When I look at Taiwan Semiconductor Manufacturing Company, better known to American investors by the ticker TSM, I see one of the most impressive businesses in the world sitting in one of the most complicated positions imaginable.
TSMC manufactures the advanced chips powering artificial intelligence accelerators, smartphones, data centers, high-performance computers, vehicles, and more devices than most of us realize. It operates the factories behind many of the technology companies that receive far more public attention. Those companies design the glamorous products. TSMC performs the extraordinarily difficult work of turning the designs into functioning silicon at scale.
This arrangement has made TSMC indispensable.
It has also made the company almost impossible to analyze with ordinary investing language.
I cannot evaluate TSM simply by looking at revenue growth, margins, and earnings estimates. Those numbers matter, but they do not tell the whole story. I also have to consider military tension across the Taiwan Strait, export controls, government subsidies, overseas construction, electricity and water availability, advanced packaging capacity, supplier concentration, customer bargaining power, and the possibility that artificial intelligence demand is simultaneously real, enormous, and capable of inspiring reckless spending.
In other words, TSM is not the kind of stock I can analyze while pretending the outside world is an irrelevant distraction.
The outside world is part of the investment thesis.
That is what makes TSM fascinating. It may be one of the clearest ways to participate in the growth of advanced computing, yet it comes attached to risks that cannot be modeled neatly in a spreadsheet.
I can calculate a range of earnings. I cannot assign a precise probability to history deciding to behave badly.
The Business Is Extraordinary
Before discussing the risks, I have to acknowledge what those risks are attached to.
TSMC is not an ordinary semiconductor manufacturer. It is the world’s leading dedicated chip foundry, producing semiconductors for companies that design chips but do not want to own and operate the fabrication plants needed to manufacture them.
A leading-edge semiconductor fabrication plant is not a warehouse with unusually expensive machinery. It is one of the most complex manufacturing environments humans have created. The process requires extraordinary precision, specialized equipment, tightly controlled materials, enormous capital investment, and years of accumulated expertise.
A company cannot decide on Monday that it would like to compete with TSMC and begin shipping advanced chips by Friday afternoon.
Money helps, but money alone is not enough. Technical knowledge, process integration, yield management, customer trust, supplier coordination, and manufacturing culture all matter. TSMC has spent decades building these capabilities.
That accumulated advantage is why companies designing advanced processors repeatedly rely on it. Customers do not merely need a factory capable of producing a chip. They need a manufacturing partner capable of producing millions of complex chips with acceptable yields, predictable performance, and a reliable development schedule.
An advanced design that cannot be manufactured economically is a very expensive diagram.
TSMC converts diagrams into commercial reality.
The company’s recent results demonstrate how valuable that role has become. In the second quarter of 2026, TSMC reported revenue of approximately $40.2 billion, up 33.7% from the year-earlier period. Its gross margin reached 67.7%, while its operating margin reached 60.3%. Management guided for third-quarter revenue between $44.6 billion and $45.8 billion, with a projected gross margin between 65% and 67%, according to TSMC’s second-quarter results.
Those are remarkable numbers for a capital-intensive manufacturer.
A company operating massive factories is not supposed to produce software-like margins while spending tens of billions of dollars on new facilities. Yet TSMC’s technological position and manufacturing execution have allowed it to capture economics that would make many supposedly lighter businesses jealous.
Naturally, this is where my suspicion begins.
The market rarely allows me to own an exceptional business without attaching a complication large enough to require its own geopolitical briefing.
Geopolitics Is Not a Footnote
The largest risk surrounding TSMC is obvious: much of its most important manufacturing remains concentrated in Taiwan.
Taiwan’s political status and its relationship with China create a risk that cannot be treated as a routine paragraph buried near the end of an investment article. China claims Taiwan as part of its territory. Taiwan operates as a self-governing democracy. The United States maintains a deliberately complicated policy intended to discourage both forced unification and unilateral moves toward formal independence.
TSMC operates inside this tension while manufacturing chips essential to the global economy.
