Taiwan Semiconductor

Taiwan Semiconductor Manufacturing Company raised its 2026 capex budget by USD 8 billion to USD 62 billion at the midpoint and full-year revenue growth target to over 40% in US dollars. It announced plans to invest an additional USD 100 billion in Arizona, cumulatively totaling USD 265 billion.

Why it matters: We believe higher full-year revenue and capital expenditure guidance ease concerns about the durability of artificial intelligence growth and supply chain bottlenecks. The capex hike is a response to hyperscalers raising their 2027 budgets. We raise our 2026 to 2030 capex forecasts by 22% on average. TSMC's expanded investment plans in Arizona show durable demand growth in US AI data center customers, and it doubles as a hedge against potential tariffs and geopolitical risks. We suspect TSMC is vague on timing to reserve flexibility in case of a temporary slowdown in AI buildout. We expect price hikes in 2027 due to higher raw material costs and tight supply. Meanwhile, the higher 2026 revenue guidance reflects AI demand benefiting both leading-edge and mature processes, plus production of more expensive 2-nanometer chips ramping up faster than expected.

The bottom line: We raise our fair value estimate for wide-moat TSMC to TWD 3,440 (USD 534 per ADR) from TWD 2,700 (USD 428 per ADR) as we believe hyperscalers' elevated spending in the coming years justifies TSMC's bigger capex budget. We believe the market is overly cautious about recent financing activities to build more data centers. Nvidia's revenue-sharing scheme may sound alarming, but the scale of neoclouds is dwarfed by stratospheric investments by hyperscalers like Microsoft and Alphabet.

Big picture: TSMC sees no material bottlenecks to expansion. We attribute this confidence to its supplier ASML's plans to increase capacity by 30% in 2027 and another 30% in 2028. Our higher TSMC capex estimates also hinge partly on ASML's ability to execute its expansion.

For third-quarter 2026 guidance, management projects revenue to grow 13.9% from the previous quarter to TWD 1.45 trillion (USD 45.2 billion) at the midpoint. The midpoints of gross and operating margin guidance are higher sequentially, at 66% and 57%, respectively. We expect results to likely come in at the higher end of guidance, with management being conservative on margin-dilutive effects from the 2nm ramp-up and non-AI demand.

Our base-case fair value estimate is USD 534 per share, at which TSMC would trade at a forward price/earnings ratio of 26 times per 2027 estimates. We use a weighted average cost of capital of 8.2% to discount our forecast cash flow for TSMC.

We project the company’s top-line compound annual growth rate at 25.1% over the next five years. Even with its dominant market share, we believe TSMC can deliver above-industry growth through a higher proportion of more valuable 10 nm to 1.4 nm logic and 28 nm to 7 nm specialty products, which are currently only produced by it and Samsung at scale. We expect Internet of Things and automotive applications to be sources of incremental demand in newer specialty products. In terms of node advancement, mass production of A14 will occur in 2028 and its derivatives in 2029.

We project 2026 gross and operating margins to be higher year on year at 66% and 54%, respectively, as the profitability of 3 nm production improves and the revenue of high-gross-margin AI products grows faster than expected. While quarterly margins may fluctuate while the company ramps up production of a new node, long-run margins should be stable, as we expect TSMC’s moat to support its pricing power for years to come. Management also aligns its interests to the 56% gross margin target by listing it as one of the criteria for performance-linked bonuses.

High-performance computing is TSMC's largest growth driver. We believe increasing in-house design of cloud computing and AI chips by US and Chinese internet giants will benefit TSMC for the next few years.

One recurring threat to TSMC's long-term growth is the loss of key personnel to competitors. In the past, TSMC has managed the issue by offering above-average salaries to employees. But to counter Chinese semiconductor companies that are poaching talent, TSMC rolled out a performance share scheme in 2021, which links financial performance and environmental, social, and governance initiatives to staff remuneration. We think this better aligns employees' interests with more stakeholders. TSMC's scheme is one of Taiwan's first to include ESG goals.

We believe TSMC’s wide moat stems from intangible assets that are realized from its leading position in advanced process technologies, or nodes, in semiconductor manufacturing. It enjoys a cost advantage stemming from its ability to spread unit research and development costs, given its dominating scale. TSMC’s long-standing leadership in node advancement comes from its ability to work with customers years in advance to identify the optimal improvements in power, performance, area, cost, and time to market, or PPACT, for each chip while maintaining fiscal discipline. This has justified higher prices than peers. TSMC’s leadership creates a virtuous cycle of premium products, premium pricing, and profits for next-generation research and development. Through seizing opportunities in smartphones in the 2010s and data centers (including AI) in the 2020s, TSMC has increased its market share to over 70% of the foundry (contract manufacturing) market and simultaneously expanded operating margins to above 40%. We project TSMC to achieve mid-20s returns on invested capital in the next five years, above its weighted average cost of capital of 8.2%.

