AI foundry king scaling fast despite margin strain
- TSMC is the leading contract chip maker for advanced processors used by AI, cloud, smartphone, and PC customers.
- Advanced technologies at 7 nm and below made up 74% of 2025 wafer revenue and 77% in Q2 2026.
- Management expects full year 2026 revenue growth to exceed 40%, driven by agentic AI and structural demand.
- N2 entered volume production in 2025, and A16 risk production is expected in 2026.
- The steep ramp of N2 technology is expected to dilute gross margins by 3 to 4 percentage points in the second half of 2026.
- The biggest risks are U.S. tariffs, export controls, Taiwan concentration, foreign exchange swings, and the cost of the Arizona buildout.
Agentic AI demand meets costly expansion
TSMC sits at the center of the AI chip boom. Most leading AI accelerators and many high-end CPUs depend on its most advanced manufacturing lines. The rise of agentic AI is creating a resurgence in CPU demand alongside accelerators. Management now expects 2026 revenue to grow above 40% and AI accelerator revenue to grow at a 54-56% annual rate over five years.
The operating data supports the bull case. In 2025, advanced technologies at 7 nm and below made up 74% of wafer revenue. In Q2 2026, that rose to 77%. N2 entered volume production in 2025, A16 risk production is expected in 2026, and A14 volume production is planned for 2028. This keeps TSMC in the lead where customers need the most power-efficient chips.
The company is also spending at a massive scale. TSMC raised its 2026 capital expenditure budget to between $60 billion and $64 billion to meet strong customer commitments. About 70-80% of that budget is for advanced process technologies, and 10-20% is for advanced packaging, testing, mask making, and other areas. This signals high demand visibility, but it raises the bar for future utilization.
The bear case centers on structural and near-term margin pressure. TSMC is adding expensive overseas capacity, including a $100 billion Arizona investment. Management expects overseas fabs to dilute gross margin by 2-3 percentage points in early stages and 3-4 points in later stages. Additionally, the steep ramp of N2 will dilute margins by 3-4 percentage points in the second half of 2026. A new U.S. 25% ad valorem tariff on certain imported advanced computing chips adds more friction. The key question is whether pricing power stays strong enough to offset these rising costs.
A foundry paid for precision and scale
TSMC is a pure-play foundry. It does not mainly sell its own branded chips. It manufactures chips designed by other companies. Customers pay TSMC to turn chip designs into wafers using advanced tools, clean rooms, process recipes, and years of manufacturing expertise.
The model works best when TSMC leads in technology and fills its fabs. Each new node costs a lot to develop and build, but high volume spreads those costs across more wafers. In Q2 2026, 2 nm, 3 nm, 5 nm, and 7 nm were 3%, 30%, 33%, and 11% of wafer revenue. This heavy tilt toward advanced nodes is why TSMC earns high margins compared with many peers.
AI and high-performance computing now drive the business. High-performance computing reached 66% of Q2 2026 revenue. Agentic AI, where software agents make more requests and use more tokens, increases demand for CPUs, networking chips, and accelerators. TSMC supplies many key customers across those designs.
The global footprint is both a selling point and a cost risk. U.S. fabs can carry a price premium and may help customers reduce policy risk. However, overseas fabs cost more than Taiwan fabs, and TSMC states they will dilute gross margins for years. Currency fluctuations can also affect profitability significantly.
Advanced nodes feed AI demand
N3, the 3 nm family
N3 contributed 24% of wafer revenue in 2025 and 30% in Q2 2026. It serves high-end smartphones, AI-related processors, and other leading-edge chips.
N5 and N4, the 5 nm family
5 nm contributed 36% of wafer revenue in 2025 and 33% in Q2 2026. These nodes remain high-volume workhorses for CPUs, GPUs, accelerators, and flagship device chips.
N7 and older advanced nodes
7 nm was 14% of 2025 wafer revenue and 11% in Q2 2026. It supports older high-performance chips and helps TSMC balance capacity across multiple nodes.
N2, the 2 nm platform
N2 entered volume production in 2025 and reached 3% of wafer revenue in Q2 2026. It is central to the next wave of AI processors, but its steep ramp will dilute second-half 2026 margins.
A16 and A14 future nodes
A16 risk production is expected in 2026. A14 development is progressing, with volume production scheduled for 2028.
Advanced packaging, including CoWoS
Advanced packaging connects powerful chips with high-bandwidth memory for AI systems. TSMC plans to spend about 10-20% of its massive 2026 capital budget on advanced packaging and testing.
High-performance computing commands the mix
Platform mix uses TSMC's Q2 2026 revenue by platform disclosure. It is a quarterly mix showing heavy concentration in high-performance computing and North American demand.
What could crack the thesis
N2 and overseas fabs dilute margins
High impact · High oddsTSMC is building expensive fabs in Arizona while heavily ramping its new N2 technology. Management expects overseas fabs to dilute gross margin by 2-3 percentage points early on, and the N2 ramp will add a steep 3-4 percentage point headwind in the second half of 2026. If customer pricing power weakens, these costs will compress returns.
Tariffs and export controls change orders
High impact · Medium oddsThe U.S. imposed a 25% ad valorem tariff on certain imported advanced computing chips. Export rules also expanded license needs for global shipments of 16 nm and below processes. Customers may shift assembly, change shipment routes, or slow orders in price-sensitive markets.
Taiwan concentration remains central
High impact · Medium oddsMost of TSMC's leading-edge scale is still located in Taiwan. A military, political, power, water, or logistics shock could halt global chip supply and TSMC's revenue. Overseas fabs reduce some risk, but they will not match Taiwan's scale in the near future.
AI demand falls short
Medium impact · Medium oddsThe current thesis depends on surging AI accelerator growth and rising silicon demand from agentic AI. If cloud companies reduce capital spending, or if accelerator supply outpaces demand, TSMC could face lower utilization just as new capacity comes online.
In one breath
How does TSMC make money?
TSMC manufactures chips designed by other companies. Its best money comes from advanced nodes like 3 nm, 5 nm, and 2 nm, where customers pay for speed, power efficiency, and dependable production.
Why is TSMC important to AI?
Most leading AI accelerators and many data center CPUs depend on TSMC's advanced manufacturing. The rise of agentic AI is driving even more demand for the CPUs and accelerators TSMC builds.
What should investors watch most closely?
Watch gross margins, N2 ramp costs, AI accelerator demand, and the cost of overseas fabs. Also monitor U.S. tariff and export control changes, because they dictate where customers can ship products.
