ASML and Taiwan Semiconductor Manufacturing Co. delivered quarterly results that showed the artificial intelligence chip expansion is still running at full speed, even as markets become more demanding about margins, cash flow and the return on heavy capital spending.
The two companies sit at critical points in the global semiconductor supply chain. ASML supplies the lithography machines needed to produce the most advanced chips, while TSMC manufactures processors for many of the world’s largest technology companies. Their latest updates suggest that demand for AI-related semiconductors has not weakened, but the market reaction also showed that strong growth alone is no longer enough to lift share prices sharply.
ASML raised its full-year 2026 revenue forecast to between 43 billion euros and 45 billion euros, up from a previous range of 36 billion euros to 40 billion euros. The Dutch equipment maker also reported second-quarter net sales of 9.33 billion euros and a gross margin of 54 percent.
TSMC, the world’s largest contract chipmaker, reported second-quarter revenue of $40.2 billion, up 12 percent from a year earlier and at the top end of its guidance. Its gross margin reached 67.7 percent, while operating margin came in at 60.3 percent. Net profit rose 77.4 percent year on year to 706.6 billion New Taiwan dollars, equal to 27.25 New Taiwan dollars per share.
Despite those strong numbers, the response from traders was restrained. Expectations had already risen sharply because of the AI hardware boom, leaving little room for disappointment. In TSMC’s case, gross margin fell short of the roughly 69 percent level many market participants had expected, turning an otherwise powerful earnings report into confirmation of the existing growth story rather than a fresh reason to reprice the sector higher.
Margins move into focus
The key change in market behavior is not a loss of confidence in AI chip demand. It is a shift in focus. Traders are now asking whether the huge spending required to build the next generation of semiconductor capacity can translate into durable profits.
TSMC raised its 2026 capital expenditure plan to between $60 billion and $64 billion, compared with an earlier range of $52 billion to $56 billion. The company said 70 percent to 80 percent of that spending will go toward advanced process nodes, while up to 20 percent will be directed to advanced packaging, testing and mask production.
That spending reflects the scale of demand from customers developing AI accelerators, central processing units, networking chips and other high-performance computing products. But it also brings higher costs. TSMC guided for third-quarter revenue of $44.6 billion to $45.8 billion, while forecasting a lower gross margin range of 65 percent to 67 percent.
Management said the margin pressure would come partly from the ramp-up of 2-nanometer technology and from the cost of overseas fabrication plants. Those factors are expected to reduce gross margin by about 3 to 4 percentage points. The company said the capacity expansion is necessary to meet customer demand, even if it weighs on near-term profitability.
That message is important for the broader technology market. AI chip demand remains strong, but the cost of keeping up with that demand is rising. New fabs, advanced packaging plants, power infrastructure and equipment purchases all require major upfront spending before they generate full returns.
ASML expands the equipment ceiling
ASML’s figures reinforced the same message from the equipment side of the supply chain. The company said it plans to expand capacity for extreme ultraviolet, or EUV, and deep ultraviolet, or DUV, lithography systems by about 30 percent for 2027 and 2028.
These machines are essential for manufacturing advanced semiconductors. EUV systems are used for the most advanced process nodes, while DUV tools remain important across a wide range of chipmaking steps and mature technologies. Because ASML has a dominant position in high-end lithography, its production capacity effectively helps determine how quickly the global chip industry can expand advanced-node output.
The company’s stronger revenue forecast for 2026 suggests that chipmakers are still reserving equipment years ahead of delivery. That is a clear sign that the industry is not preparing for a near-term downturn in AI semiconductor spending.
At the same time, ASML’s position gives it pricing power. Chief Executive Christophe Fouquet has indicated that demand for the company’s most advanced machines remains extremely tight, with key orders stretching well into future production periods. When supply is constrained, equipment pricing becomes a sensitive issue for customers already facing high capital costs.
TSMC chief executive C.C. Wei has made clear that preserving margins is a priority as the company absorbs higher costs across equipment, energy, labor and overseas expansion. Negotiations between the largest chip manufacturers and their suppliers are likely to become more important as the industry moves deeper into the next phase of AI infrastructure construction.
The second wave of AI chips
The results from ASML and TSMC also point to a broadening of the AI semiconductor cycle.
The first stage of the AI boom was heavily centered on graphics processing units, or GPUs, used for training and running large AI models. The next stage is spreading across a wider range of hardware, including CPUs, custom accelerators, networking chips, memory products and advanced packaging.
TSMC said revenue from high-performance computing products rose 20 percent from the previous quarter and accounted for 66 percent of total sales. The company also said chips made on 7-nanometer technology and below contributed 77 percent of wafer revenue.
Those numbers show how deeply advanced-node manufacturing is now linked to AI infrastructure. The most powerful AI processors require leading-edge manufacturing, and their performance also depends on the ability to connect multiple chips through advanced packaging technologies.
One of the most important bottlenecks is chip packaging. TSMC’s advanced packaging technology, including CoWoS, is used to combine processors and high-bandwidth memory in ways that improve speed and efficiency. The company said current advanced packaging capacity remains tight enough to limit some customer growth, and it is adding lines to ease that constraint.
