ASML ramps up production, TSMC doubles down: the “second wave” of AI chips—why does the market still think it’s not good enough?

Over the past week, the strained AI hardware sector has been waiting for a shot strong enough to turn sentiment around.

What ASML and TSMC then delivered were basically strong fundamentals financial results that still failed to fully satisfy elevated expectations: the former significantly raised its full-year revenue and gross margin guidance and began increasing lithography tool capacity for 2027–2028; the latter kept revenue, gross margin, and operating profit margin at historical highs, while raising full-year capital expenditures in one jump to $60 billion to $64 billion.

In theory, this should be the ideal combination for AI semiconductors—an equipment company proving customers are still placing orders, and the leading wafer foundry demonstrating that orders are turning into revenue and that it’s willing to keep pouring huge sums into expansion.

But the market’s response did not match the strength of the results.

The reason isn’t that the fundamentals of the two companies deteriorated—it’s that expectations for the AI industry chain have already been pushed to an unusually high level. The market is no longer satisfied with “demand is still strong”; instead, it wants every earnings report to be revised upward again, every gross margin/profit margin item to break through the ceiling, and all massive capital expenditures to be immediately converted into higher profits.

This is also why ASML and TSMC’s earnings reports send two seemingly contradictory but actually consistent signals: the AI semiconductor expansion cycle is still continuing, and some key links are even accelerating; but the way the capital market is pricing this cycle has moved from validating demand to validating returns.

I. ASML and TSMC both ramp up: the expansion cycle is far from over

ASML was the first to unveil the answer for this earnings season.

The company’s net sales in the second quarter reached €9.326 billion, ahead of the prior guidance of €8.4 billion to €9.0 billion; gross margin was 54%, and net profit was €2.918 billion. It then raised its third-quarter sales guidance to €11.0 billion to €12.0 billion and lifted its full-year 2026 sales expectation from €36.0 billion to €40.0 billion to €43.0 billion to €45.0 billion.

More important than the single-quarter numbers, however, is that ASML has started adjusting equipment capacity plans for the next two years. The company plans to increase low numerical aperture EUV capacity by 30% in 2027, on top of about 65 tools in 2026, and also to raise DUV immersion equipment by 30% from about 130 tools. At the same time, ASML is studying the possibility of further expanding in 2028.

The lithography equipment supply chain is complex and deliveries take a long time. ASML will not rashly increase production capacity for two years later based solely on fluctuations in orders over one or two quarters. Such an expansion plan implies that wafer fab customers are locking in advanced process and high-end memory capacity for 2027–2028 in advance.

One day later, TSMC provided corresponding validation from the wafer manufacturing side.

In the second quarter, the company generated revenue of $40.2 billion, up 12% quarter-over-quarter, at the top end of the prior guidance range of $39.0 billion to $40.2 billion; gross margin was 67.7%, slightly above the top end of guidance, and operating margin reached 60.3% for the first time. Net profit was NT$706.56 billion, up 77.4% year over year, and earnings per share were NT$27.25.

Revenue mix is also continuing to tilt toward AI and advanced processes. In the second quarter, high-performance computing (HPC) revenue grew 20% quarter-over-quarter and accounted for 66% of company revenue. Advanced nodes of 7 nanometers and below accounted for 77% of wafer revenue; among them, 3nm and 5nm contributed 30% and 33%, respectively, and 2nm, which is in the ramp-to-volume phase, contributed 3% of wafer revenue for the first time.

More signal-rich still is capital expenditures. TSMC raised its 2026 capital expenditure plan from the prior $52 billion to $56 billion up to $60 billion to $64 billion. About 70% to 80% will be used for advanced processes, and about 10% to 20% for advanced packaging, testing, mask manufacturing, and other steps.

The company also increased its full-year US dollar revenue growth expectation, from “over 30%” to “slightly above 40%.” Management said AI-related demand remains extremely strong, and demand signals from cloud service providers and from downstream customers are still positive.

ASML is preparing to increase lithography equipment capacity, while TSMC expands wafer manufacturing and advanced packaging capabilities through higher capital expenditures.

So when the equipment leader and the world’s largest foundry both raise future investment, at least one thing can be confirmed: AI semiconductor capital expenditures have not entered a contraction cycle, and the supply chain is still preparing capacity for demand acceleration in the coming years.

II. With results this strong, why does the market still feel it’s not enough?

The issue is that the market is waiting for more than a “meeting targets” kind of earnings report.

Because TSMC releases revenue data every month, the $40.2 billion revenue in the second quarter has already been largely digested by the market. Therefore, before the earnings release, the real expectation gap lies in the gross margin, third-quarter guidance, and how far capital expenditures can be raised.

From this perspective, TSMC’s second-quarter gross margin of 67.7%—though above the company’s earlier guidance range of 65.5% to 67.5%—is roughly in line with mainstream expectations after the market’s upward revisions; it did not satisfy some investors’ more aggressive views approaching 69% or even higher.

For the third quarter, the company expects revenue of $44.6 billion to $45.8 billion, which implies about a 12% quarter-over-quarter increase at the midpoint; but gross margin guidance is lowered to 65% to 67%, with a midpoint of about 66%.

A decline in gross margin does not necessarily mean demand is weakening.

TSMC expects that the rapid ramp to volume for 2nm in the second half will dilute gross margin by about 3 to 4 percentage points. Continued overseas wafer fab expansions will also keep increasing depreciation and manufacturing costs. Strong advanced process demand, higher capacity utilization, and improvements in manufacturing efficiency can only partially offset these pressures.

In other words, TSMC is facing a typical high-demand expansion paradox: the stronger the demand, the more the company needs to buy equipment in advance, build wafer fabs, and introduce new processes; and the higher the capital expenditures, the earlier depreciation, overseas production costs, and new-node ramp pressures will show up in profit margins.