We believe the gap between TSMC and its smaller peers remains wide. TSMC and Samsung Electronics are the only two foundries that sell 5- and 3-nanometer chips at scale to external customers, and Samsung has lost market share due to yield issues at 3 nm. Although Intel’s 14A node may emerge as a viable competitor to TSMC’s A14 node in 2028 with strong US government support, we think this will end up with Intel replacing Samsung as TSMC’s alternative instead of dethroning TSMC. The gap is starker with smaller foundries, as the latter tend to stop at 14 nm or 28 nm nodes, which were first mass produced over a decade ago.

Another of TSMC’s intangible assets is its close relationships with industry giants, which help justify investments in the next nodes. These relationships and TSMC’s lead in advanced nodes convince customers to share detailed road maps, which reduces the odds of over/underinvestment and helps optimize efforts to improve chip PPACT for various applications like consumer electronics, high-performance computing, and automotives. TSMC is the only leading-edge foundry that does not have its own branded products and has limited licensed intellectual properties, so customers can take comfort in the lack of conflict of interest and equal access to new technology that TSMC offers.

TSMC enjoys structural cost advantages compared with Intel and Samsung, especially in R&D. TSMC only serves external customers, while Intel and Samsung have to firewall client-owned designs from in-house ones. The firewall comes with additional compliance costs and potential duplicate efforts in R&D (at least two sets of designs for each product). The second cost advantage is TSMC’s massive scale (8 times Samsung, 6 times Intel Foundry) to push down burgeoning unit R&D costs for cutting-edge chips. The actual scale difference may be higher, as we estimate TSMC has an overwhelming 90% share in AI computing chips, which makes catching up more difficult than it seems. TSMC also spends almost double on R&D versus Intel, which gives it a better chance to address diverse client needs in cutting-edge AI, smartphone, and autonomous driving. The third advantage is lower R&D personnel costs in Taiwan. It would be hard for Intel to match TSMC’s costs, as it may face pushback from the government and other stakeholders if it outsources R&D to low-cost regions, including Taiwan.

TSMC shows signs of network effects and switching costs, but we reckon they are not durable enough to qualify as a moat source. Rather, we view them as customer considerations that reinforce TSMC’s intangible assets. Surging costs of designing cutting-edge chips make it costly to maintain multisourcing arrangements. IBS estimates chip design costs to exceed USD 500 million from 3 nm onward, and historically, newer nodes have always cost more despite more automation. This means that when TSMC wins orders for a chip model, customers are disincentivized to switch suppliers in the middle of a product cycle, which could be one year for smartphone chipsets and two years for AI data centers. Automotive and industrial chips have longer product lifecycles, with some exceeding 10 years to qualify as a stand-alone switching costs moat source, but they are cheaper to design with the use of mature nodes, so multisourcing is common.

A side effect of TSMC’s dominant market share is that the company doubles as a marketplace between electronic design automation firms and chip designers, which gives rise to some aspects of network effect. EDA firms offer intellectual property blocks that serve as ready-to-use portions of chip designs for mundane tasks and are specific to a single node of a single foundry. As a part of EDA firms’ revenue is based on licensed fees per chip sold, they prioritize IP blocks that are likely to be widely adopted. The wide clientele of TSMC makes it attractive for EDA firms to prioritize compatibility with TSMC’s processes, and TSMC’s current and prospective customers will find the wide range of available IP blocks valuable for product flexibility. The marketplace is especially strong at AI data centers, where TSMC has around 90% market share and bundles in-house advanced packaging offerings.

TSMC should consistently earn higher gross margins than competitors thanks to its economies of scale and premium pricing justified by cutting-edge process technologies.

TSMC wins when customers compete to offer the most advanced processing systems using the latest process technologies.

TSMC will benefit from more semiconductor firms embracing the fabless business model and internet giants designing their own data center chips.

Although TSMC is the foundry leader, each generation of process technology matures and commoditizes quickly, forcing the company to deal with pricing pressure.

The rise of AI may allow Samsung to bundle logic and memory chip sales and win customers through vertical integration.

Samsung and Intel are committed to heavy capital spending with the support of the US government. SMIC and other state-supported Chinese foundries also lurk as potential threats.