Taiwan Semiconductor (TSM) Stock Analysis — October 2026

Taiwan Semiconductor Manufacturing Company Limited, or TSMC, is the industrial partner that turns other companies’ chip designs into physical products. Its revenue comes from manufacturing, packaging, testing, and related engineering work, while its costs come from the factories, equipment, people, materials, and research needed to deliver them.

1.1 — What it is and what it sells

TSMC sells usable chips made to a customer’s design, not mainly a standard TSMC-branded chip.

A customer gives TSMC the plan for a chip’s circuits and functions. TSMC, acting as a foundry, manufactures that plan on a wafer, a thin, round slice of silicon. The factory builds many copies of the design on the wafer, then cuts them into individual dies. A die is the piece of silicon containing the chip’s circuitry. TSMC tests those dies and places working ones into protective packages that provide electrical connections.

The customer is therefore buying saleable output, not merely time on a machine. Some dies fail because of defects or electrical problems, so yield—the percentage that works—matters. Wafer size, die size, processing time, and yield determine how many chips TSMC can sell from each production run.

TSMC also sells access to its process technologies. A process technology is the manufacturing method used to create the circuits. A node, such as 3 nanometer, is mainly a name for a generation of that method, not a promise that every feature measures exactly 3 nanometers. A newer node aims to improve the trade-off among speed, power use, and chip area. TSMC’s N3E process, for example, is described as improving those measures compared with N5. Customers value not only the factory capacity but also the process and design tools that let them use it.

Packaging is the second major offering. TSMC’s 3DFabric family connects separate chips or pieces of chips so they work as one system. CoWoS places processor chips and high-bandwidth memory on an interposer, a dense connecting layer. InFO uses fine wiring in a fan-out package. SoIC stacks chips vertically, while COUPE combines optical components with electronic control chips. These methods can put more computing, memory, or communication capability into a package and move data across shorter connections.

The company reports one foundry segment, which includes its manufacturing and related work. The table shows foundry revenue, foundry operating income, and total revenue across the years shown.

Results by segment, as the company reports them

NT$ millions

That is the product: a customer’s design converted into tested dies and, when requested, a connected package. The next question is how the payments and costs move through that process.

1.2 — How it makes money

TSMC’s money comes in as customers buy preparation, wafer manufacturing, packaging, testing, and related services.

The relationship starts before a wafer enters the factory. A customer such as NVIDIA, a chip designer, supplies its design and works with TSMC to adapt it to a chosen process. TSMC may prepare masks, precision stencils that transfer each layer of the design onto the wafer. Because a chip has many layers, mask preparation is an upfront engineering task. TSMC can also offer shared prototyping in which several designs use one mask set, reducing the initial cost.

The main production bill covers the repeated work of making the wafer. TSMC uses specialized equipment to coat, print, etch, measure, and clean the silicon. That work consumes raw silicon wafers, gases, chemicals, photoresist, electricity, and water. If the customer buys the broader service, TSMC then packages and tests the resulting chips. Some work happens inside TSMC and some through outsourcing arrangements.

The cash does not necessarily arrive long before production. Customers generally do not place firm orders far in advance, although some provide temporary receipts to retain capacity. Those receipts help TSMC plan, but they are not the same as guaranteed final demand. The company must build factories and buy equipment before it knows exactly how busy they will be.

That timing creates operating risk. TSMC spent US$40.9 billion on capital expenditures in 2025. In its third-quarter 2026 call, it reiterated its capital budget for 2026 of US$60–64 billion. When utilization is high, committed factory costs spread across more wafers. When demand falls, buildings, tools, and staff still cost money while fewer wafers produce revenue. New technology ramps can also weigh on margins.

What remains after operating costs is operating income. In simple terms, the flow is: customers pay for preparation and production; TSMC pays for inputs, employees, research, equipment, and capacity; the remainder belongs to the operating result. The next question is who those customers are and how much bargaining power they have.

1.3 — Who pays

TSMC’s customers are companies that design chips, manufacture them, or do both.

A fabless company designs and sells chips but outsources factory production. An integrated device manufacturer designs chips and also operates some manufacturing of its own. Either type may hire TSMC for a particular design. Building a leading-edge factory requires enormous capital, specialized engineers, process development, and enough demand to keep the plant busy. Outsourcing lets a designer focus its resources on chip architecture and software while TSMC spreads factory costs across hundreds of customers.

The arrangement is closely coordinated. A customer adapts its design to TSMC’s process, while TSMC plans equipment and capacity around the customer’s needs. That creates a durable relationship, but it also means large customers retain negotiating power. In 2025, the largest customer generated 19% of revenue and the second-largest generated 17%. The ten largest customers together represented about 78% of revenue.

Customers also do not generally commit firm orders far in advance. Some provide temporary receipts to reserve specified capacity. The terms matter because the receipts can be refunded or offset against accounts receivable by mutual consent when the agreement’s conditions are met. A reservation therefore helps planning but is not identical to an irrevocable long-term order.

TSMC’s revenue table groups customers by region. The United States is the largest reported region in 2025, but that describes where customers are grouped, not necessarily where the wafers are physically made. The table shows how reported revenue is distributed among Taiwan, the United States, China, Japan, Europe, the Middle East and Africa, and other regions.

Revenue by region, as the company reports it

NT$ millions

TSMC is powerful because many designers need its scale and process expertise, but concentration means a change in a few major customers’ demand can affect revenue and seasonality. Testing is one of the services that helps complete those customer orders.

1.4 — Testing services

Testing verifies that the chips TSMC makes are ready for delivery.

Testing begins while the dies are still on the wafer. Electrical equipment checks which locations work and meet the required standards. After the wafer is cut, good dies are assembled into packages, and the finished packages can be tested again. Some testing is performed by TSMC and some is outsourced.

This two-stage check matters because a processed wafer is not automatically a usable product. Manufacturing defects or electrical failures can leave individual dies unable to perform, and assembly can introduce another point where inspection is needed. Testing protects the customer from receiving a batch that looks complete but contains unusable chips.

TSMC can sell testing alongside wafer fabrication and packaging, or include it in a broader manufacturing relationship. It is a revenue stream, although wafer fabrication remained approximately 86% of 2025 net revenue. Testing is smaller than the core factory business, but it helps TSMC deliver verified output.

The business therefore has a clear economic shape: customers bring designs and payments; TSMC converts them into tested, packaged chips; and the company spends heavily on the people, materials, research, equipment, and capacity required to make that conversion reliable.

Taiwan Semiconductor Manufacturing Company Limited, or TSMC, is the factory behind chips designed by companies such as Apple and Nvidia. Its market is large and growing, but its lead rests on a difficult promise: make other companies’ designs reliably, cheaply and at enormous scale.

