TSMC to Expand Production in the US with $265 Billion Investment, AI Valuations Increasingly Focus on Cash Flow

TL;DR

• TSMC's Q2 earnings and full-year guidance reinforced AI chip demand, with its annual capital budget raised to between USD 60 billion and USD 64 billion.
• Market debate centers on US factory construction costs, the 2nm ramp-up, and long-term capital returns; strong demand does not equate to stress-free profit margins.
• Associated tickers: TSM, NVDA, AMD, AVGO, ASML.

After reporting its Q2 earnings on July 16, TSMC raised its full-year revenue and capital budget guidance and continued to advance its US expansion plans.

This should have been a standard positive for the AI supply chain. Revenue is growing rapidly, the share of advanced process nodes is increasing, 2nm is starting to contribute revenue, and demand from US customers was described by management as multi-year structural demand. However, the market's focus has shifted from "securing orders" to "can orders translate into sufficiently good returns."

The reason is straightforward. The stronger the AI demand, the heavier the capital investment TSMC must make. Building factories, purchasing equipment, and expanding advanced packaging all require upfront spending. US factory construction also entails higher costs, longer ramp-up times, and a more complex execution environment.

TSMC Chairman C.C. Wei offered a relatively optimistic view during the earnings call. He emphasized that the company's investment is driven by multi-year demand from leading US customers, and that TSMC will maintain reasonable gross margins necessary for long-term expansion, without sudden, significant price hikes due to supply tightness.

Strong Earnings Don't Eliminate Return Concerns

The Q2 data itself was very strong. TSMC officially reported revenue of USD 40.20 billion, a 33.7% increase year-over-year. In New Taiwan Dollars, revenue grew 36.0% year-over-year. Net profit was NTD 706.56 billion, up 77.4% year-over-year.

The revenue composition also points directly to AI. Advanced process technologies, namely 7nm and below, accounted for 77% of wafer revenue. 2nm contributed 3% in its first quarter. High-end GPUs, custom AI chips, and high-performance computing chips all depend on this type of capacity.

The message from the earnings report was clear. AI demand has already entered TSMC's revenue structure, not just remained an order narrative. The company also raised its 2026 US dollar revenue growth forecast to slightly above 40% and increased its full-year capital budget to between USD 60 billion and USD 64 billion.

But revenue growth isn't the only variable. The higher the capital budget, the more investors will ask: when will this money turn into capacity, when will that capacity generate revenue, and can that revenue sustain the historically high profit margins?

AI Demand Intensifies the Expansion Cycle

TSMC's current expansion isn't about replenishing capacity in a typical inventory cycle; it's more like repositioning for the AI supply landscape of 2026 to 2028.

AI chips require not only advanced process nodes but also advanced packaging. Advanced packaging can be understood as efficiently integrating multiple chips together, enabling faster coordination between computation, memory, and data transmission. For AI servers, packaging is equally crucial in determining final supply.

The company disclosed that 70% to 80% of the full-year capital budget will be allocated to advanced process technologies, approximately 10% to specialty technologies, and 10% to 20% to advanced packaging, testing, and mask-making.

The US expansion is progressing under the same logic. According to company announcements and media reports, after TSMC's additional USD 100 billion investment in Arizona, its cumulative committed investment in the US stands at approximately USD 265 billion. The new investments will be directed towards several advanced logic fabs and advanced packaging facilities.

C.C. Wei's emphasis on not implementing sudden, large price increases should also be viewed within this framework. This doesn't mean TSMC lacks pricing power, but rather that the company prioritizes long-term customer relationships and capacity commitments. For Nvidia, AMD, and Broadcom, securing stable capacity is more important than short-term pricing.

US Expansion Raises the Cost Benchmark

The scale of US investment is the most sensitive part of this revaluation. It reduces market concerns about regional capacity concentration while providing the US AI supply chain with a stronger domestic manufacturing narrative.

However, building factories in the US is not a free safety net. Labor, construction, supply chain integration, and management costs are typically higher than in Taiwan, and new fabs must navigate yield and utilization rate ramps during the start-up phase. Lower yields translate directly to higher unit costs.

Therefore, the US expansion carries a dual meaning. It helps mitigate geopolitical risks, making both clients and the US government more inclined to view TSMC as part of the AI infrastructure. At the same time, it could potentially compress overall profit margins, diluting the cost advantage derived from the "Taiwanese ecosystem coupled with ultra-high efficiency."

This could alter TSMC's valuation anchor. The market previously paid a premium for its technological scarcity and high profit margins. Now, it also needs to assess whether the capacity post-US expansion can maintain a sufficiently high return on invested capital.

Gross Margins and Cash Flow Judge the Expansion

What TSMC can currently demonstrate is strong AI demand, accelerating US expansion, and entry into a phase of higher capital intensity. However, these conditions alone do not guarantee superior long-term returns.

The near-term pressure is on gross margins. The company's earnings call commentary indicated that the steep ramp of 2nm in the second half of 2026 is expected to dilute gross margins by approximately 3 to 4 percentage points. The dilution from overseas fab ramps is initially estimated at 2 to 3 percentage points, potentially widening to 3 to 4 percentage points later on.

This shows that margin pressure is already incorporated into management's guidance. Even with rapid revenue growth, the costs associated with the 2nm ramp and overseas fab startups could put阶段性 pressure on profit margins. If the market values TSMC as a high-margin leading-edge foundry, a downward revision in gross margins could amplify stock price volatility.

The medium-term variable is free cash flow. CFO Wendell Huang emphasized strong operational cash flow, indicating that the expansion isn't driven by cash constraints. Nonetheless, a capital budget of USD 60-64 billion will consume substantial cash, and the scope for dividends and shareholder returns will continue to face scrutiny.

The long-term depends on the yield and utilization rates of the US fabs. If US advanced process and packaging capacity ramps up steadily and is consistently absorbed by AI customers, the high capital expenditure will translate into stronger barriers to entry. If AI capital spending slows after 2027, or if US fab costs remain higher than expected for an extended period, TSMC's growth story could be repriced based on a discount on returns.

The most important thing to track for TSMC right now isn't whether quarterly revenue continues to hit new highs, but the quality of profits generated from these high investments. The AI super-cycle has provided the rationale for expansion; the market will use gross margins, free cash flow, and the progress of US fab ramps to reassess whether the company can still deliver the high-quality returns of the past.