TSMC’s Growth Rate Is Not Normal for Its Size

Four months in a row. That is how long Taiwan Semiconductor’s monthly revenue has kept accelerating, and the latest print showed growth of 53.3% from a year earlier. I have watched a lot of chip companies report a good quarter and then cool off. I have not watched many post four straight months of speeding up while already generating tens of billions in quarterly sales. TSMC trades around $434.67 as I write this, close to its 52-week high of $478, and the question worth asking is not whether the business is good. It obviously is. The question is whether the stock still gives you room to be right.

My thesis is simple: TSMC’s growth rate is unusual for a company its size, the moat behind it is real, and the price has moved fast enough that the easy money in the re-rating is probably gone. I would still hold it, just smaller than the headline numbers alone would suggest.

Four straight months of acceleration

Start with the quarter that already happened. Revenue in the three months ended June 29 came to NT$1.27 trillion, up 36% from a year earlier and up 12% from the prior quarter. That is the fifth straight quarter of accelerating year-over-year growth, and it followed a full 2025 in which revenue rose to NT$3.81 trillion, up from NT$2.89 trillion in 2024. Forward estimates imply earnings per share growth of 47% from the trailing $13.40 to roughly $19.63. Companies this large usually grow in the high single digits. A rare great year gets them into the teens. TSMC has been compounding closer to 35% for over a year, off a base that was already the largest foundry on earth, and the monthly cadence into August did not show any sign of that slowing down.

I want to be precise about what is driving it, because “AI demand” gets used as an explanation for almost everything these days and that makes it easy to wave off. TSMC does not sell chips under its own name. It manufactures them for Nvidia, AMD, Apple, Broadcom and a long list of others, which means its growth is really a read on how much AI computing capacity the entire industry is trying to buy at once. When that number moves 36% in a quarter, it is not one company’s forecast. It is an industry-wide bet, made with real purchase orders.

The capacity nobody else can build

Here is why there is nowhere else for that demand to go. TSMC and Samsung are the only two foundries selling 5-nanometer and 3-nanometer chips to outside customers at meaningful volume, and Samsung has struggled with yield at 3nm, according to Morningstar’s equity research on the sector. Intel’s 14A node might become a real competitor around 2028 with US government backing, but even in that scenario the realistic outcome is Intel replacing Samsung as the second option, not unseating TSMC from the top. Everyone else in the industry stopped advancing nodes years ago and now serves the mature, low-margin end of the market.

That is the moat in a sentence: two viable suppliers at the leading edge, and one of them is stretched. Nvidia’s own reliance on outside manufacturing is part of why TSMC’s order book looks the way it does, and the same logic extends to AMD, which depends on the identical fabs for its competing chips. Neither company can simply choose a cheaper supplier. There is not one.

Margins are expanding even as the buildout grows

Gross margin came in at 59.9% in fiscal 2025, up from 56.1% the year before, and operating margin reached $1,936.1 billion on $3,809.1 billion of revenue for a full-year operating margin near 51%. That is worth sitting with. A company spending tens of billions on new fabs is usually diluting its own margins in the process, not expanding them.

It is not doing that by accident. TSMC’s customers reserve capacity years ahead and pay for it whether or not the chips ship on schedule, which cushions the fab-building costs against the usual semiconductor boom-bust pattern. TSMC’s chief financial officer told analysts in July that 2026 capital spending would rise to between $60 billion and $64 billion, up from the prior plan of $52 billion to $56 billion, with 70% to 80% of it going toward advanced nodes, as CNBC reported. The company also lifted its total committed US investment to $265 billion, spread across at least four more Arizona fabs. Overseas fabs, historically, run at lower margins than TSMC’s home base in Taiwan for years after they open, because yields take time to mature and local labor and construction costs are higher. If Arizona output scales up faster than expected, that gap could start showing up in consolidated margins before the revenue benefit fully lands. I would treat any margin compression tied to US expansion as noise rather than a thesis-breaker, but it is the kind of number I would want to see broken out separately once those fabs are running at volume.

