betting on volume; the companies one layer up are still waiting for the sums to work.

The signal matters because of who is sending it. ASML holds a monopoly on the extreme-ultraviolet lithography machines required for leading-edge logic. TSMC manufactures the overwhelming majority of the world's most advanced processors. When either of them moves, the rest of the supply chain reads it as weather forecasting. When both move in the same direction at once, the industry treats it as something close to a fact.

What are ASML and TSMC actually signalling?

The two companies have indicated that they expect chip production to keep growing, which implies continued demand for advanced manufacturing equipment and leading-edge wafer capacity. ASML's machines sit at the front of every advanced fab build; an expansion signal from the toolmaker is effectively a statement that its customers intend to keep ordering and keep fabricating.

TSMC's parallel signal points the same way. The foundry's capacity plans anchor the expectations of chip designers, substrate suppliers and packaging houses downstream. Expansion talk from the foundry ordinarily precedes capital commitments from everyone who depends on its wafers.

The precise scale of the new commitments is not detailed in the reporting available. What is confirmed is the direction: both companies are pointing toward more capacity, not less.

Why does the AI question stay open?

The uncomfortable part is the second half of the headline. Capital expenditure on AI infrastructure has run far ahead of demonstrated revenue from AI products and services. Chipmakers, cloud providers and hyperscalers have spent as though the demand curve were certain. The returns, so far, are not.

This creates a familiar asymmetry. The equipment maker and the foundry book their revenue early in the cycle, when the capital is spent. The AI developers who must justify all of it book theirs late, if the applications sell at all. ASML and TSMC can be right about near-term volumes and still leave their customers exposed to a demand shortfall two or three years out.

History offers a cautionary template. The telecom build-out of the early 2000s financed enormous equipment orders on the strength of projected internet traffic. The traffic eventually arrived. Several of the companies that paid for the build-out did not survive to see it.

Who carries the risk if AI revenue stalls?

The risk does not sit evenly. ASML sells machines against multi-year order books; its exposure is a cancellation risk, which historically has been modest. TSMC carries more capital weight in fabs and capacity, but serves a customer base broad enough to absorb single-project failures.

The greater exposure sits with the buyers: the firms converting balance sheets into GPUs, data centres and power infrastructure on the promise of future AI income. If that income stays thin, the expansion now being signalled will read, in retrospect, like the top of a cycle rather than the floor of a new one.

There is also a geopolitical edge to the expansion. Leading-edge capacity decisions by TSMC and EUV tooling decisions by ASML shape where advanced chips can be made at all, and both firms operate under export controls and government incentive regimes that did not exist a decade ago. Expansion signals from these two companies are read as closely in Washington, Brussels and Beijing as in the markets.

What happens next?

The test will be arithmetic, not sentiment. If AI product revenue rises to meet the capital already committed, ASML and TSMC's expansion will look prescient and under-sized. If it does not, the industry will have built capacity for a demand curve that exists mostly in investor presentations. The next two quarters of AI revenue disclosure, set against the equipment order books, will tell us which story we are in.