TSMC Commits Another $100 Billion to the U.S. as AI Demand Reshapes Its Expansion Plan

The company is not adding capacity during a weak cycle or in anticipation of a distant recovery. It is expanding after record quarterly results, rising orders from major U.S. customers and growing pressure on the global supply of advanced chips and packaging.

The central question is no longer whether TSMC needs more capacity. It is whether demand will remain strong enough to support a U.S. manufacturing program whose scale and cost are far above the company’s original Arizona plan.

A $265 Billion U.S. Manufacturing Footprint

The additional investment will finance four semiconductor fabrication plants and two advanced packaging facilities in Arizona. TSMC’s planned U.S. network will expand to 12 facilities, including six new sites covered by the latest commitment.

The new fabs will focus on 2-nanometer and more advanced chips, placing Arizona inside TSMC’s leading-edge production roadmap rather than limiting it to older technologies. The packaging facilities are also important because high-performance AI accelerators increasingly depend on close integration between processors and high-bandwidth memory.

TSMC initially treated Arizona as a limited geographic diversification project. The additional investment changes that role. The state is becoming a second large-scale manufacturing base designed around demand from American customers.

President Donald Trump’s administration presented the announcement as part of a broader effort to strengthen domestic semiconductor production. For TSMC, however, customer orders appear to be the more immediate driver. CEO C.C. Wei said the additional investment would support strong multi-year demand from leading U.S. clients.

Capital Spending Jumps as AI Orders Accelerate

TSMC raised its expected 2026 capital expenditure to $60–64 billion, up from its previous forecast of $52–56 billion.

The Reuters chart shows capital spending of roughly:

• $36 billion in 2022
• $30 billion in 2023
• $30 billion in 2024
• $41 billion in 2025
• $60–64 billion planned for 2026

After two years of lower spending, the planned 2026 budget represents a step change rather than a gradual recovery. At the midpoint of the forecast, annual capex would rise by about 51% from 2025 and more than double the level recorded in 2024.

The increase covers more than the Arizona project. TSMC is expanding capacity in Taiwan, the U.S. and Japan as demand grows for AI accelerators, custom processors and advanced packaging. Still, the new U.S. commitment shows where a substantial share of the company’s longer-term spending will go.

Wei described AI-related demand as “extremely robust” and said the need for computing power continues to grow. He expects demand to remain strong through approximately 2029 or 2030, giving TSMC a multi-year horizon against which to plan its new fabs.

Record Earnings Support the Buildout

TSMC reported NT$1.27 trillion, or about $39 billion, in quarterly revenue, an increase of 36% year over year. Net profit reached a record NT$706.6 billion, or approximately $22 billion, rising 77% from the previous year and exceeding analysts’ expectations.

The company also raised its 2026 revenue growth forecast to slightly above 40%, compared with its earlier guidance of more than 30%.

The results explain why TSMC is accelerating construction instead of protecting cash. Profit is rising faster than revenue, while the company sees enough future demand to increase capital spending by billions of dollars within the same year.

This does not remove the financial risk. New fabs take years to complete and begin generating depreciation costs before reaching full utilization. The current earnings surge gives TSMC room to absorb those costs, but the economics of the expansion will still depend on how consistently customers use the new capacity.

Arizona Moves Closer to the Leading Edge

The first Arizona project was announced as a single $12 billion fab. TSMC later increased its commitment to $40 billion, then $65 billion. A further $100 billion expansion raised the total to $165 billion, while the latest announcement takes it to $265 billion.

The final step is the largest. TSMC is adding as much to the program in the latest announcement as it committed during the entire previous expansion.

The technology mix is changing as well. The additional fabs are expected to manufacture chips at 2 nanometers and below, bringing Arizona closer to the company’s most advanced operations in Taiwan. This matters for Nvidia, Apple, AMD and developers of custom AI processors, which rely on leading-edge nodes to improve performance and energy efficiency.

Arizona will not replace Taiwan as the center of TSMC’s operations. The company’s engineering base, supplier network and most advanced production infrastructure remain concentrated there. The U.S. project instead adds a large alternative manufacturing base near some of TSMC’s most important customers.

Packaging Becomes Part of the Capacity Problem

The inclusion of two advanced packaging facilities addresses a bottleneck that has become increasingly visible during the AI boom.

Producing the processor itself is only one part of manufacturing an AI accelerator. The chip must be integrated with high-bandwidth memory and other components through complex packaging technologies. Limited packaging capacity can restrict shipments even when sufficient wafer production is available.

TSMC’s expansion therefore targets both sides of the constraint: fabrication and packaging. Keeping more of that process in Arizona could reduce transport between production stages and give U.S. customers a more complete domestic supply chain.

It also increases the cost and complexity of the project. TSMC must develop a local network of engineers, equipment providers, material suppliers and specialized contractors capable of supporting several stages of advanced manufacturing.

Higher U.S. Costs Put Margins Under Scrutiny

A larger American footprint provides supply-chain resilience, but it is unlikely to match Taiwan’s cost structure in the near term. Construction, labor and operations are generally more expensive in the U.S., while Taiwan offers TSMC a dense network of suppliers and experienced semiconductor workers. New Arizona facilities will need time to improve production yields, train staff and reach efficient utilization levels.

That makes gross margin one of the most important metrics to watch as the expansion progresses. Record profit currently gives the company room to invest, but sustained overseas manufacturing costs could offset part of the benefit from strong AI pricing and volume.

Government incentives and tax credits can reduce the initial burden, though they cannot guarantee competitive production. The factories will ultimately need sufficient customer volume to justify both their construction and ongoing operating costs.

The decision to add four fabs suggests TSMC has received stronger demand signals than the public order data alone reveals. Large customers typically reserve capacity years before a facility enters production, particularly for advanced nodes with limited alternatives.

A Larger Bet on a Longer AI Cycle

TSMC’s new investment plan rests on two connected assumptions: demand for AI computing will continue rising through the end of the decade, and major U.S. chip designers will increasingly want part of their supply manufactured closer to home.

The company’s latest results support the first assumption. Revenue rose 36%, profit increased 77%, and management raised its annual growth forecast above 40%. The capex increase shows that TSMC is preparing for demand beyond the current quarter rather than treating the surge as temporary.

The second assumption is harder to measure. U.S. customers value geographic diversification, but Arizona production must still meet their expectations for price, yield and delivery. A strategically located fab has limited commercial value if its output remains materially more expensive than comparable production in Taiwan.

TSMC is using record AI earnings to build that capacity before the market needs it. By the time the new facilities open, the company will either have created a profitable second center for advanced chip production or committed too much capital to demand that failed to develop as expected.