First Eagle Trims TSMC: The Signal Is Supply, Not Demand

When a value-oriented fund with $59.92 billion in portfolio value reduces a position, the market reflexively reads it as a thesis break. First Eagle's reported reduction in its TSMC stake - and the larger ~2 million share cut disclosed in its Q1 2026 13F - triggered the standard headline cycle: Is AI demand peaking? Is TSMC overvalued? Is the foundry cycle turning?

The question is wrong. The actual driver is structural, and it has nothing to do with whether hyperscalers will keep buying AI chips. It has everything to do with what TSMC is building and how much it costs.

First Eagle does not trim positions because consensus sentiment shifts. It trims them when the capex-to-revenue trajectory implies that future returns on invested capital will drift below what the entry price assumed. That appears to be the calculus here. And the data supports it.

The Capex Problem

TSMC's trailing twelve-month capital expenditure stands at $45.87 billion. The company's initial 2026 guidance was between $52 billion and $56 billion. As of mid-July 2026, TSMC raised that to $64 billion and added $100 billion to its cumulative U.S. investment footprint. That is a capex trajectory that has nearly doubled from $28.9 billion in 2024.

Revenue is growing, too. TTM revenue growth is 23.8% year-over-year. Free cash flow came in at $34.72 billion for the trailing twelve months - solid, but leaving a $46 billion capex bill that consumes the majority of operating cash flow ($80.59 billion TTM). The free cash flow conversion rate - the portion of operating cash flow that remains after capital spending - has fallen to roughly 43%. That means for every dollar of operating cash TSMC generates, 57 cents goes back into building capacity.

The implication is fairly straightforward. A company spending $64 billion a year on capacity expansion cannot sustain that intensity unless revenue growth stays at or above 25% for the next three to four years. If it doesn't, return on invested capital - currently at 29.7%, which is exceptional - will face structural headwinds. Not because demand collapses, but because the asset base grows faster than earnings.

This is not a demand story. It's a supply discipline question inverted: TSMC isn't constraining supply to protect pricing. It's flooding the industry with new capacity because it believes the AI cycle justifies it. The risk is that the cycle doesn't last long enough to amortize the investment.

Utilization, Margins, And The Dilution Clock

TSMC management highlighted near-term margin dilution and packaging constraints on its Q4 2025 earnings call. That guidance was given before the capex boost to $64 billion. More spending means more capacity coming online, which means utilization growth slows.

The company's gross margin sits at 64.38% and operating margin at 56.29%. Both are best-in-class. But margin dilution on a $140+ billion revenue base is not cosmetic - it represents billions in reduced operating income. At a 31.5x trailing P/E, the market has already priced in sustained top-quartile profitability. Even modest dilution compresses the earnings multiple unless revenue growth fully offsets it.

There is a second-market split that matters here, and it's one the headline readers miss. TSMC's advanced-node foundry (3nm, 2nm, N2P) is structurally different from its specialty and mature-node business. Advanced-node capacity, particularly CoWoS advanced packaging, remains the constraint. TSMC is allocating 70-80% of its 2026 capex to advanced process technologies and 10-20% to packaging, testing, and mask making. The company has also pulled forward its Arizona Fab 2 timeline, targeting high-volume manufacturing in the second half of 2027 - up from a more conservative schedule.

The constraint is migrating from wafer fabrication to packaging. TSMC knows this. The capex split reflects it. But the packaging expansion is capital-intensive, yield-sensitive, and takes years to reach stable economics. That means the next 18-24 months of capacity growth will be lumpy, and margin dilution will be front-loaded before new capacity reaches efficient utilization.

First Eagle's Pattern, Not The Headline

First Eagle's position history with TSMC is instructive. The fund added shares in November 2025, increasing its stake by 6,983 shares - buying into strength during the AI-driven rally. Its Q1 2026 filing disclosed a reduction of 2,036,189 shares. What matters is the pattern, not the exact share count. First Eagle built the position during the early AI capex wave, then began reducing as TSMC's market cap approached $2.15 trillion and its capex guidance kept rising. This is consistent with a value investor who bought a company at one capex-to-revenue ratio and is now uncomfortable with the next one. The entry assumption no longer matches the exit reality.

TSM remains among First Eagle's five largest holdings as of the latest disclosed quarter, alongside Alphabet, Meta, Becton Dickinson, and Fomento Economico Mexicano. The trim is a reduction, not an exit. That distinction matters: it signals discomfort with the current valuation-to-capex ratio, not a loss of conviction in the AI cycle.

Investor Takeaway

The structural thesis is this: TSMC is not facing a demand problem. It is facing an investment intensity problem. A $64 billion annual capex budget is sustainable only if AI-driven foundry demand stays at current growth rates for the next three to four years. If it does, utilization absorbs the new capacity, margins stabilize post-dilution, and the stock earns its 31.5x multiple. If AI capex growth slows to high-single digits - still strong by historical standards, but insufficient for this buildout pace - then ROIC drifts lower and the multiple compresses.

What to watch: TSMC's quarterly utilization rate and node-level ASP trends. If utilization holds above 90% on advanced nodes while ASPs stay firm, the capex is justified. If utilization flattens or ASPs soften while new Arizona and Japan capacity ramps, First Eagle's trim will look prescient rather than premature.

The foundry cycle is bifurcating again. Advanced nodes with packaging capacity will remain constrained through 2027. Mature and specialty nodes face the opposite risk - excess capacity from slower cycle turns. TSMC's positioning across both markets is the real story. First Eagle appears to be pricing in the risk that the advanced-node premium narrows as capacity catches up to demand. Whether that happens depends entirely on whether the AI capex cycle outlasts the buildout.