TSMC Arizona was the canonical test case for whether the United States could host leading-edge semiconductor manufacturing. The doubts were many: cost differential to Taiwan, supply-chain depth, workforce availability, and the cultural overlay of operating a Taiwanese fab in the American Southwest. The first fab is now in volume production at the leading-edge node it was scoped to produce, at yield levels described internally as acceptable, with deliveries to customers including Apple and AMD. The first real data point for the US onshoring thesis is positive.
Key takeaways
- TSMC Arizona Fab 21 is in volume production at the leading-edge node.
- Yields are reportedly within acceptable range of Taiwan equivalents.
- Customer deliveries — Apple, AMD — are in volume.
- Cost differential remains real but smaller than feared.
What worked and what did not
Workforce challenges were real but largely addressed by importing experienced Taiwanese engineers during ramp; the supply-chain depth problem persists but has been partially closed by Asia Pacific supplier expansion into Phoenix; and the cost differential is real but offset by CHIPS Act subsidies and customer-supported pricing.
- Workforce. Imported Taiwanese expertise plus US engineer onboarding worked.
- Supply chain. Asia Pacific suppliers establishing local Phoenix footprint.
- Cost. Higher than Taiwan but subsidized to be acceptable.
What this means for policy
The proof-of-concept is meaningful. US-based leading-edge production at acceptable yield was not assumed possible by skeptics. The follow-on fabs and the Intel comparison case become more credible. CHIPS Act subsequent funding rounds have stronger empirical foundation.
Why the cost gap matters
Without sustained subsidy, the cost differential makes incremental fabs harder to justify.
What Intel learns
The TSMC Arizona success raises pressure on Intel Foundry to demonstrate equivalent capability.
Output profile
Fab 21 is operating at scale.
| Year | Wafer starts/month | Yield (Y/N) |
|---|---|---|
| 2024 (ramp) | ~10k | Mixed |
| 2025 | ~18k | Acceptable |
| 2026 | ~22k | Target |
The first real data point for US leading-edge onshoring is positive. The next question is replication.
Frequently asked questions
Is the cost gap closing?
Slowly. Subsidy still bridges it.
Does this help Intel Foundry?
It raises the bar — both helpful and challenging for Intel.
What about Fab 21B and C?
Both progressing. The follow-on case is now stronger.
The bottom line
TSMC Arizona has changed the empirical case for US semiconductor onshoring. The skeptics' prior was largely correct on cost but wrong on feasibility. The policy implications are large.
