While we’re keeping an eye on Nvidia's valuation, and exploring some fears on the AI bubble, TSMC is spending $42 billion in 2025.
That number should make you pause. This isn't routine capital allocation.
The company is launching nine new facilities this year—eight fab plants plus one advanced packaging facility. For context, TSMC historically opens five facilities annually. They're running 80% above their normal pace.
Here's what the market isn't pricing in: This expansion centers on CoWoS packaging capacity, not just chip production. TSMC plans to scale CoWoS output from 36,000 wafers monthly to 130,000 by the end of 2026. That's 3.6x growth in 18 months.
The Bottleneck Nobody Talks About
Nvidia $NVDA doesn't have a chip problem. It has a packaging problem.
The same constraint hits AMD $AMD and Broadcom $AVGO.
All three depend on CoWoS, Chip-on-Wafer-on-Substrate technology, to build their high-performance AI chips.
No other supplier can replicate TSMC's CoWoS process at scale. IC design firms wait for TSMC allocation because they have no alternative.
This creates a structural advantage. TSMC isn't just selling manufacturing capacity. It controls the critical path for AI infrastructure deployment.
Consider the timeline. Current CoWoS capacity of 36,000 wafers monthly can't support projected AI chip demand through 2026.
Hyperscalers—Amazon $AMZN, Microsoft $MSFT, Google $GOOGL—are building data centers that require chips TSMC can't produce fast enough today. The gap between demand and supply isn't closing with incremental adjustments. TSMC's 3.6x capacity expansion is the minimum response to visible order books.
The 2nm Timeline
Fab 25 in Taiwan goes online for sub-2nm production by 2028. That facility alone represents multi-billion dollar investment in next-generation process technology.
Current leading-edge production runs on 3nm and 4nm nodes.
The 2nm transition matters because it enables another performance leap for AI accelerators. Training models at frontier scale requires computing power that doesn't exist yet. The infrastructure being built now, data centers, power systems, and cooling, assumes chips that won't ship until TSMC delivers 2nm at volume.
That gap matters more than most investors recognize.
Companies ordering 2nm capacity today are making capital allocation decisions based on TSMC's execution timeline.
If TSMC delays, their entire AI infrastructure roadmap shifts.
If TSMC delivers early, competitors without secured capacity fall behind.
The Taiwan Pressure Intensifies
China deployed warships and fighter jets around Taiwan in what observers called the largest military exercises to date.
This matters for TSMC investors because it's not theoretical anymore. The exercises showed China's capability to isolate Taiwan if tensions escalate. Semiconductor supply chains that depend on Taiwan face concrete disruption risk, not just abstract geopolitical scenarios.
The market's 15-25% discount on TSMC shares reflects this reality.
That discount either represents opportunity or appropriate risk pricing, depending on your assessment of near-term conflict probability.
The ETF Exposure Question
Three semiconductor ETFs offer different TSMC exposure profiles: $SOXX, $SMH, and $SOXL.
SOXX holds broad semiconductor exposure with moderate TSMC weighting.
SMH concentrates holdings in equipment manufacturers and foundries.
SOXL provides 3x leveraged exposure to semiconductor indices, amplifying both upside and volatility.
But here's the complication: Taiwan concentration risk.
TSMC's expansion happens primarily in Taiwan, despite plans for Arizona and Japan facilities. Geopolitical tension around Taiwan creates tail risk that ETF investors can't diversify away if they want pure-play foundry exposure. The discount applied to TSM shares relative to fundamental value reflects this political premium.
Some investors route around this by owning TSMC's equipment suppliers instead. ASML and Applied Materials capture TSMC's capex surge without direct Taiwan exposure.
When TSMC commits $42 billion to facility expansion, a significant portion flows to lithography equipment from ASML and deposition tools from Applied Materials.
The math works differently depending on your view of Taiwan risk.
If you believe cross-strait tensions remain manageable through 2028, direct TSMC exposure offers better value.
If you price in a meaningful probability of supply chain disruption, the equipment suppliers provide cleaner exposure to TSMC's growth without the geography risk.
TSMC currently trades at a 15-25% discount due to China/Taiwan tension. How are you positioning?