This is the part where investors often divide into two camps.
One group behaves as though military conflict is imminent and any investment in TSM is financial skydiving without a parachute.
The other group argues that TSMC is so strategically important that no rational actor would seriously disrupt it.
I find both positions too comfortable.
The first assumes catastrophe with more confidence than the available evidence supports. The second assumes rational restraint will always prevail, a theory history has tested repeatedly with mixed results.
TSMC’s importance may discourage conflict because the economic consequences would be enormous. The semiconductor ecosystem sometimes receives the flattering description of a “silicon shield,” suggesting that Taiwan’s technological importance provides strategic protection.
But a valuable asset is not automatically a protected asset.
Strategic importance can create deterrence. It can also create leverage, competition, and pressure. The same factories that make Taiwan economically essential also make global dependence on Taiwan politically uncomfortable.
That is why governments and customers are encouraging geographic diversification.
For me, the correct approach is neither panic nor denial. I treat geopolitical risk as a low-frequency but potentially extreme-impact event. I do not pretend I can forecast it precisely, and I do not construct a position so large that one event could permanently damage my portfolio.
No discounted cash flow model can make invasion, blockade, sanctions, or severe regional disruption feel appropriately civilized.
The Risk Is Broader Than a Full-Scale Invasion
When people discuss Taiwan risk, the conversation often jumps immediately to invasion.
That makes for dramatic headlines, but it can distract from less dramatic scenarios that might still hurt TSMC shareholders.
A blockade, expanded military exercises, airspace restrictions, shipping disruptions, cyberattacks, infrastructure sabotage, or prolonged political confrontation could interfere with operations and investor confidence without becoming a conventional war.
Financial markets would not wait for a formal declaration before reacting.
If shipping routes became uncertain, insurers increased rates, customers accelerated inventory purchases, or foreign employees began leaving the region, the effects could appear well before physical damage occurred.
TSMC itself identifies geopolitical tension, cyberattacks, supply-chain disruption, sabotage, utility failures, and natural disasters among the threats capable of interrupting operations. Its 2025 annual report explains that prolonged disruption affecting the company, its customers, or its suppliers could reduce manufacturing capacity and materially harm performance.
That disclosure is not evidence that disaster is expected. Annual reports contain long inventories of unpleasant possibilities because attorneys enjoy imagining every conceivable form of ruin.
Still, the risks are not fictional.
Semiconductor production depends on continuity. A fabrication plant cannot be casually shut down, moved across town, and reopened after lunch. Production cycles are long, processes are tightly calibrated, and interruptions can damage output already moving through the facility.
The danger is not limited to whether a building survives. The surrounding system must continue operating.
Employees need access. Electricity must remain stable. Water must remain available. Chemicals, gases, replacement parts, masks, wafers, and specialized equipment must continue arriving. Customers must be able to receive finished products.
A semiconductor fab is less like an isolated fortress and more like an extraordinarily expensive organism connected to a global circulatory system.
Overseas Expansion Reduces One Risk and Creates Others
TSMC is responding to geographic concentration by expanding outside Taiwan.
The company has projects in Arizona, Japan, and Germany. Its first Arizona fab entered high-volume production in late 2024, and TSMC has announced a much larger American expansion involving additional wafer fabs, advanced packaging facilities, and a research-and-development center.
The company has described total planned U.S. investment of $165 billion. It expects that, once the expansion is completed, roughly 30% of its capacity for 2-nanometer and more advanced technologies will be located in Arizona.
That is not a minor satellite operation. It is an attempt to build a substantial semiconductor manufacturing cluster in the United States.
Japan provides another piece of diversification. TSMC’s first Kumamoto facility began volume production in late 2024, initially emphasizing specialty technologies. The company later said the second Kumamoto fab was planned to use 3-nanometer technology to serve demand influenced by AI. In Germany, TSMC is developing a Dresden facility focused primarily on automotive and industrial applications.
As an investor, I like the strategic logic.