A customer list with one very large name

Concentration is the risk that gets discussed least and matters most. Morningstar estimates TSMC’s single largest customer accounted for 19% of 2025 revenue, with the top four customers together making up roughly half of sales, and that Nvidia overtook Apple as the largest customer sometime in 2025. A foundry this dependent on one buyer’s capital spending plans is exposed to that buyer’s cycle, not just the industry’s. If Nvidia’s own order pace slows for reasons that have nothing to do with chip demand generally, say a change in how hyperscalers time their AI capex, TSMC would feel it before the broader “AI theme” showed any cracks elsewhere.

That is my one explicit uncertainty here: I do not know how quickly the AI buildout among Nvidia’s own largest customers, the hyperscalers, will slow once current data center construction catches up to today’s backlog. Nobody outside those companies really does. TSMC’s revenue is a lagging read on decisions made 18 to 24 months earlier, because of how far in advance capacity gets reserved, so a slowdown in AI capex commitments today would not show up in TSMC’s monthly numbers for a while.

Two multiples that do not fully agree

TSMC trades at 32.4 times trailing earnings, well above its five-year average of 25.0 times, but only 22.1 times forward earnings. Read those two together and the market is not saying the stock is expensive. It is saying the trailing multiple looks rich only because trailing earnings have not caught up to where estimates expect them to go. The price-to-sales ratio tells a similar story: 16.0 times current sales against a five-year average of 10.4, which is a much bigger gap than the earnings multiples show. Sales growth is real and measurable. The multiple expansion on top of it is a bet on durability, and bets on durability are exactly the part of a thesis that breaks first when growth merely slows rather than reverses.

One data point argues for caution on timing. TSMC’s July 16 earnings report beat expectations again, and the stock still fell -2.3% that day, against an average earnings-day move of 2.9%. When a beat produces a decline bigger than the stock’s typical earnings reaction, it usually means the bar going in was already set at “great,” not “good.” Applied Materials, which sells the equipment TSMC and its peers use to build these fabs, has told a version of the same capacity story from the supply side, and I would watch that read-through alongside TSMC’s own numbers rather than in isolation.

Metric	TSMC now	Five-year average
P/E (trailing)	32.4	25.0
P/E (forward)	22.1	n/a
P/S	16.0	10.4
P/B	11.1	7.2
Quarterly revenue growth	36%	n/a
Gross margin	59.9%	56.1% (prior year)

Where I would size the position

Short interest sits at just 0.6% of float, which tells you almost nobody is positioned against this story right now. That is not bullish or bearish on its own, but it does mean there is little skeptical money to absorb a disappointment if one shows up. The dividend yield of 0.80% is not why anyone owns this stock; it is a capital-intensive growth compounder, not an income position, and treating it like one would be a mistake.

Analyst targets average $552, ranging from $440 to $650 across 7 estimates, with 86% rating it a buy. Price targets alone rarely move my thinking. The spread here is informative anyway: even the low estimate sits close to today’s price, which suggests the sell-side has mostly caught up to the growth story rather than being ahead of it. How I size any position once the underlying stock’s swings get large applies directly here, because TSMC’s earnings-day moves of 2% to 4% are modest by chip-stock standards, but the stock’s five-fold-plus run over the past two years means a normal-sized dollar position today carries more concentration risk than the same position would have carried a year ago.

What would change my mind

The specific thing that would make me wrong is a break in the monthly revenue streak, not a single soft quarter. TSMC has posted five straight quarters of accelerating growth. The first month where year-over-year growth actually decelerates, rather than just growing off a bigger base, is the number I would treat as the real signal, more than any single earnings call. Until that print shows up, on the monthly disclosures TSMC releases in the first few days of each month, I would rather own this at a smaller position size than the growth rate alone justifies than avoid it because the multiple has already moved.

Analysis and opinion only, not investment advice. Figures come from Taiwan Semiconductor’s filings on SEC EDGAR and its investor relations site; valuation multiples are approximate and were checked on September 18, 2026.