What The Capex Cycle Tells Us
TSMC doesn't spend $42 billion on speculation. This capital deployment reflects contracted revenue visibility extending years forward.
Semiconductor capex cycles typically run 18-24 months ahead of revenue recognition. Equipment orders placed today support production that ships in 2026 and 2027. TSMC's willingness to expand this aggressively indicates customer commitments that justify the investment.
Compare this to Intel's recent capital discipline. Intel $INTC cut capex guidance after demand visibility deteriorated.
TSMC is doing the opposite—accelerating investment when peers are pulling back. The CoWoS expansion is particularly revealing. Packaging capacity doesn't generate headlines like 2nm process technology.
But it's the constraint that determines whether Nvidia can ship H200 systems on schedule. TSMC increasing CoWoS capacity 3.6x suggests they're seeing order commitments that make current capacity inadequate for 2026 demand.
The Applied Materials Angle
Applied Materials ships the deposition and etching equipment TSMC needs for new fabs. When TSMC announces eight new fabrication facilities, Applied Materials books orders for hundreds of process tools per fab.
The lag matters for timing. TSMC's 2025 capex translates to Applied Materials revenue in 2025-2026, with service revenue extending beyond initial installation. Investors buying Applied Materials today are positioning ahead of that revenue flow.
The same logic applies to ASML, which holds a monopoly position in extreme ultraviolet lithography. Every leading-edge fab requires multiple EUV systems. TSMC's expansion requires dozens of additional machines. ASML's order book extends through 2026 largely because of TSMC's committed capex.
This creates a different risk profile than direct TSMC ownership. Equipment suppliers face cyclical risk, if TSMC's customers cancel orders, future capex cuts hurt Applied Materials and ASML.
But they avoid Taiwan-specific political risk that affects TSMC's valuation multiple.
The Geopolitical Discount
TSMC trades at a valuation discount to historical norms relative to earnings growth. The market applies a geopolitical risk premium that ranges from 15-25% depending on cross-strait tensions.
This discount creates opportunity if you believe the risk is overpriced. TSMC's technology leadership and customer dependency create structural moats that should command premium valuations. Instead, political uncertainty compresses multiples below where fundamentals suggest fair value sits.
The Arizona facilities don't solve this near-term. TSMC's US expansion won't reach meaningful volume until 2025-2026, and even then represents a small fraction of total capacity. Taiwan remains the center of TSMC's production through the end of this decade.
Investors need to decide if the discount compensates for the risk. There's no analytical framework that provides a clean answer. You're pricing political outcomes that depend on decisions by governments, not corporate management.
If China escalates drills around Taiwan next month, what is your first move?
What This Means
Three approaches make sense depending on risk tolerance.
Direct TSMC ownership provides purest exposure to the AI infrastructure buildout. You accept Taiwan risk in exchange for the company controlling the critical manufacturing bottleneck for next-generation chips.
Equipment supplier exposure through ASML and Applied Materials captures TSMC's capex without direct political risk. You give up some upside but avoid the tail risk of supply chain disruption.
ETF exposure through SMH or SOXX diversifies across the semiconductor value chain. You get TSMC upside with reduced concentration risk, but you also own companies with different growth profiles that may dilute returns.
TSMC's $42 billion capex commitment is sizing infrastructure for AI demand that hasn't fully materialized yet. They're building ahead of the curve because their customers need capacity guarantees to make their own multi-billion dollar infrastructure investments.
That forward positioning creates asymmetry. If AI infrastructure deployment accelerates, TSMC's capacity expansion looks conservative in retrospect. If it slows, they're stuck with underutilized facilities and compressed returns on capital.
The signal from TSMC's capital allocation is clear: they're seeing demand visibility that justifies unprecedented expansion. Whether that conviction translates to shareholder returns depends on execution and external factors beyond management control.
But the setup is worth watching.
When the company that controls the manufacturing bottleneck for AI chips commits $42 billion to expansion, that's not a marginal bet.
That's infrastructure for a market that doesn't exist yet at the scale they're building for.
What's catching investor attention today: $1.28 Trillion Poured Into ETFs in 2025. The Biggest Wealth Shift in a Decade
Disclaimer: This is not financial or investment advice. Do your own research and consult a qualified financial advisor before investing.