2.1 — The market it sells into

TSMC sells manufacturing capacity, not a chip brand. A designer sends TSMC instructions for its chip, and TSMC supplies the factories, equipment, materials and expertise to reproduce that design on silicon wafers.

The market is large, but its size depends on what is counted. The worldwide semiconductor market excluding memory reached $611 billion in 2025. TSMC’s broader “Foundry 2.0” measure reached $305 billion in 2025 because it also includes packaging, testing, mask-making and other chip manufacturers.

That explains why TSMC can have a 40% share of one market and 70.4% of the top ten foundries’ combined revenue in the fourth quarter of 2025. The first figure uses a broad denominator; the second focuses on conventional foundry sales among the largest companies. Trade rules also shape this market because tariffs and incentives can change where customers build and source chips. The market’s size is therefore real, but its boundaries matter.

2.2 — Where the market is heading

Artificial-intelligence demand is pulling chip manufacturing toward advanced computing and packaging. AI accelerators perform the calculations used to train and run AI systems, while high-bandwidth memory supplies data quickly; advanced packaging brings these parts together.

Global semiconductor sales excluding memory rose 19% in 2025, and TSMC expects about 10% annual growth through 2030. But growth is uneven. High-performance computing made up 68% of TSMC’s revenue in the third quarter of 2026, while smartphones made up 21%.

That mix shows both the opportunity and the risk. Cloud companies can keep ordering chips if data-center investment remains profitable, but power, memory, networking or financing constraints could delay deployments. Consumer and price-sensitive products can weaken at the same time. AI may enlarge the market without making every part of it healthy.

2.3 — The rivals

Samsung Electronics Co. Ltd. is the leading alternative many customers would name first, while Intel Corporation is another major challenger with an advanced process and a foundry business. Their total-company revenue gives scale, but not a clean comparison with TSMC because Samsung also sells memory and electronics, while Intel sells chips under its own brand.

The process race shows why a public announcement is not the same as a proven substitute. TSMC’s N2 process entered high-volume manufacturing in the fourth quarter of 2025 and supplied 8% of wafer revenue in the third quarter of 2026. Intel’s 18A process entered high-volume manufacturing in late 2025, and Samsung planned to ramp mobile products on its second-generation 2-nanometer process in the second half of 2026.

Other foundries compete in older, specialized or regionally important technologies. They may not challenge TSMC for the newest AI chips, but they can win valuable work in particular products. The boundary between rival and supplier can also shift: GlobalFoundries agreed to supply silicon interposers for TSMC’s CoWoS packaging system.

The table lists the named competing companies and compares their financial and market measures.

The rivals

2.4 — What protects it

TSMC is protected by the hard combination of manufacturing know-how, scale, customer trust and switching costs. A yield is the share of chips on a wafer that pass testing, so a factory with better yield produces more saleable chips at a lower cost.

Moving a design is difficult. Engineers must adapt it to a new factory’s rules, build test versions, solve defects and qualify production. TSMC connected that process across 305 technologies, 12,682 products and 534 customers in 2025.

Its pure-play model adds trust. TSMC makes customers’ designs but does not sell competing chips under its own name. Samsung and Intel also serve outside customers, but they sell their own products, so neutrality can matter when customers share sensitive designs.

This protection demands constant execution. Advanced processes accounted for 74% of wafer revenue in 2025, so a slow or costly new process could pressure results before its capacity earns back its investment. The moat is strong because it must keep moving.

2.5 — How process advances shape competition

A smaller process protects TSMC only when it becomes dependable, affordable production. A process is the full recipe and design system used to build chips; its node number is mainly a generation label, not a complete measure of performance.

Yield improves during a process ramp, the period when a new technology moves from early production to large, repeatable output. TSMC’s N2 ramp reduced gross margin by 1.3 percentage points in the third quarter of 2026, even as the process contributed 8% of wafer revenue.

That trade explains the competitive test. TSMC, Intel and Samsung have all reached meaningful milestones, but customers care about performance, power, cost, yield and delivery at scale. A node wins only when engineers qualify it and customers put products into volume production. TSMC’s advantage is therefore execution, not the smallest number on a roadmap.

2.6 — Advanced packaging and chip stacking

Packaging has become part of chip performance and can become a bottleneck after wafers are finished. Modern AI products often combine logic chips, memory stacks and connecting layers in one package rather than using one large piece of silicon.

TSMC’s CoWoS service places system-on-chip dies and high-bandwidth memory on a silicon interposer, which acts like a dense miniature circuit board. This arrangement can provide more computing power and memory bandwidth, but it also requires suitable substrates, materials, testing and customer approval.

That creates a second supply chain. TSMC expanded CoWoS and SoIC volume production as AI demand grew, while GlobalFoundries agreed to make interposers for the CoWoS ecosystem. Packaging gives TSMC more work per finished system and closer customer ties, but it also makes shipments depend on constrained components and a rival supplier’s execution.

2.7 — Customer concentration and end-market cycles

TSMC has hundreds of customers, but its largest customers and their investment cycles still have an outsized effect on revenue. High-performance computing includes data-center processors and AI accelerators, while smartphones depend more on handset launches and replacement cycles.

The difference is visible in TSMC’s mix: high-performance computing represented 58% of revenue in 2025 and 68% in the third quarter of 2026. The ten largest customers accounted for 78% of 2025 revenue.

Many accounts therefore provide breadth, but a few programs provide much of the volume. Large customers help TSMC justify new factories because their product roadmaps offer visibility. They also create risk: a delayed launch, inventory correction or design change can leave expensive capacity underused. Switching is possible, but alternatives must have the right process and customer qualification.

2.8 — Taiwan production and disruption risk

Overseas fabs reduce geographic risk gradually, but Taiwan remains the center of TSMC’s manufacturing system. A fab is not merely a building: it needs trained staff, equipment, process recipes, electricity, water, chemicals, gases and nearby suppliers.

TSMC’s managed facilities had annual capacity above 17 million 12-inch-equivalent wafers in 2025. Its expansion includes production in Taiwan, China, Arizona and Japan, with additional overseas processes planned for later years. Taiwan was estimated to hold 66% of global advanced foundry capacity in 2024.

That concentration cannot be replaced immediately. An overseas fab must install the right process, produce reliable wafers and qualify them with customers before it can substitute for Taiwan. Earthquakes, droughts, power incidents or other utility failures could affect several connected sites and suppliers at once. Overseas investment improves resilience over time, but Taiwan still carries the system’s greatest concentration.

2.9 — Process advances and production scale

TSMC’s moat depends on turning expensive research and construction into reliable output that earns a return. Scale helps because the company can spread equipment, factory and development costs across many customers and products.