That’s also the most important thing to understand in this earnings report.

Judging from management’s comments on pricing strategy in the earnings briefing, TSMC is not aiming to push short-term gross margin to the limit when supply is tight. The company emphasizes that it is a long-term partner to its customers and will not squeeze customers by abruptly hiking prices; instead, it aims to maintain profitability at a level sufficient to support long-term expansion.

This means TSMC is more inclined to keep a balance between pricing power, customer relationships, and continued expansion rather than cashing in all of the scarcity premium at once. From an industry perspective, that is clearly a positive signal, but from a short-term trading perspective, it implies investors need to accept a reality: AI demand remains strong, but it does not necessarily mean that every additional dollar of revenue will immediately translate into higher gross margin.

As a result, the market’s lukewarm reaction to TSMC’s earnings report cannot be simply interpreted as AI demand peaking. A more accurate explanation is that in an environment where expectations are already extraordinarily elevated, strong performance has become a necessary condition for valuation, but no longer automatically creates a new upward catalyst.

After the earnings were released, strong performance did not immediately translate into sustained sector gains, which also reflects that investors are absorbing margin pressures and overly high expectations.

III. Only by pairing ASML and TSMC can you see the “second wave” of AI chips

If you look at ASML and TSMC’s earnings together, the outline of the so-called “second wave” of AI chips is clearer than before.

It isn’t a return to “total shortage of all chips,” nor is it a simple repeat of the two-year run centered on Nvidia GPUs. Instead, the supply bottlenecks are continuing to spread across the entire AI system.

ASML’s EUV and DUV tools determine how quickly advanced process capacity can be expanded; TSMC’s 3nm and 2nm determine how much wafer capacity can be obtained for GPUs, CPUs, and custom ASICs. HBM determines memory bandwidth, and advanced packaging such as CoWoS determines whether compute chips, memory, and high-speed interconnects can ultimately be assembled into data-center products that can be delivered.

If any one of these links fails to expand enough, it will slow down shipments of the entire AI system.

Even management at TSMC has stated clearly that advanced packaging capacity is currently so tight that it limits customer growth. It is working to close the gap between demand and capacity, while also welcoming alternative packaging solutions to give customers more options.

Meanwhile, AI demand is also spreading from a single accelerator to a wider range of chip types.

TSMC believes the development of Agentic AI is again increasing the importance of CPUs in data centers. Regardless of whether customers use x86, Arm, or RISC-V architectures, the advanced chips behind them are still mostly manufactured by TSMC. This means future AI capital expenditures will not only flow to GPUs, but will continue to drive demand for CPUs, network chips, memory, and advanced packaging.

Management’s outlook on long-term demand is also constructive. TSMC believes AI-related trends will remain strong through 2029–2030, during which period it does not rule out intermittent fluctuations, but the long-term direction has not changed. On the earlier judgment of compound growth of more than 50% for AI-related businesses, management did not provide new specific numbers; it only said that the demand trend is stronger than previously expected.

But that does not mean all semiconductor companies will benefit evenly.

• ASML (ASML.M) directly benefits from lithography equipment demand and advanced process expansion;
• Applied Materials (AMAT.M), Lam Research (LRCX.M), and KLA (KLAC.M) each benefit from equipment demand such as deposition, etching, and inspection, though the order fulfillment timing may differ;
• TSMC (TSM.M) controls advanced wafer manufacturing and packaging capacity, making it the key carrier for AI chip capacity expansion;
• SK hynix (SKHY.M), Micron (MU.M), and Samsung Electronics assume HBM and high-end memory supply;
• Cloud providers with in-house chip capabilities—Nvidia (NVDA.M), AMD (AMD.M), Broadcom (AVGO.M), as well as Amazon (AMZN.M), Alphabet (GOOGL.M), Microsoft (MSFT.M), and Meta (META.M)—collectively determine how fast end-demand can ultimately grow.

They are in the same capital expenditure cycle, but they have completely different technological barriers, capacity constraints, profit structures, and valuation levels. Therefore, the “second wave” of AI chips is more likely to be a structural market opportunity rather than the entire hardware supply chain synchronously rising again.

In the next stage, the market will focus more on which companies truly control scarce, hard-to-replicate capacity, which companies are just following customers to increase capital expenditures, and which companies can continue to improve free cash flow and capital returns after expanding capacity.

And after ASML and TSMC, the next key validation will once again fall on cloud service providers such as Microsoft, Amazon, Google, and Meta. After all, even if equipment companies expand, and wafer fabs invest, in the end cloud providers still need to keep raising capital expenditures—and prove that ever-growing AI infrastructure can generate real model calls, enterprise revenue, and cash flow returns.

Written at the end

Objectively speaking, ASML answered whether wafer fabs still want to buy equipment, and TSMC further proved that customer orders are enough to keep pushing the company to add both wafer and advanced packaging capacity.

From this perspective, the semiconductor industry cycle for AI has not topped out.

After all, capacity for equipment is expanding, advanced packaging remains tight, and TSMC even raised its full-year capital expenditures to the highest $64 billion—these are not signals a contracting industry would release.

But the reason the market still feels it’s not enough is that the problem in the next phase has changed. Previously, investors needed to confirm whether AI demand is truly real; now, demand is difficult to deny. What the market wants to know is how much capital needs to be put in to meet these demands, and how much profit margins and cash flow those investments can ultimately produce.

Therefore, the “second wave” of AI chips may already have begun, but it will not be a simple replay of the first round of the rally.

The real scarcity is no longer just companies that can provide more chips. It’s companies that both control critical capacity and, even after massive expansion, can continue to maintain pricing power, profit margins, and capital returns.