2026-10-15

TSMC Posts Record Profit for a Sixth Straight Quarter, Beats on Revenue - So Why Did the Stock Fall?

Written & reviewed by the TradeSmrt editorial team · 2026-10-15

In this article

What Happened
Why the Stock Fell Despite the Beat - the 2nm Margin and Capex Math
What to Take Away From This
FAQ
Sources

What Happened

On October 15, Taiwan Semiconductor Manufacturing Co. - the world's largest contract chipmaker and the single most important supplier behind the AI buildout - reported third-quarter 2026 results that looked, on the surface, like a clean beat. Revenue came in at $46.7 billion (NT$1.49 trillion), up roughly 41% year-over-year, topping the high end of TSMC's own guidance range of $44.6-45.8 billion and beating Wall Street's consensus by about $1.3 billion, or 2.9%. Net income hit NT$763.1 billion, up 69% year-over-year, marking the company's sixth consecutive quarter of record profit. Earnings per ADR came in at $4.60, ahead of the roughly $4.42-4.45 analysts had modeled.

The catch was quality, not quantity. Gross margin landed at 66.4%, down 1.3 percentage points from the prior quarter's 67.7%. Management attributed the slide mainly to the ramp of its 2-nanometer process - the same cost drag it had already flagged last quarter, when it warned that 2nm start-up costs would dilute second-half gross margin by 3 to 4 percentage points. Operating margin, at 58.7%, actually came in above the top end of guidance, a bright spot the market largely overlooked. By segment, high-performance computing - chips feeding AI accelerators and data centers - made up roughly 68% of total revenue, underscoring just how dependent TSMC's growth has become on the AI buildout.

Guidance was aggressive too: TSMC guided fourth-quarter revenue to $52.3-54.3 billion, about 14% above the third quarter and, if hit, the largest single quarter in company history. And yet TSM shares kept sliding through and after the release. The stock had already fallen about 3.4% on October 14 alongside a broader chip selloff triggered by China's CXMT claiming a DRAM technology breakthrough, which hammered Samsung and SK Hynix shares in Seoul. Rather than reversing that move, the earnings print only added to it, as investors zeroed in on the margin trajectory and a capital-spending plan that keeps growing, instead of the headline beat. It's a pattern that's repeated through this earnings season: companies beat, raise guidance, and the stock sells off anyway - a "sell the news" reflex that has become almost a house style for AI-chip megacaps this year.

Why the Stock Fell Despite the Beat - the 2nm Margin and Capex Math

To make sense of the reaction, it helps to start with TSMC's unusual position in the market. It is the sole leading-edge foundry that essentially the entire AI chip ecosystem - Nvidia, AMD, Apple, Broadcom - depends on. That means TSMC's results aren't just a single-stock story; they function as a leading indicator for how real AI infrastructure demand actually is. The fact that HPC revenue climbed to 68% of the mix, with record guidance for Q4, is a genuinely bullish demand signal - at least on the order-book side of the ledger.

What worries the market is the cost side. The 2-nanometer node is TSMC's newest leading-edge process, and it's the node that next-generation Nvidia GPUs and Apple's flagship mobile chips are transitioning onto. Every node transition carries lower initial yields and underutilized capacity before production stabilizes, and those costs flow directly into gross margin. The "3-4 point dilution" warning management gave last quarter is now showing up in the actual numbers, and investors are nervous precisely because this isn't a one-time charge - it's a structural cost that will likely persist for several more quarters as the ramp continues. Expanding overseas fabs in Arizona and Kumamoto adds to the same drag, since production costs there run higher than in Taiwan.

The second concern is capital spending. TSMC has raised its 2026 capex guidance twice this year already - from an initial $52-56 billion range at the start of the year to $60-64 billion by the time of its second-quarter report. That's roughly 50% higher than last year's spending, a bet that doubles as both conviction in the AI supercycle and leverage: if AI demand softens even modestly, the earnings hit from that spending base would be outsized. TSMC's own management has fielded "is this a bubble" questions on nearly every earnings call this year. CEO C.C. Wei has been candid about it, saying on a recent call that he is "also very nervous" about the possibility, while adding that after cross-checking customers' demand forecasts against real data-center construction, power availability, and deployment schedules, he came away "quite satisfied" that the demand is real. In other words, even the company building the chips isn't dismissing the bubble question outright - it's just concluded, after its own diligence, that the demand looks structural rather than speculative.