Customers and governments want production distributed across more regions. Local capacity may reduce certain transportation and political risks. Building near important customers can strengthen relationships and make TSMC even harder to replace.
But diversification is not free.
TSMC has said overseas fabs carry higher costs than its operations in Taiwan. Labor, construction, utilities, regulation, supply-chain maturity, and local manufacturing ecosystems all affect profitability. Taiwan has spent decades developing dense networks of suppliers, engineers, contractors, and technical institutions around semiconductor production. That ecosystem cannot be copied by issuing a press release and cutting a ribbon.
A fab may be located in Arizona, but it still needs suppliers, technicians, equipment support, chemicals, materials, and a workforce capable of maintaining demanding production schedules.
Replicating the building is easier than replicating the ecosystem.
TSMC expects support from government incentives, customer commitments, pricing, and operational scale to help manage the higher cost. I believe the company has a reasonable chance of succeeding because its execution history is exceptional.
Reasonable is not the same as painless.
Overseas expansion could pressure margins, consume capital, create delays, and make operations more complicated. Geographic resilience may improve while economic efficiency declines.
The investor receives greater security and a larger bill.
AI Demand Is Both the Opportunity and the Test
The artificial intelligence boom has transformed TSMC’s growth outlook.
AI accelerators accounted for a high-teens percentage of company revenue in 2025, and management entered 2026 expecting another year of strong growth. Advanced AI systems require leading-edge logic chips, high-bandwidth memory controllers, and sophisticated packaging capable of combining multiple computing and memory components.
TSMC benefits from nearly every part of that demand.
It manufactures advanced processors. It provides leading process technologies. It supplies advanced packaging services such as Chip-on-Wafer-on-Substrate, commonly known as CoWoS. As AI chips become larger and more complex, packaging becomes increasingly important.
The chip is no longer the whole story.
Multiple components must be integrated with extraordinary precision while managing heat, power, bandwidth, and physical scale. TSMC’s packaging roadmap includes larger CoWoS configurations designed to combine more computing dies and high-bandwidth-memory stacks. In April 2026, the company said a 14-reticle CoWoS platform capable of integrating roughly 10 large compute dies and 20 HBM stacks was planned for production in 2028, according to its North America Technology Symposium announcement.
That roadmap tells me AI demand is changing the physical architecture of computing.
It also tells me TSMC must keep spending aggressively before it knows exactly how demand will develop several years from now.
This is the central AI risk.
If demand remains powerful, the company may struggle to add enough capacity. If demand disappoints, it may discover that it built very expensive factories for customers whose enthusiasm peaked before their business models matured.
Either scenario creates pressure.
Capacity Shortages Are a Good Problem Until They Are Not
Investors often describe tight capacity as a good problem.
That is true in the same way that being too popular is a good problem. It remains pleasant until everyone is angry that I cannot return their calls.
TSMC has repeatedly discussed strong demand for advanced nodes and CoWoS packaging. Front-end fabrication capacity may be constrained, but advanced packaging can create another bottleneck after the wafer has been manufactured.
The factory can produce an advanced die, yet the customer still cannot ship the final accelerator unless the packaging ecosystem can integrate it with memory and other components.
When demand exceeds supply, TSMC gains pricing power and visibility. Customers may provide commitments that help justify expansion. High utilization supports margins.
But shortages also create risks.
Customers may delay products. They may redesign systems to reduce dependence on constrained technologies. They may fund alternative suppliers. Governments may increase support for domestic competitors. TSMC may spend rapidly to satisfy demand just as the market begins changing.
Capacity decisions are especially difficult because semiconductor factories require long lead times. Management must estimate future demand years before the resulting capacity becomes available.
The company is effectively placing enormous wagers on technology roadmaps that do not yet exist as commercial products.
So far, TSMC has managed this challenge better than almost anyone.
Still, no management team can eliminate the cycle. It can only manage the cycle more intelligently than competitors.
What If AI Spending Is Ahead of AI Economics?
I believe AI demand is real.
I also believe real demand can become overextended.