TSMC shipped 15.0 million 12-inch-equivalent wafers in 2025, compared with 12.9 million in 2024. It planned capital spending of $60 billion to $64 billion in 2026, while overseas fabs were expected to reduce gross margin by about 2 percentage points that year.

Customers do not always need the newest process. They can extend an older technology, divide a design into chiplets or use packaging to combine components. But products that value speed, power efficiency and density can justify the newest generation and help fund the next one.

The real test is the conversion from promise to economics: customer qualifications, design wins, improving yield, rising shipments, revenue contribution and margins after ramp costs. TSMC’s protection is not one invention or one factory. It is the accumulated system that makes advanced chips at scale, and keeps customers from wanting to leave.

Taiwan Semiconductor Manufacturing Company Limited, or TSMC, built its business by making chips designed by other companies. Its capital decisions follow from that choice: spend heavily on factories and technology, then ask whether the resulting capacity earns enough to support customers, shareholders, and the next investment.

3.1 — How it got here

TSMC’s first capital decision was to make other companies’ designs, not its own branded chips. When TSMC was founded in Taiwan in 1987, it created the “pure-play foundry” model: a company that manufactures chips for customers without competing with them through its own chip brand.

A chip is a tiny electrical circuit that processes, stores, or controls information. The customer designs that circuit and sends TSMC a detailed blueprint. TSMC builds repeated copies of it in layers on a round silicon wafer. It cuts the wafer into individual pieces, called dies, tests them, and packages the working ones into finished chips.

This arrangement lets several designers share an expensive factory. It is like several restaurants sharing one specialised kitchen: each supplies its own recipe, while the kitchen owner buys and maintains the ovens. TSMC combines orders from many customers, keeping its equipment productive while customers avoid building their own factories.

Trust therefore became part of the economics. Customers had to believe TSMC would protect their designs, avoid competing with them, and provide capacity when needed. Public listings in Taiwan in 1994 and New York in 1997 gave TSMC more access to capital as it expanded. By 2025, it made 12,682 products for 534 customers. Its original bet was working only because customer demand could keep its shared factories full.

3.2 — The big decisions

TSMC now invests in both the newest manufacturing technology and factories closer to customers, accepting higher costs for capacity and geographic reach. A fab is a highly controlled factory where machines build circuits layer by layer. Advanced fabrication can take 14 to 20 weeks. It can involve 400 to 1,400 steps, so a small defect can waste many dies.

That makes expansion a long-term bet. TSMC announced its first Arizona fab in 2020 and a second in December 2022. By July 2026, it had added US$100 billion to a stated U.S. investment plan of US$265 billion. The U.S. government also agreed to provide up to US$6.6 billion in direct support. The first Arizona fab entered high-volume production in late 2024, while the second targeted production in the second half of 2027.

The cost is already visible in profitability. Margin dilution means a new factory temporarily leaves less profit from each dollar of sales because construction, labour, utilities, logistics, and training cost more before the site is fully productive. TSMC expected overseas-fab expansion to reduce gross margin by about 2 percentage points in 2026, with the effect potentially reaching 3%–4% in the following years. The company accepts that cost for supply closer to customers and capacity outside Taiwan.

TSMC is also advancing its leading-edge process. Its 2-nanometer process entered high-volume manufacturing in the fourth quarter of 2025 with good yield. Yield is the share of chips that work well enough to sell. By the third quarter of 2026, 2-nanometer wafers generated 8% of wafer revenue, showing commercial adoption even though the investment’s full return remains unknown.

Finally, TSMC agreed to invest about ¥282 billion in a Japanese image-sensor venture with Sony, which develops the technology and will control the company. An image sensor turns light into an electronic picture. Volume production is expected in 2029, so this is a further manufacturing commitment, not yet a completed source of foundry revenue.

3.3 — Where the cash went

TSMC has mainly directed its cash toward factories and dividends while keeping a growing reserve. Free cash flow is the money left after operating costs and factory investment. Management can then retain it, pay shareholders, repay debt, or invest further.

The table tracks free cash flow and the payments, investments, financing, and other uses that followed it.

Follow the money: where the free cash flow went

NT$ millions

Capital expenditure is the main reinvestment engine. It pays for production facilities, equipment, and the research and development needed to enlarge or upgrade capacity. TSMC’s capital expenditure reached NT$1,272.4 billion in 2025, up from NT$956.0 billion in 2024. Its investing cash use also included financial assets, which are not the same as building a factory.

Debt adds financing but does not create chip revenue. TSMC issued and repaid debt in 2025 while its net cash—cash and short-term investments minus debt—rose. The company also approved a dividend of NT$7.00 per ordinary share for the second quarter of 2026 and retained a 2026 capital-spending budget of US$60 billion to US$64 billion. The important question is whether future operating cash flow can fund both commitments.

3.4 — Capital allocation history

TSMC’s long-term record shows a preference for manufacturing capacity and process technology, with dividends rising as the business generated more cash. Its original model turned one company’s factory spending into shared infrastructure for many chip designers.

That network is becoming global. Taiwan remains central, while Arizona, Japan, and Germany add capacity for different customers and markets. TSMC’s first Arizona fab and first Japanese specialty fab began volume production in 2024. But overseas diversification costs more, and the factories must still reach strong yields and enough utilisation to earn adequate returns.

The balance sheet shows that TSMC has generally expanded without exhausting its resources. Net cash increased across the five years through 2025 even as dividends rose. That supports the view that the company has funded major expansion from a powerful operating business rather than sacrificing financial resilience.

The harder conclusion is more limited. TSMC has repeatedly converted large commitments into working factories, advanced processes, and customers’ products, so the strategy has worked operationally. The final test is still ahead: whether Arizona, two-nanometer production, and the newer overseas and specialty projects generate returns that justify their cost.

Taiwan Semiconductor Manufacturing Company Limited, or TSMC, makes chips for other companies rather than selling products under its own name. To decide whether its leaders and owners are on an outside shareholder’s side, we must examine who controls decisions, who checks them and who bears the results.

4.1 — Who runs it

TSMC is run by C.C. Wei, a company veteran whose role gives him direct influence over both management and the board.

Wei spent most of his career inside TSMC, leading operations, technology, business development and the co-COO office before becoming CEO. The board elected him chairman while he remained CEO in 2024. His experience fits the company’s central problem: customers commit to chip production years ahead, while TSMC must spend billions before those chips reach volume production.

That combination can make decisions faster and more coherent, but it concentrates authority. The same person leads the executives doing the work and the board meant to supervise them, so succession planning and independent questioning matter more.

