ASML Raises Forecast on AI Chip Demand

ASML Holding posted Q1 2026 net sales of €8.8 billion on 15 April — €300 million above LSEG consensus — and immediately raised its full-year guidance, projecting 2026 revenue of €36 billion to €40 billion against the prior forecast of €34 billion to €39 billion. The revision, upward on both ends by €2 billion, arrived the same week that reports confirmed more than half of planned U.S. AI data centre construction projects have been delayed or cancelled due to power grid constraints and hardware supply shortages. The machine that makes the machines, it turns out, is running ahead of schedule.

ASML is the sole manufacturer of extreme ultraviolet (EUV) lithography systems — the machines without which advanced semiconductors below the 7-nanometre threshold cannot be produced. That monopoly positions ASML as a leading indicator for the entire AI hardware chain. When ASML's orders surge, foundries are committing capital to expand the physical capacity that turns chip designs into silicon. When they dry up, the AI model buildout eventually stalls for want of compute.

By the numbers, the quarter was strong across every metric. Q1 net income came in at €2.8 billion, beating the LSEG estimate of €2.5 billion, with a gross margin of 53.0 percent — the top of the guidance range. ASML plans to ship 60 low-NA EUV systems in 2026, a 25 percent increase over its projected 2025 deliveries, CEO Christoph Fouquet said on a 15 April earnings call. China's share of total sales fell to 19 percent in Q1 from 36 percent in Q4 2025, reflecting the latest round of U.S.-led export restrictions that took effect in early 2026. Fouquet said the company's overall order intake "continues to be very strong" despite the China compression, as customers in Taiwan, South Korea, and the United States accelerate capacity commitments.

“By the numbers, the quarter was strong across every metric.”

The guidance revision is unambiguously good news for the foundry ecosystem. Taiwan Semiconductor Manufacturing Company, Samsung, and Intel have all accelerated planned capacity announcements this quarter. TSMC's Arizona expansion — Fabs 21 and 22 — requires a steady supply of ASML's high-NA EUV systems beginning in 2027, and the ability to confirm a 25 percent shipment increase this year signals that the delivery bottlenecks that bedevilled customers in 2024 and 2025 are easing. For cloud providers — Amazon, Microsoft, and Google — who have collectively committed over $600 billion in AI infrastructure spending through 2027, confirmed EUV delivery schedules are the prerequisite for knowing when their custom silicon roadmaps become real production assets.

ASML's own stock slipped 2.1 percent on 15 April despite the earnings beat, reflecting investor anxiety about China rather than the underlying business strength. China represented 36 percent of ASML revenue as recently as Q4 2025; the drop to 19 percent in Q1 is not a market decision — it is a compliance decision. U.S. lawmakers introduced legislation in February 2026 that would bar ASML from selling any lithography equipment to Chinese chipmakers, including the less-advanced deep-ultraviolet machines not yet covered by existing controls. If that bill passes in its current form, ASML loses access to roughly $7 billion in annual Chinese revenue. The company does not comment on individual legislative proposals.

The memory chip shortage is a separate drag on the ecosystem. Prices for high-bandwidth memory — the type stacked directly on AI accelerators — have risen to record highs this year, partly because TSMC and Samsung face competing demands from logic and memory customers for the same EUV time slots. Some AI accelerator programmes have slipped their production ramp schedules by one to two quarters as a result, according to industry consultancy TechInsights.

The news ASML cannot control is the AI data centre power problem. A Lawrence Berkeley National Laboratory analysis published this week projects that AI data centre power demand in the United States will reach 12 percent of national grid capacity by 2028, up from 4 percent in 2024. More than half of planned data centre projects announced since 2024 have been delayed or cancelled — not for lack of capital but for lack of electricity. If the power constraint does not ease through new grid capacity or more efficient chip architectures, demand for the processors ASML's machines produce could plateau before the physical infrastructure catches up.

The next meaningful data point is ASML's Q2 2026 results, due in mid-July. If China restrictions tighten further before then, and if the legislative bill advances through committee, the gap between €36 billion and €40 billion suddenly narrows in ways the current consensus has not fully absorbed. For now, the machine that makes the machines says the AI cycle has runway left. The machines that run on electricity are telling a more complicated story.