Those ideas are not contradictory. Railroads transformed transportation and still produced financial bubbles. The internet changed civilization and still punished investors who paid anything for companies with “.com” in their names.
Technological importance does not guarantee disciplined capital allocation.
Major cloud providers and technology companies are spending extraordinary sums on AI infrastructure. They are building data centers, acquiring accelerators, securing electricity, and signing long-term agreements.
TSMC sits near the foundation of this spending cycle.
If AI systems generate durable productivity improvements and profitable services, the demand could continue for years. Models may require more computing power. Inference may expand across consumer and enterprise applications. Custom silicon may proliferate as large customers optimize their own workloads.
In that scenario, TSMC may become even more important.
But I also have to ask what happens if customers collectively overbuild.
What if model improvements begin requiring less computing power? What if cheaper inference reduces the need for premium accelerators? What if enterprise adoption progresses more slowly than expected? What if customers discover that AI revenue does not justify the infrastructure spending required to generate it?
TSMC does not need AI to fail completely to experience pressure.
Growth merely has to fall short of the assumptions embedded in capacity plans and investor expectations.
This is a crucial distinction. A company can continue growing while its stock declines because the market expected growth to be faster, longer, or more profitable.
The greatest risk to a popular investment is not always disaster.
Sometimes it is ordinary success arriving after investors paid for perfection.
Customer Concentration Deserves Attention
TSMC serves a broad range of semiconductor designers, but a relatively small number of large customers account for significant portions of revenue.
That concentration reflects the economics of leading-edge chip design. Developing advanced processors is extremely expensive, so only a limited group of companies can produce enough volume to justify the newest manufacturing technologies.
Large customers provide scale and visibility. They also possess bargaining power.
If one major customer changes product timing, adjusts inventory, shifts designs, or loses market share, TSMC can feel the impact. If several customers are exposed to the same AI spending cycle, apparent diversification may be less comforting than it looks.
Different customer names do not always mean different economic risks.
A cloud company designing a custom accelerator, a processor supplier selling AI chips, and a memory-controller designer may all depend on the same wave of data-center investment.
TSMC’s pure-play foundry model helps reduce direct competition with customers. Unlike a vertically integrated semiconductor company, it does not design competing branded processors for the same markets. Customer trust is one of its greatest strengths.
But trust does not eliminate strategic self-interest.
Customers continually evaluate cost, capacity, technology, and supply security. Some may use alternative foundries for less advanced products. Others may support competitors to preserve negotiating leverage or reduce dependence on one supplier.
TSMC’s technology leadership makes switching difficult at the leading edge.
Difficult does not mean customers will stop trying.
The Supply Chain Extends Far Beyond TSMC
TSMC may be the central manufacturer, but it does not control every component required to produce advanced chips.
It relies on specialized lithography systems, deposition and etching equipment, metrology tools, industrial gases, chemicals, wafers, photomasks, packaging substrates, and countless other products.
Some of these markets are highly concentrated.
Extreme ultraviolet lithography provides the most famous example. The machines required for the most advanced semiconductor processes are produced by a very small supplier base, with critical subsystems coming from additional specialized companies.
If equipment deliveries are delayed, TSMC cannot simply order a comparable machine from a nearby office-supply store.
Advanced manufacturing depends on precision tools with long production cycles and globally distributed components. Export restrictions, transportation problems, supplier accidents, natural disasters, and sudden demand increases can all create bottlenecks.
The same applies to materials.
A disruption affecting one specialized chemical may have consequences far beyond the supplier’s apparent size. Semiconductor supply chains contain numerous products that are inexpensive relative to the finished chip but essential to producing it.
The value of an input does not measure the damage caused when it disappears.
TSMC works extensively with suppliers, maintains inventory strategies, qualifies alternatives, and develops continuity plans. Its scale gives it influence, and its strategic importance encourages suppliers to prioritize it.
But no company can fully diversify every dependency without creating enormous cost and complexity.
Efficiency and resilience are often natural enemies forced into an uncomfortable partnership.