The rest of the senior team also has deep company experience. Y.P. Chyn brings experience in product development and overseas operations; Y.J. Mii came through research and development; and finance chief Wendell Huang has handled corporate finance inside and outside TSMC. Together, their backgrounds match a business that must coordinate factories, research, equipment, finance and customer commitments.

The table lists the named officers, their roles and the dates they began serving.

Who runs it

That experience gives TSMC operational depth, but the board must ensure that depth does not become dependence on one leader.

4.2 — The board

TSMC’s board has a large independent majority, but its combined chair-and-CEO role makes genuine challenge especially important.

The board appoints and can dismiss officers, reviews management performance and succession, and oversees strategy, risk, capital spending and dividends. Its audit, compensation, and nominations committees give those responsibilities a formal structure.

Several independent directors have led major companies outside semiconductor manufacturing, including Xerox, Sunoco and Applied Materials. Their experience can help them question technology spending, capital discipline and executive performance. But independence is a formal classification, not a promise that a director will disagree when disagreement is needed.

The National Development Fund, Executive Yuan, Taiwan’s government investment fund, also has a seat. That makes public-policy interests visible inside the boardroom, but it remains a minority voice. The independent directors therefore provide an important counterweight, provided they control their own agenda and continue to question management.

The table lists each director, the year service began and whether the director is classified as independent.

The board

The board is the main check on Wei’s concentrated authority, and ownership determines who ultimately chooses that board.

4.3 — Who owns it

TSMC’s ownership shows a meaningful Taiwanese government connection, but no controlling shareholder in the figures shown.

The National Development Fund is the largest named holder. Its stake gives the government a voice and a board representative, but not the power to dictate company decisions alone. That distinction matters: influence can serve public interests without amounting to control.

TSMC has ordinary common shares listed in Taiwan. U.S. investors can buy American Depositary Shares, or ADSs, on the New York Stock Exchange. An ADS represents underlying common shares, so it does not create a separate operating claim. Holders can receive dividends and vote, subject to the depositary process, but an ordinary minority investor cannot direct factory investments or elect the board alone.

Senior executives own shares, yet none approaches the position of a controlling founder. Their holdings give them financial exposure to TSMC’s performance, while the dispersed ownership base leaves management, the government fund and public investors reliant on governance checks rather than personal control.

The table lists the fund, executives and other principal holders, their roles, share classes, values, recent transactions and dates.

Who owns it: insiders and principal holders

Ownership therefore creates alignment, but insider transactions show how individual leaders actually act on that alignment.

4.4 — Insider buying and selling

The disclosed insider trades are small personal signals, not a reliable vote of confidence from the whole leadership team.

One operations vice president sold ADS equivalents, while several executives and a director bought them. A purchase means an insider used personal money to acquire shares; a sale means an insider reduced a holding. Neither explains the person’s motive by itself.

Trade size, frequency, role and total holding provide the useful context. A series of routine purchases by one executive does not equal a coordinated commitment by the leadership team. One sale does not establish that the seller expects the business to weaken. The trades matter, but they are weaker evidence than the company’s compensation design and long-term results.

4.5 — Pay, and promises kept

TSMC links executive pay to company performance through salary, bonuses, profit sharing and stock awards, but stock awards alone do not prove strong alignment.

The company added an annual long-term incentive plan in 2025 for executives and critical employees. Awards depend on individual and company performance, shareholder returns relative to peers and ESG results, and are ultimately used to buy TSMC shares. That structure connects pay with several measures rather than one short-term target.

The table shows the stock awards, buybacks, cash-flow comparisons and changes in the share count.

Stock pay and the share count

NT$ millions

TSMC has also set ambitious financial goals. An earlier target called for 10–15% annual U.S.-dollar revenue growth through 2025. It also called for ROE of at least 20% across cycles. In 2025, revenue grew 35.9% in U.S. dollars. Gross margin reached 59.9% that year.

The current objective is revenue CAGR approaching 25% over the 2024–29 period. It also targets gross margin of at least 56% and high-20s ROE through the cycle. These measures answer different questions: margin shows what remains after manufacturing costs, ROE measures returns on owners’ capital, and free cash flow shows what remains after operations and investment. A strong year cannot by itself prove that a multi-year promise has been kept.

Those promises matter because TSMC’s largest management decisions involve spending enormous sums before the returns are certain.

4.6 — Management’s investment choices

TSMC’s biggest bet is that demand for advanced chips will remain strong enough to justify a vast expansion of manufacturing and packaging capacity.

A fabrication plant, or fab, makes chips on silicon wafers. A node such as 3-nanometer or 2-nanometer describes a generation of manufacturing technology, not simply a measurement. Building a fab requires land, clean rooms, specialized equipment and years of process development.

In July 2026, Wei said TSMC would invest an additional US$100 billion in Arizona, including fabs for advanced technologies and advanced-packaging facilities. The company also raised its capital budget to US$60–64 billion. CFO Wendell Huang said spending from 2026 to 2028 would be significantly higher than in the previous three years.

The strategy brings TSMC closer to U.S. customers and government partners, which can strengthen supply-chain resilience. But TSMC says overseas fabs can dilute margins. The test is whether customer demand and factory utilization justify the extra cost, and whether packaging capacity arrives alongside wafer capacity.

TSMC’s pure-play foundry model means it manufactures customers’ designs without selling competing designs under its own brand. In 2025 it made 12,682 products for 534 customers. That neutrality encourages chip designers to trust TSMC with valuable intellectual property, but it also makes customer road maps essential evidence for new capacity.

AI processors show why packaging matters. They may combine computing chips with high-bandwidth memory, and TSMC’s CoWoS service places those components together on an advanced interconnection layer. TSMC can therefore have enough wafer capacity but still fail to ship enough complete processors if packaging is short.

Management is betting on durable demand, disciplined construction and successful yields. For shareholders, the central question is whether TSMC can preserve its returns while expanding around the world.

TSMC’s accounts describe a contract manufacturer whose growth depends on technology and factory discipline, not on selling a branded chip. Read together, the statements show a powerful cash-generating business making an unusually large bet on future capacity.

5.1 — Growth

TSMC grows by making more wafers, moving customers to newer processes and serving more demanding computing markets.

TSMC is a foundry: a contract manufacturer for chip designers. NVIDIA, Apple and AMD design chips, then pay TSMC to manufacture them. TSMC owns the factories, equipment and production expertise, while customers bring the designs. Its growth therefore depends on winning programs and keeping its factories full.

A wafer is a thin silicon disk containing many copies of chips. TSMC reports shipments in 12-inch-equivalent units so different wafer sizes can be compared. A process node such as 7-nanometer or 5-nanometer names a generation of manufacturing technology. Newer generations can fit more transistors into less space or use less power, but they can also begin with poor yields, meaning that too many chips fail testing.

