Foundry and Advanced Packaging Industry Explained: The U.S. View
Most of America's leading chip companies design chips but do not make them. Manufacturing happens at foundries, and at the leading edge that mostly means TSMC in Taiwan. The dependence runs both ways: customers based in North America accounted for 79% of TSMC's revenue in the third quarter of 2026, and high-performance computing, mainly AI and data center chips, made up 68%. TSMC's revenue that quarter was $46.72 billion, up 41% from a year earlier, at a 66.4% gross margin, and it guided $52.3–54.3 billion for the fourth quarter. Its newest 2-nanometer process already provided 8% of wafer revenue, and nodes of 7nm and below 79%. For AI chips, making the processor is only half the job: advanced packaging places it side by side with high-bandwidth memory, and packaging capacity has been one of the tightest steps in the AI supply chain. SEMI expects spending on foundry and logic equipment to rise 18.9% in 2026 to $78 billion.
- TSMC
- Q3 2026 revenue $46.72B, up 41%; North America 79% of revenue
- AI share
- High-performance computing was 68% of TSMC's Q3 2026 revenue
- Leading edge
- 2nm was 8% and 7nm-and-below 79% of TSMC wafer revenue in Q3 2026
How the chain works
TSMC revenue mix, Q3 2026
| Share | What it means | |
|---|---|---|
| By customer region: North America | 79% (China 5%) | U.S. chip companies are TSMC's core customers |
| By platform: high-performance computing | 68% (smartphone 21%) | AI and data center chips now dominate |
| By node: 7nm and below | 79% of wafer revenue | Leading-edge nodes make most of the money |
| By node: 2nm | 8% of wafer revenue | The newest node is ramping |
Designers and manufacturers
Nvidia, Apple, AMD, Broadcom and Qualcomm design chips and outsource manufacturing. That model lets them spend on design while the foundry spends the tens of billions needed for fabs. TSMC's capital spending was $13.84 billion in the third quarter of 2026 alone and $40.64 billion in the first nine months of the year.
TSMC's numbers show how central U.S. customers are: in the third quarter of 2026, customers based in North America provided 79% of its net revenue, up from 76% a year earlier, while China fell to 5% from 8%. High-performance computing made up 68% of revenue and grew 22% from the previous quarter.
Leading-edge nodes
Each new manufacturing node packs more transistors into the same area and uses less power, which matters for AI chips that run at their limits. In the third quarter of 2026, TSMC's 2nm process provided 8% of wafer revenue, 3nm 33%, 5nm 28% and 7nm 10%; together, 7nm and below made up 79%.
Ramping a new node is costly: TSMC said its gross margin of 66.4% was 1.3 points lower than the prior quarter, mainly because of dilution from the 2nm ramp. It guided fourth-quarter revenue of $52.3–54.3 billion and a gross margin of 65–67%.
Packaging: the AI bottleneck
An AI accelerator is not a single chip. The processor is placed next to stacks of high-bandwidth memory on a shared base, so data can move between them quickly. This advanced packaging needs its own equipment and lines, and capacity has repeatedly limited how many AI chips can ship.
Equipment forecasts reflect the shift. SEMI expects foundry and logic wafer fab equipment spending of $78.0 billion in 2026, up 18.9%, test equipment up 31.0% to $15.3 billion and assembly and packaging equipment up 9.6% to $6.7 billion, with advanced packaging for AI and HBM among the drivers.
How the U.S. view differs from the China view
The Chinese version of this page is about building a domestic foundry and packaging industry under export controls: Chinese foundries cannot buy the most advanced tools, so they push older nodes harder and invest heavily in packaging to make up the gap.
From the U.S. side the issue is concentration risk. America leads in chip design but relies on Taiwan for most leading-edge manufacturing and packaging, which is why the U.S. has offered subsidies to bring fabs and packaging lines onto U.S. soil.
2026: AI takes over the foundry
By the third quarter of 2026, high-performance computing had grown to 68% of TSMC's revenue, and North American customers to 79%.
TSMC's revenue rose 41% year on year to $46.72 billion, and it guided higher again for the fourth quarter as 2nm ramped.
The more AI chips U.S. companies sell, the more they depend on one foundry and its packaging lines. Capacity at those steps, not design, sets how many chips reach the market.
- U.S. AI chip growth runs through foundry and packaging capacity.
- New nodes bring higher revenue but dilute margins during the ramp.
- Concentration in a few suppliers is the chain's main risk.
Common misreadings
MythU.S. chip leaders make their own chips.
RealityMost outsource; North America was 79% of TSMC's Q3 2026 revenue.
MythSmartphones are still the main foundry market.
RealityHigh-performance computing was 68% of TSMC revenue, smartphones 21%.
MythMaking the chip is the whole job.
RealityAI accelerators also need advanced packaging with high-bandwidth memory.
MythNew nodes immediately raise margins.
RealityTSMC's margin fell 1.3 points in Q3 2026, mainly from the 2nm ramp.
Demand, capacity and geopolitics: scenarios
| Scenario | Main effect | How it passes through |
|---|---|---|
| AI chip demand keeps rising | Leading-edge and packaging capacity | Longer lead times, more capex |
| Packaging capacity catches up | AI chip shipments | More chips ship per quarter |
| Cross-strait tension rises | Supply risk | Customers push for capacity elsewhere |
| More U.S. fabs open | Supply location | Less reliance on Taiwan, higher costs |
| Tariffs on imported chips | U.S. buyers | Higher costs or faster onshoring |
| AI budgets are cut | Foundry utilization | Slower growth at leading nodes |
Supply and demand by segment
| Segment | Supply rigidity | Concentration | New demand | Policy role | What to watch |
|---|---|---|---|---|---|
| Chip design | Low | High (a few leaders) | Yes | High (export controls) | Product launches |
| Leading-edge foundry | Very high (years, tens of billions) | Extreme (TSMC) | Yes (AI) | High (subsidies, geopolitics) | Node mix, capex |
| Advanced packaging | High | Very high | Yes (AI) | High | Packaging capacity |
| Mature-node foundry | Medium | Medium | Low | High (China competition) | Utilization |
Supply rigidity: how long new capacity takes and whether money can speed it up. Concentration: whether output sits in a few countries or companies. The more rigid and concentrated, the more prone to shortages.
How to track it
| Indicator | What it tells you | Where to find it |
|---|---|---|
| TSMC monthly revenue | Foundry demand | TSMC investor relations, monthly, free |
| TSMC quarterly management report | Mix by node, platform and region | TSMC investor relations, free |
| Equipment spending forecasts | Capacity additions | SEMI forecasts, free |
| Semiconductor equipment | Who builds the fabs | U.S. view: chip equipment |
Sources
- TSMC: Third quarter 2026 results (2026-10-15)
- TSMC: 3Q26 management report (2026-10-15)
- I-Connect007 citing SEMI: equipment sales forecast to reach record $229 billion in 2028 (2026-07-15)
Last updated 2026-10-18.