Taiwan’s Natural Resources Add Another Layer
Geopolitical tension attracts the headlines, but Taiwan also faces natural and infrastructure risks.
Earthquakes are a recurring concern. Semiconductor equipment is extraordinarily sensitive, and even when facilities are designed to withstand significant seismic activity, production interruptions and equipment recalibration can affect output.
Taiwan also experiences typhoons and periodic drought conditions.
Chip manufacturing requires large quantities of ultrapure water. TSMC has invested in water recycling, conservation, and alternative supplies, but regional scarcity remains an operational concern.
Electricity is equally important.
Advanced fabs consume enormous amounts of power, and AI-related manufacturing growth will increase those needs. TSMC requires electricity that is not merely available but stable. Even brief disturbances can affect complex production processes.
These risks rarely receive the same emotional reaction as cross-strait conflict. There are no military maps or dramatic speeches.
Yet an investor should not ignore a mundane problem simply because it lacks theatrical lighting.
Water shortages, grid constraints, construction delays, or energy-price increases can pressure operations and costs without producing an international crisis.
The world does not need to end for a quarterly margin to disappoint.
Export Controls Create Strategic and Commercial Friction
Semiconductors have become central to national-security policy.
The United States and allied countries have imposed restrictions intended to limit China’s access to advanced chips and manufacturing technology. TSMC must navigate these rules while serving a global customer base and operating across multiple jurisdictions.
Compliance is not optional.
If regulations prevent TSMC from manufacturing certain products for specific customers, the company can lose revenue. If restrictions expand, the affected market may grow. If rules differ across countries, compliance becomes more complicated.
There is also a long-term competitive consequence.
Restrictions designed to limit access to advanced technology may encourage China to invest even more heavily in domestic semiconductor capabilities. Those efforts may not quickly replace TSMC at the leading edge, but they could increase competition in mature and specialty nodes.
TSMC therefore faces pressure from both directions.
It must follow restrictions that limit some business opportunities while watching restricted markets devote more resources to developing alternatives.
At the same time, TSMC’s strategic importance gives it influence and access to government support. The company benefits from incentives tied to overseas manufacturing, including support for its Arizona expansion.
Government involvement can reduce financial burdens.
It can also attach expectations.
Once semiconductor capacity becomes national policy, factory decisions are no longer determined exclusively by economics. Governments care about domestic employment, security, technology transfer, and supply guarantees.
TSMC may gain subsidies while losing some freedom to pursue only the cheapest manufacturing location.
There is no free government money. There is government money with excellent documentation and long-term interests.
Margin Pressure Is the Risk I Can Actually Measure
Geopolitical risk is dramatic but difficult to price. Margin pressure is less exciting and far easier to observe.
TSMC’s overseas fabs are expected to be more expensive than its Taiwanese operations. The company is also spending heavily on advanced nodes, packaging, research, and construction.
New technologies usually create initial margin dilution.
Early production may have lower yields. Equipment depreciation begins. Utilization may take time to rise. Engineers must solve manufacturing problems while customers refine designs.
As production scales and yields improve, economics can strengthen. TSMC has demonstrated this pattern repeatedly.
But the company is now expanding across more technologies and geographies simultaneously.
It is ramping 2-nanometer production, developing future processes, increasing CoWoS capacity, building in Arizona, expanding in Japan, and progressing in Germany.
Execution complexity is growing.
TSMC’s second-quarter 2026 gross margin of 67.7% was exceptional, and third-quarter guidance remained strong. I would not assume those margins represent a permanent floor. Foreign exchange, overseas costs, technology ramps, utility prices, pricing decisions, and capacity utilization can all move the figure.
The danger is not necessarily that margins collapse.
The danger is that investors begin treating unusually favorable conditions as normal and value the stock accordingly.
I prefer to assume that excellent businesses still experience difficult quarters.
That belief has saved me from several expensive emotional relationships with earnings estimates.
Valuation Can Turn Strength Into Risk
A risk analysis is incomplete if it ignores price.