The process mix has moved toward newer technology. In 2025, advanced technologies of 7-nanometer and smaller represented 74% of wafer revenue, while 3-nanometer alone represented 24%. That is revenue mix, not profit by process: expensive tools and weak early yields can delay the economic payoff.

Demand has shifted as well. High Performance Computing, or HPC, includes data-center processors, accelerators, networking and scientific systems; artificial intelligence is one part of it, not the whole category. HPC’s share rose while smartphones’ share fell, changing the source of demand. Revenue growth in U.S. dollars also exceeded growth in New Taiwan dollars during 2025, so currency translation reduced reported New Taiwan dollar revenue growth.

The results table records revenue, growth, profit, margins, earnings per share and dividends from 2021 through 2025.

Results

NT$ millions

The company’s own-measures table records wafer shipments and the process and end-market mix from 2021 through 2025.

The company's own measures

The growth story therefore ends with a test: do richer products and fuller factories produce durable profit?

5.2 — Profit

TSMC’s margins show the power of scale, but reported net income can contain gains that have little to do with wafers.

Gross margin is the share of sales left after production costs. Operating margin also subtracts research and other operating expenses; net margin includes taxes, financing and other items. In 2025, operating income grew faster than revenue, which means the company converted more sales into operating profit as demand and mix improved.

Factory loading helps explain that result. A fab carries buildings, tools and engineers whether production is high or low. When more wafers pass through, fixed costs are spread across more output. A new process can reverse the benefit at first because equipment is expensive and yields are still improving.

Net income needs a closer reading. Tax true-ups reduced tax expense in each year from 2023 through 2025, making reported earnings and margins better than they would have been under a normalized tax rate. Earthquake damage, a later insurance settlement and the sale and remeasurement of TSMC’s Vanguard investment also affected comparisons. The investment gain increased earnings without showing stronger foundry sales.

TSMC is evaluating IFRS 18, a new accounting presentation rule taking effect in 2027. Future operating-profit comparisons may therefore need to be reconciled with earlier presentations.

The adjusted table sets out tax, income and the company’s other adjusted measures from 2021 through 2025.

The company's own adjusted measures, and tax

NT$ millions

Effective tax rate — Earlier tax benefits of NT$32,775.5 million, NT$26,042.7 million and NT$56,886.1 million lowered tax rates; without them, rates were 16.4%, 19.5% and 19.8%, respectively.

Profit is meaningful here, but the next question is how much of it becomes cash after the factories are funded.

5.3 — Cash

TSMC turns operating profit into cash, but its factories consume a large share of that cash.

Cash from operations begins with earnings and adjusts for non-cash expenses and changes in receivables, inventory and other working-capital items. Free cash flow is what remains after capital spending: the cash paid for factories, tools and other long-lived assets.

Stock-based pay is deducted to produce “real free cash flow,” because paying employees in shares does not use cash immediately but does dilute shareholders. The adjustment is small here compared with the company’s overall cash generation.

The cycle matters. Real free cash flow was much stronger in 2025 than in 2023, but capital spending also remained enormous. Working capital tied up cash in 2025 after releasing cash in 2024. That movement can separate cash flow from profit for a year without changing the underlying factory economics.

The free-cash-flow table follows operating cash, working capital, capital spending, stock pay, free cash flow and the company’s own count from 2021 through 2025.

Follow the money: real free cash flow

NT$ millions

Strong cash generation makes expansion possible; the balance sheet shows how much of that future cash is already spoken for.

5.4 — Balance sheet and debt

TSMC has ample liquidity, but its purchase commitments reveal a substantial future claim on cash.

Using December 31, 2025 balances, cash and short-term investments exceeded debt by a wide margin. The current ratio, which compares current assets with current liabilities, was comfortably above one, so near-term liquidity is not the main constraint on the factory plan.

TSMC’s debt is mainly fixed-rate New Taiwan dollar and U.S.-dollar corporate bonds. Rates ranged from 0.41% to 4.63%, and maturities ran from less than one year to 35 years. Fixed rates reduce the immediate effect of higher interest rates, although maturing bonds still require repayment or refinancing.

Purchase commitments are different from debt. They cover equipment and other purchases not yet received, so they are not yet recorded as liabilities. At year-end 2025, these obligations totalled NT$1.535 trillion, with NT$1.220 trillion due within a year. They show where part of the cash cushion is already headed.

The balance-sheet table sets out liquidity, assets, liabilities, debt, leases, equity and related measures from 2021 through 2025.

Balance sheet

NT$ millions

The balance sheet can fund the plan; the latest quarter shows whether the plan is already earning its cost.

5.5 — The latest quarter

The latest quarter showed powerful year-over-year growth, alongside an early margin cost from the two-nanometer ramp.

In the quarter ended September 2026, revenue rose 51% from a year earlier. Operating income, net income and earnings per share rose even faster, while operating cash more than doubled and free cash flow also improved. The strength therefore reached sales, profit and cash rather than appearing in only one line.

The sequential picture was less smooth. Gross margin fell by 1.3 percentage points from the June quarter to the September quarter. TSMC attributed the decline to the initial two-nanometer ramp. Two-nanometer represented 8% of wafer revenue in the third quarter of 2026, while three-nanometer remained a larger contributor.

That combination is normal for a new process: TSMC must absorb equipment, early yields and ramp costs before volume and efficiency improve. Overseas expansion may add further pressure in the next few years. The next test is whether two-nanometer scale and stronger loading recover the margin.

The last-eight-quarters table sets out revenue, margins, profit, earnings per share, operating cash and free cash flow.

The last eight quarters

NT$ millions

The quarter confirms demand; utilization will determine whether the new capacity converts that demand into returns.

5.6 — Factory utilization and capacity

TSMC earns its investment back only when its expensive factories and packaging lines stay busy.

Utilization is the share of available capacity actually in use. A fab’s buildings and equipment are fixed costs, so high utilization spreads them across more wafers and makes profit rise faster than sales. Low utilization has the opposite effect.

TSMC’s managed manufacturing capacity exceeded 17 million 12-inch-equivalent wafers in 2025. Shipments were lower, but those figures do not create a precise utilization rate because their timing and scope may differ. A company-wide average could also hide a full three-nanometer line beside an underloaded older process.

Demand planning reduces, but does not remove, the risk of overbuilding. TSMC studies customer demand, data-center construction, power and component availability before approving investment. Customer plans are not necessarily binding purchase guarantees, and demand can still arrive late.