This matters because AI systems are not built from processors alone. A full AI server requires accelerators, memory, power components, networking equipment, packaging and cooling systems. A shortage in any part of that chain can delay the deployment of complete systems, even when demand from cloud customers is strong.
Supply chain benefits spread wider
The expansion plans at ASML and TSMC are likely to support demand for other semiconductor equipment and materials suppliers. Companies involved in deposition, etching, inspection and process control are positioned to benefit from the construction and upgrading of fabs.
Applied Materials, Lam Research and KLA are among the companies tied to those stages of the manufacturing process. Their tools are used throughout advanced chip production, from layering materials on wafers to removing unwanted material and inspecting defects.
Memory suppliers are also central to the AI buildout. SK Hynix, Micron and Samsung Electronics supply high-bandwidth memory, or HBM, which is used alongside AI accelerators. HBM has become one of the most important components in AI servers because model training and inference require rapid movement of large amounts of data.
The relationship between logic chips, memory and packaging is becoming more complex. Faster processors require faster memory. Faster memory requires advanced packaging. Advanced packaging requires more capacity and more specialized equipment. That interdependence helps explain why capital spending is rising across the semiconductor supply chain rather than in one narrow segment.
Cloud spending remains the next test
The next major test for the AI chip cycle will come from the world’s largest cloud service providers. Microsoft, Amazon, Google and Meta have been major buyers of AI infrastructure, and their capital expenditure plans will help determine whether the current semiconductor buildout remains on track.
For chip suppliers, the key question is no longer only whether cloud companies want more hardware. It is whether that hardware can produce measurable business results through higher usage, stronger revenue and better cash generation.
AI model training remains expensive, and AI inference at scale can also carry large operating costs. Cloud companies must balance the need to secure scarce computing resources with the pressure to show that spending is producing returns.
If cloud demand continues to rise, ASML’s equipment expansion and TSMC’s fab buildout may prove well timed. If customer spending slows or shifts toward more efficient use of existing infrastructure, margins across the supply chain could face renewed pressure.
For now, TSMC’s longer-term view remains positive. The company expects AI-related demand to continue through 2029 and 2030, with a compound annual growth rate above 50 percent. Management cautioned, however, that growth may not move in a straight line and that periodic fluctuations are still possible.
Power demand becomes a larger constraint
The AI hardware boom is also increasing pressure on electricity grids. As more advanced chips are deployed in data centers, power availability is becoming a strategic issue for technology companies, utilities and governments.
Industry estimates show that data centers consumed roughly 415 terawatt-hours of electricity in 2024, equal to about 1.5 percent of global power generation. That share is expected to rise as AI workloads expand and as more servers are installed around the world.
The stress is already visible in some regions. In Ireland, large data centers have been reported to consume more than 21 percent of the country’s electricity supply. That level of demand has raised concerns about grid capacity, energy prices and the pace at which new data facilities can be approved.
Power constraints could influence where new AI infrastructure is built. Data center operators may increasingly favor locations with reliable electricity, access to renewable generation, favorable regulation and enough transmission capacity. Semiconductor demand may remain strong, but the ability to deploy chips at scale will also depend on energy infrastructure.
This adds another layer to the return question now facing the sector. Buying chips is only one part of the cost. Companies also need buildings, cooling systems, networking equipment and long-term electricity supply. If power becomes scarce or more expensive, the economics of AI computing could change.
Shared compute draws attention but carries risk
The high cost of AI computing has also increased interest in shared hardware networks and alternative cloud platforms. Some industry participants estimate that the shared computing market has recently reached about $12.2 billion in value, as users look for cheaper ways to access graphics processors and other high-performance chips.
Standard cloud providers can charge up to $8 an hour for access to a single high-end graphics chip, depending on availability, location and contract terms. Shared hardware networks claim they can reduce some of those costs significantly by connecting users with idle or underused computing resources.
Supporters of these systems argue that lower prices could attract developers, smaller companies and research groups that cannot afford large long-term cloud contracts. Some also link these platforms to digital utility tokens, although that part of the market remains volatile and carries regulatory, operational and liquidity risks.
The rise of shared compute does not change the core semiconductor story. Demand for advanced chips remains strong. But it does show that customers are searching for more efficient ways to use expensive hardware, particularly as AI workloads become more power-hungry and cost-sensitive.
For traders, the important distinction is between real demand for computing power and speculative claims built around that demand. Shared infrastructure may become part of the broader AI ecosystem, but its long-term role will depend on reliability, pricing, regulation and actual usage.
Capacity is still rising, but the bar is higher
ASML’s stronger order outlook and TSMC’s higher capital spending plans confirm that the upstream and midstream parts of the AI chip supply chain are still expanding. The industry is building more equipment capacity, more advanced wafer capacity and more packaging capacity.
That does not mean share prices will automatically move higher. Markets have already priced in a large amount of AI growth. The latest results make clear that traders are now looking more closely at execution, margin protection and capital discipline.
The AI semiconductor cycle has not entered a downturn. The evidence from ASML and TSMC points to continued expansion. But the story is maturing. The easy question was whether demand existed. The harder question is how efficiently the industry can convert that demand into profits, cash flow and durable returns over the next several years.
To act on this AI chip outlook, explore institutional-grade crypto trading tools with Toobit TradFi today.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