TSMC may be an exceptional company, but an exceptional company can still become a poor investment when purchased at an unrealistic valuation.
The stock’s valuation reflects expectations about AI growth, technological leadership, margin durability, and continued execution. The more investors believe TSMC is indispensable, the more they may be willing to pay.
Indispensability is not the same as invulnerability.
If geopolitical tensions increase, the market may apply a larger discount even if earnings remain strong. If AI growth slows, valuation multiples may contract before revenue declines. If overseas expansion pressures margins, investors may reconsider how much they are willing to pay for each dollar of earnings.
The stock could therefore fall without the business becoming bad.
This is emotionally difficult because investors like moral clarity. We want a declining stock to mean management failed or the thesis collapsed.
Sometimes the business performs well and the price simply started from a level that required something closer to divine intervention.
When I consider TSM, I separate business quality from entry price.
I can admire the company and still wait.
I can own the stock and still recognize that future returns depend on what I paid.
I can believe AI will transform computing without assuming every AI-related security will outperform.
Enthusiasm is not valuation discipline.
How I Think About the Risk
I divide TSM’s risks into three categories.
The first category contains potentially catastrophic risks: military conflict, blockade, severe sanctions, or prolonged disruption in Taiwan. I cannot forecast these events reliably, so I manage them through position sizing and diversification.
The second category contains operational risks: overseas construction delays, supply shortages, packaging constraints, utility interruptions, natural disasters, export-control compliance, and lower manufacturing yields. These are serious but manageable, especially given TSMC’s experience and resources.
The third category contains investment risks: paying too much, overestimating AI demand, assuming peak margins will persist, or confusing technological importance with guaranteed shareholder returns.
This third category receives less attention because it lacks aircraft carriers.
It may be the risk most likely to hurt ordinary investors.
I cannot control geopolitics. I can control whether I chase a stock after a dramatic rally. I cannot control AI infrastructure budgets. I can control whether my portfolio assumes they will grow forever.
I cannot make Taiwan geographically inconvenient to China.
I can avoid putting half my retirement account into one company because a chart looked persuasive.
My Bottom Line on TSM
TSMC sits at the center of several powerful trends.
Advanced computing requires smaller, faster, more efficient chips. AI systems require enormous processing power and increasingly sophisticated packaging. Major technology companies want custom silicon. Governments want secure semiconductor capacity. Customers want access to leading manufacturing without building their own fabs.
TSMC benefits from all of this.
Its technology leadership, manufacturing skill, customer trust, and scale form one of the strongest competitive positions I can find in the public market.
But strength does not eliminate risk. It changes the shape of risk.
The company’s importance makes it valuable, yet its geographic concentration makes it strategically sensitive. AI demand creates growth, yet it requires huge capacity investments before future consumption is certain. Overseas expansion improves resilience, yet it raises costs and execution complexity. Supply-chain leadership provides bargaining power, yet the company still depends on specialized equipment and materials it cannot produce alone.
This is why I do not view TSM as either an obvious bargain or an uninvestable geopolitical trap.
I view it as a world-class business that requires world-class risk awareness.
For me, the investment case depends on three beliefs.
First, AI and high-performance-computing demand must remain structurally strong even if annual growth fluctuates.
Second, TSMC must preserve its manufacturing and packaging leadership while successfully expanding capacity across several regions.
Third, the valuation must compensate me for risks that may never appear in an earnings forecast.
If those conditions are satisfied, TSM can remain one of the most compelling long-term semiconductor investments available.
If I ignore any one of them, I am not analyzing the stock.
I am admiring the company.
There is a difference.
TSMC manufactures some of the most advanced technology on Earth, but owning TSM still requires an old-fashioned skill: accepting that great opportunities and serious dangers often occupy the same address.
I do not need to predict every outcome.
I need to respect the range of outcomes, size the position accordingly, and resist the urge to mistake confidence for certainty.
That may not produce the most exciting investment thesis.
It may produce one I can survive.
Comments
Post a Comment