Packaging can become the next bottleneck. CoWoS places multiple chips and memory close together so they can communicate quickly. TSMC can have enough wafers but still lack enough packaging capacity to deliver a finished accelerator. TSMC was working to double CoWoS capacity, while demand remained higher than supply in the third quarter of 2026.

Capacity must therefore be judged by output and loading, not by buildings alone. That leads to the central investment question: what return will the new factories earn?

5.7 — Returns on factory and equipment investment

TSMC’s investment case depends on turning heavy spending into productive, well-loaded capacity before the technology cycle changes.

A new fab requires cash years before it sells wafers. Capital expenditure rose from 2021 through 2025, while real free cash flow also strengthened. That shows the existing business can fund expansion, but it does not prove that the newest factories will earn attractive returns.

TSMC’s own test is whether revenue grows faster than the capital base and whether new tools become productive. Management expects capital spending from 2026 through 2028 to be significantly higher than in the prior three years, while arguing that revenue growth should outpace it if execution is strong.

The optimistic case is that artificial-intelligence demand, customer commitments and improving yields fill the new capacity. The sceptical case is that customers delay products, demand forecasts overshoot or overseas fabs ramp slowly. In that case, cash leaves first and depreciation rises before revenue arrives.

The evidence will be practical: shipments after the build, loading of two-nanometer and CoWoS capacity, and operating cash that continues to cover investment without weakening the balance sheet. TSMC can afford its plan. Only later cash generation will show whether the plan was productive.

Taiwan Semiconductor Manufacturing Company Limited, or TSMC, is a foundry: it manufactures chips designed by other companies. Its stock price reflects not only today’s profits, but also a bet that its technology, factories, and customer relationships will keep earning attractive returns.

6.1 — What an investor pays today

TSMC’s current valuation combines a valuable operating business with a large cash cushion, while asking investors to fund an expensive factory cycle.

Enterprise value estimates what investors are paying for the operating business after subtracting cash and adding debt. If a shop costs $100 but contains $20 in cash and carries no debt, its business has effectively cost $80. TSMC’s September 30 balance sheet showed more cash and marketable securities than interest-bearing debt, so its enterprise value is below its stock-market value. Lease liabilities can change that comparison if one calculation treats them as debt and another does not.

P/E, or price-to-earnings, compares the value of the shares with profit attributable to shareholders. Its forward figure is below its trailing figure, which means investors expect earnings to rise. EV/revenue compares the operating business with sales, but says less about how much profit those sales produce.

The cash cushion helps, but TSMC’s factories need repeated spending on equipment and construction. Look across the measures for the balance between cash, expected earnings, and reinvestment.

What an investor pays today

US$ millions

The next question is whether this valuation looks unusual against TSMC’s own past.

6.2 — Against its own past

TSMC’s history shows that a multiple can fall when profits are temporarily squeezed and recover when growth looks durable.

Semiconductor factories have heavy fixed costs. A fab is like a hotel whose building, machinery, engineers, and utilities must be paid for even when many rooms are empty. When customers cut orders, lower production spreads those costs across fewer wafers, so profit falls faster than sales.

Worldwide semiconductor sales fell 8.2% in 2023 as customers worked through excess inventory. TSMC’s valuation contracted during that adjustment, then recovered as demand for advanced processes and artificial-intelligence computing strengthened.

The earnings recovery also supported the re-rating. From 2024 to 2025, TSMC’s revenue rose 31.6% and net income rose 46.4%. Investors were therefore paying for a business they believed had both stronger profits and a more durable mix of leading-edge work.

The table shows the five-year pattern. A year-end P/E cannot by itself separate earnings growth from investors’ willingness to pay a higher multiple.

Five years of history

share prices in US$ per ADS

That cycle matters when TSMC is compared with rivals.

6.3 — Against rivals

TSMC’s premium reflects its position as the leading manufacturer of chips designed by other companies, but the comparison is not simple.

A company such as Nvidia or Apple may design a processor without owning the enormous factories and production expertise needed to make it. TSMC turns those designs into physical chips on silicon wafers. Customers qualify its processes over years, so scale, reliable yields, and trust become difficult for competitors to copy.

The closest peers are pure-play foundries, which manufacture customer designs rather than selling their own branded processors. Even they differ: UMC and GlobalFoundries focus more on mature and specialty processes, while SMIC faces a different technology and market environment. Integrated companies such as Samsung and Intel also have businesses beyond contract manufacturing.

TSMC captured 72.5% of global foundry revenue in Q2 2026. In 2025, processes of 7nm and smaller generated 74% of its wafer revenue. Labels such as 3nm and 5nm identify generations of manufacturing technology that pack more computing into less space and can reduce power use. They also demand greater research, capital spending, and manufacturing skill.

The table compares TSMC and its rivals by revenue, margin, market value, and trailing P/E.

The rivals

The premium is therefore a claim about durable technological and customer advantages, not proof that any particular multiple must last.

6.4 — How the price moved

TSMC’s share price has risen sharply, but the path reflects both enthusiasm about AI demand and sudden changes in risk.

The share-price tables show a strong long-term advance alongside recent volatility. The PHLX Semiconductor Index, a benchmark for chip stocks, gained 47.05% over the trailing three years through October 16, 2026, while TSMC’s performance was much stronger.

Artificial-intelligence computing has become a concrete demand driver. In Q3 2026, high-performance computing supplied 68% of revenue. CoWoS, TSMC’s advanced packaging method, places powerful logic chips close to high-bandwidth memory so they can communicate quickly. Because AI chips are large and demanding, packaging capacity can become a bottleneck after wafer production.

The stock also responds to events outside TSMC’s orders. Tariff uncertainty weighed on shares in April 2025, showing how policy headlines can change the price before earnings change. TSMC represented 8.5% of the PHLX Semiconductor Index at the end of July 2025, so broad index buying or selling may amplify movements, though it does not prove the cause of any one move.

Look first at the recent trading range.

The share price today and over 52 weeks

US$

Then compare the stock’s returns across the different periods.

Share price performance

The price history leads to the central question: what future does today’s price require?

6.5 — What the price implies

Today’s valuation assumes that TSMC can grow through a major investment cycle without giving away the returns that made it valuable.

At the October 18, 2026 snapshot, a forward P/E of 24.2 means investors pay $24.20 for each dollar of profit expected over the next year. That can make sense if profit soon grows to $2, but not if it remains at $1. The ratio therefore prices profits beyond the forecast year.

Recent results support the growth case. In Q3 2026, revenue rose 50.9% from a year earlier, while 7nm-and-smaller technologies supplied 79% of wafer revenue. But TSMC expects a 2026 capital-spending budget of US$60–64 billion.