The third variable is advanced packaging capacity, specifically CoWoS, the technology used to stitch together the high-bandwidth memory and compute dies inside AI accelerators. By some analyst estimates, Nvidia alone has locked up more than half of TSMC's total CoWoS capacity, and the broader packaging pipeline is reportedly booked out through 2027. That's bullish in one sense - demand clearly outstrips supply - but it also means TSMC's capacity expansion is struggling to keep pace with how fast AI chip orders are growing. Investors are watching nervously for the moment that balance tips: overbuild capacity ahead of a demand air pocket, and margins get crushed from the other direction; underbuild, and Nvidia and AMD's own shipment volumes become the bottleneck.

Put together, the message from this earnings report is that demand has been confirmed, but the cost of serving that demand is not trivial. The backward-looking numbers - revenue, profit - were about as clean as they come. The forward-looking ones - margin trajectory, capex intensity - are where the market's skepticism lives, and that's exactly why a record quarter still produced a falling stock.

What to Take Away From This

Separate the headline numbers from the margin guidance. A revenue and profit beat doesn't guarantee a stock pops if margins are sliding within a range management already flagged. When reading any earnings report, check the sequential change in gross and operating margin alongside whatever "known" cost headwinds the company previously disclosed - that's often where the real signal is.
Upstream suppliers are leading indicators for downstream customers. A company that sits at the top of a supply chain, the way TSMC does for the entire AI chip industry, reveals how solid its customers' demand really is well before those customers report their own numbers. TSMC's segment mix (HPC at 68%) and forward guidance are useful inputs when sizing up Nvidia, AMD, or any other name further down the chain.
Rising capex is a double-edged signal. Two guidance increases in a single year show strong conviction in AI demand, but they also raise the stakes if that demand cools even slightly. Timing the peak of a capex cycle is one of the hardest calls in semiconductor investing, and it's worth tracking even when you don't own the stock directly.
A "sell the news" reaction can be about positioning, not the earnings themselves. When a stock is already under pressure heading into a report - as TSM was here, following the Korea memory shock - good results may not be enough to attract fresh buying, and existing selling pressure can simply continue. Context matters as much as the print.
Capacity constraints like CoWoS cut both ways. A sold-out packaging pipeline looks bullish for demand today, but it's also a reminder that customers have an incentive to diversify toward alternative suppliers over time - a dynamic worth watching rather than assuming away.

FAQ

If TSMC's results were strong, why did Nvidia and AMD shares move on this too?

TSMC is the sole leading-edge foundry actually manufacturing Nvidia's AI GPUs and AMD's data-center chips. Its guidance and capacity - especially in advanced packaging like CoWoS - effectively set the ceiling on how much Nvidia and AMD can physically ship. That makes TSMC's margin and capex commentary a read on the health of the entire AI supply chain, which is why it ripples into customer stocks even when TSMC itself is the one reporting.

If the 2nm node is squeezing margins, why is TSMC pushing ahead with it so aggressively?

In foundry manufacturing, leading-edge process leadership is the main lever for long-term pricing power and customer retention. Yield rates are typically low and costly early in any new node's life, then improve as production matures and margins recover. TSMC is willing to absorb short-term dilution to make sure customers like Nvidia and Apple have no reason to shift volume to a rival foundry such as Samsung.

Does this earnings report signal a turning point for chip stocks broadly?

It's too early to say. The report confirmed that AI chip demand itself remains solid, but it also deepened concerns about the cost side - margin dilution and capex intensity. Coming on the heels of this week's CXMT DRAM breakthrough claim and the Korea memory stock selloff, broader semiconductor sentiment was already fragile, so one company's report isn't enough to call a trend reversal. Watch how Nvidia, AMD, and Micron frame demand and margins in their own upcoming reports for more confirmation.

Sources

This article is an original synthesis and analysis based on the reporting below, not a reproduction of the original articles. Please check the sources directly for the latest exact figures.

⚠️ This article is for informational purposes only and is not investment advice. Market conditions change constantly, so please verify the latest data before making any investment decisions.