That spending hurts cash flow before it helps earnings. TSMC pays for land, buildings, and equipment upfront, then records machinery costs gradually through depreciation as the fabs begin operating. Overseas expansion adds another test: management expects about 2% gross-margin dilution in 2026, with 3–4% possible in following years.

The price works if new fabs fill with valuable work and eventually earn enough to fund the next generation while rewarding shareholders.

6.6 — How to read valuation measures

No single ratio tells you whether TSMC is cheap, because each one sees a different part of the business.

P/E tests how much investors pay for profit, but profit changes with utilization, pricing, product mix, currency, and the semiconductor cycle. Enterprise value accounts for debt and cash, while EV/revenue compares the operating business with sales. EV/EBITDA excludes depreciation, which can make a fab owner look inexpensive even though its equipment wears out and must be replaced.

Price-to-book shows the value assigned to TSMC’s recorded industrial assets, but not fully to its process knowledge or customer qualification. Free-cash-flow yield shows cash left after capital spending, yet one year can look weak simply because equipment bills arrived early. Dividend yield shows cash returned to shareholders, not the cash still needed for expansion.

The right conclusion comes from cross-checking the measures: profitability, operating value, assets, cash generation, and distributions should tell a consistent story.

Taiwan Semiconductor Manufacturing Company Limited, usually called TSMC, makes chips for other companies rather than selling a competing chip brand. Its next phase combines surging demand for advanced computing with a costly expansion that must eventually earn its keep.

7.1 — What management has guided

TSMC’s own forecast says demand is strong enough to justify faster growth and heavier investment.

Management expects revenue in the coming quarter to rise about 14% from the previous quarter. Its outlook also gives gross margin, the share left after manufacturing costs, and operating margin, the share left after operating costs; both may weaken as new processes and overseas fabs ramp.

The annual forecast has moved steadily higher. TSMC began the year expecting 2026 revenue growth above 30%. By October, it guided to 44% growth in U.S. dollar terms. That sequence shows management committing money before all the new capacity is producing chips. The confidence is strongest in artificial-intelligence and high-performance computing, while consumer and price-sensitive markets face higher component prices and economic uncertainty.

The guidance table shows the assumptions and targets that support this plan.

Guidance in force

7.2 — What analysts expect

Analysts broadly expect TSMC to keep growing, but they disagree about how large the next investment wave must be.

Guidance is TSMC’s forecast, analyst estimates are outsiders’ models, and actual results are what the company eventually reports. For the coming quarter, TSMC guided to US$52.3–54.3 billion of revenue. Analysts make their own forecasts from assumptions about customer orders, artificial-intelligence demand and construction speed.

The latest quarter exceeded both analyst expectations and the upper end of TSMC’s earlier revenue guidance. That supports the near-term demand story, but one result cannot establish a recurring pattern.

Analysts’ spending models for the next two years vary widely. Outside estimates range from US$81 billion to US$105 billion. These are scenarios, not purchase orders. They may prove conservative if demand stays strong, or aggressive if customers slow investment or data-center power becomes a bottleneck. The central uncertainty is therefore not whether TSMC plans to spend more, but how much demand the added capacity will actually serve.

7.3 — The spending ahead

TSMC is a contract manufacturer: customers design chips, and TSMC manufactures them for a fee.

A customer supplies a design, chooses a manufacturing process and receives wafers or packaged chips that can go into servers, phones and other products. A fab is a chip factory. A wafer is a round silicon disk containing many chip copies, called dies. TSMC must expand the factory, equipment and packaging stages together.

The 2026 capital budget focuses mainly on advanced process technologies, with further spending on specialty technologies, packaging, testing and mask making. A nanometer node is a manufacturing generation; newer generations can make more capable or efficient chips, but require expensive new equipment and recipes.

TSMC said 2-nanometer contributed 8% of Q3 wafer revenue. Its ramp initially reduced margins because yield—the share of working dies—starts lower and improves as engineers learn the process.

The company is also adding three new 3-nanometer fabs despite its former practice of stopping expansion once a node reached its target capacity. That decision reflects strong demand, but it creates risk if customer plans change. Advanced packaging and testing add another constraint, while customers must still build and power the data centers that use the chips. TSMC’s checks on those sites connect factory spending to the physical reality behind customer forecasts.

7.4 — Can they pay for it

TSMC can fund the expansion, but a fab consumes cash years before it produces its full revenue.

The company first pays for land, buildings and specialized equipment. It then qualifies processes and improves yields before customer orders generate the expected cash. In Q3 2026, operations generated about NT$920 billion. The company spent NT$443 billion on capital expenditure and paid NT$156 billion in dividends.

At the latest spending pace, net cash equaled about 6.7 quarters of capital expenditure. That is a static comparison, not a forecast: it ignores future operating cash, changing investment, borrowing and dividends. TSMC can also meet near-term debt maturities from current resources and future operating cash. The real question is whether the new factories will produce enough profit to justify the capital tied up in them.

The first table shows cash, investments, debt and net cash across recent quarters.

Cash and debt, last eight quarters

NT$ millions

The second table isolates the debt coming due within twelve months.

Debt coming due after the latest fiscal year end

NT$ millions

7.5 — What to watch

The next results must show whether demand and infrastructure are keeping pace with construction.

Monthly sales and quarterly revenue will test the outlook, but margins matter too because new nodes and overseas fabs can make rapid growth less profitable. TSMC’s checks on U.S. data-center construction and power supply provide a physical test of customer forecasts: desired chips cannot become sustained revenue until customers can install and operate them.

Regulatory permissions and customer screening create another risk. In October 2024, TSMC disclosed a possible diversion of a customer chip to a restricted entity. Its Nanjing factory now relies on a renewable annual U.S. export license after its earlier authorization ended in December 2025.

A patent investigation involving Marlin, a patent claimant, ended on August 6, 2026, after the parties acted jointly, but the settlement terms remain undisclosed. These matters could affect shipments or costs, so investors must watch for disclosed outcomes rather than assume a numerical impact.

7.6 — Overseas expansion: cost and value

Overseas fabs offer local supply and strategic flexibility, but they begin with an economic disadvantage.

A new factory must recruit workers, train teams, qualify equipment and build a supplier network. Its fixed costs are spread across fewer wafers until production reaches scale, so each chip costs more than one made in an established Taiwan fab.

TSMC expects overseas fabs to dilute gross margin by about 2% in 2026. It says the effect could widen to 3%–4% in the following years as more U.S. capacity ramps. Customers may accept higher prices for local supply, and governments may provide incentives, but neither support proves that a fab will earn a competitive return.

The test is therefore twofold. Overseas capacity must deepen customer relationships and geographic resilience, while improving its yields, utilization and cost structure enough to justify the capital invested.

7.7 — Growth and capital measures to follow

The best evidence will show whether new capacity becomes saleable chips rather than merely impressive construction.

Yield measures how many dies on a wafer meet specifications. A new node usually begins with lower yield, then improves during its ramp. High and rising yield matters because it turns more of each costly wafer into products customers can buy.

Investors should also follow utilization, process mix and end-market mix. Broadening demand makes growth less dependent on one application or buyer, while concentration makes it more fragile. TSMC reported 305 distinct process technologies. It manufactured 12,682 products for 534 customers. Those figures show breadth, but they do not reveal how much revenue comes from any one customer or process.

The useful conclusion is not simply that TSMC is adding capacity. It is whether customers use that capacity, whether yields become competitive and whether the resulting revenue earns an adequate return.

7.8 — Usual financial measures

The right financial measures connect TSMC’s growth to the assets and cash required to produce it.

Revenue growth shows expanding demand, but it can also reflect higher prices or a richer process mix. Gross margin reflects factory utilization, process mix, currency, ramp costs and location. Operating margin adds research and other operating costs, while net margin also reflects tax, financing and currency effects.

Free cash flow is operating cash less capital expenditure. It may fall during a fab buildout before the new capacity produces revenue. Net debt to EBITDA compares debt after cash with earnings before interest, taxes, depreciation and amortization; interest coverage and the current ratio show additional financing flexibility.

Return on equity, dividend yield, price-to-earnings and EV/EBITDA help describe shareholder returns and valuation. None can show whether a new fab will earn its cost of capital. That answer depends on utilization, yields, future revenue and the timing of investment.

8.1 — The verdict

The probability-weighted forecast reaches 468.34 in October 2027, for a 15% return from today including dividends. It reaches 580.53 in October 2028, a 43% return. By October 2029, it reaches 742.49, an 84% return.

The 2029 range is 298.51 to 1163.24. Today’s reference price is 412.61. The spread reflects the possibility that both earnings and the valuation multiple move sharply in opposite directions.

What the shares could be worth

USD per TSM depositary share at each date; in brackets the return since today's 412.61, dividends included

8.2 — The model

Prices and per-share figures in this section are converted from TWD at 0.0313 USD per TWD (ECB euro reference rates and the Federal Reserve's H.10 rate, 9 October 2026); one TSM depositary share represents 5 shares of the issuer.

The forecast uses an enterprise-value approach because TSMC is a high-capex wafer manufacturer. Capital spending was about 33% of revenue in 2025, well above the 10% threshold, so EBITDA and net cash capture the economics more directly than an equity-only measure.

At each date, the forecast applies an EV/EBITDA multiple to the preceding twelve months’ EBITDA, subtracts net debt, and divides by the projected share count. The multiple therefore captures changing expectations about growth, utilization and execution, while net cash gives TSMC credit for cash accumulated after investment and dividends.

The share count moves only slightly. Awards create dilution, while occasional repurchases can offset it; no recurring buyback programme or new share sale is assumed. That restraint fits the company’s history, where diluted shares changed very little across the cycle.

8.3 — The drivers

EBITDA is the main source of upside and downside. Base EBITDA is TWD 4,500,000m in October 2027 and 5,475,000m in October 2028. It reaches TWD 6,512,000m in October 2029. The bear row falls to TWD 3,575,000m in 2028, while the bull row reaches TWD 6,080,000m. The bear assumes weaker utilization and ramp costs; the bull assumes strong AI demand, better N2 yields and fuller CoWoS capacity.

Estimated EBITDA fell about 7% in 2023. It then rose 36% in 2024 and 32% in 2025, showing both cyclical risk and operating leverage. Management’s 44% 2026 revenue-growth outlook and planned doubling of N2 capacity in 2027 support the base growth path, but overseas fabs and new-process ramps make the margins deliberately conservative.

The base share count rises from 25,934m in 2027 to 25,936m in 2028. It reaches 25,938m in 2029. The bear ends at 25,939.5m, while the bull ends at 25,935m. The latest outstanding and diluted count is 25,932m, and the historical diluted average changed only 0.3m between 2021 and 2025. The paths therefore allow for awards without making dilution a major driver.

Base dividends per share are 28 in 2027 and 32 in 2028. They reach 36 in 2029. The bear holds the annual rate at 28, while the bull reaches 44 by 2029. Annual dividends rose from 11 in 2021–22 to 22 in 2025, and the 2026 target is 24. The policy of steadily increasing dividends supports the base case, but heavy investment justifies the bear case.

The base multiple is 17x in 2027 and 18x in 2028. It reaches 19x in 2029. The bear falls to 10x in 2028, while the bull reaches 25x in 2029. Today’s 18.41x exceeds the five-year average of 14.89x, although leading-edge scale and AI exposure justify some premium. The historical fall to 7.10x in 2022 shows how quickly valuation can contract when the cycle turns.

Base net debt moves to -3,793,000m in 2027 and -5,242,000m in 2028. It reaches -7,384,000m in 2029. The bear and bull rows reflect slower or faster cash accumulation. TSMC starts with net debt of -2,985,401m, while 2025 operating cash flow exceeded capital spending by a substantial amount. The forecast assumes elevated capex, 3% interest, 18.1% depreciation and amortization relative to revenue, and a 17.5% tax rate, so cash growth remains sensitive to utilization.

The forecast's table

base case, then bear / bull in brackets; TWD millions, per share in TWD, unless shown

8.4 — Bear and bull

The bear case requires AI investment to slow, data-center customers to defer deployments, and new capacity to remain underloaded. The shares fall from 316.21 in 2027 to 298.51 in 2029. Its probability is 25% because TSMC has already experienced a material downturn, while high fixed investment increases the cost of weaker utilization.

The base case requires strong advanced-computing demand, successful N2 and CoWoS expansion, and gradual absorption of overseas-fab costs. It rises from 484.67 in 2027 to 791.30 in 2029. Its 55% probability reflects strong current guidance balanced against the present premium valuation and execution risk.

The bull case requires demand, N2 yields, packaging throughput and utilization to outperform together. It rises from 613.62 in 2027 to 1163.24 in 2029. Its probability is 20% because the upside depends on several uncertain successes while capex is increasing sharply.

8.5 — What would change the view

The forecast would move toward the bear case if customer capex or data-center power constraints delayed AI orders in 2027–2028. A slow overseas-fab ramp, higher N2 costs or weak utilization through 2029 would reduce margins, cash flow and the valuation multiple.

It would move toward the bull case if AI and HPC demand exceeded expectations, N2 capacity doubled with improving yields, and CoWoS expansion raised saleable output in 2027–2028. Sustained growth and strong returns on the investment programme would support a premium multiple by October 2029.