If you only glance at Taiwan Semiconductor Manufacturing’s (TSM) chart and headline multiples, it’s easy to assume the stock must be stretched after vaulting past the trillion‑dollar mark. The 20x price-to-earnings ratio is pretty lofty for this type of capital intensive business.
Look a layer deeper, though, and you see a business that sits at the fulcrum of the AI build‑out, is expanding capacity at the exact bottlenecks constraining industry growth, and still has levers on price and mix that most investors don’t fully see.
So, are investors fully valuing its scarcity, pricing power, and operating geometry over the next 24–36 months?”
30 Billion Reasons to Buy TSMC
In Q2 2025, TSMC delivered $30.1 billion in revenue and a 58.6% gross margin, with advanced nodes (3/5/7nm) accounting for 74% of wafer revenue and high‑performance computing (HPC, which includes AI) representing 60% of total revenue.
The top line for next quarter was expected to land between $31.8 and $33.0 billion and lifted overall growth to around 30% for the year.
By late August 2025, TSMC’s market cap hovered around $1.2 trillion, with a trailing P/E in the high‑20s, a forward P/E near the mid‑20s, and an EV/EBITDA ~15x. That’s elevated versus lagging foundries, but modest versus AI leaders trading at far higher growth‑adjusted multiples despite heavier platform risk.
The market recognizes TSMC’s centrality to AI, but it’s not awarding software‑like multiples. The multiple is more “high‑quality industrial technology” than “hyper‑growth narrative stock.”
Unusual Currency Situation Is an Edge
Nearly all revenue is in USD while ~75% of COGS is NT‑dollar‑denominated. Management quantified the sensitivity by stating that every 1% NT appreciation knocks about 1% off reported NT$ revenue and ~40 bps off gross margin.
With the NT$ notably going up this year, gross margin headwinds create real economics. Management reiterated six profitability levers, including technology lead, pricing, utilization, cost reduction, mix, FX. Long‑term gross margin of 53%+ remains the target despite foreign exchange. In addition, overseas fabs dilute margins now, but TSMC is scaling its way down the cost curve.
Management was clear that overseas ramp will dilute gross margin by 2%–3% annually in early years, widening to 3%–4% later as Arizona, Kumamoto, and Dresden scale. That sounds scary until you pair it with explicit pricing power and higher‑margin mix (3nm/2nm + advanced packaging), as well as learning‑curve effects as volume rises. The company’s long‑term 53%+ margin target explicitly incorporates this dilution.
TSMC is fixing the bottleneck, which is AI advanced packaging in CoWoS/SoIC. The AI compute boom is constrained at the back‑end: substrate, interposers, and advanced packaging throughput. TSMC has been doubling CoWoS capacity, signaled additional price increases for advanced packaging, and is planning U.S. advanced packaging capacity by 2028. This is directed at the industry choke point and so pricing and mix are set to benefit.
Add to this the fact that the cluster strategy in the U.S. is bigger than investors think. Management described a multi‑fab cluster buildout in the U.S., with volume ramp at Fab 1 (N4), 3nm tooling for Fab 2, and additional fabs in planning. Beyond politics, clustering creates economies of scale, improves yields, and can structurally lower the “overseas penalty” over time.
While peers position High‑NA EUV as an imminent necessity, TSMC reiterated it doesn’t need High‑NA for A16 (1.6nm‑class) or A14 (1.4nm‑class). Delaying High‑NA adoption avoids a very costly depreciation hit while still advancing performance/power via nanosheets and backside power delivery. Fewer ultra‑expensive tools now means less capex drag and better returns.
Valuation through the right lens
In a base case, the top line grows at about 30% this year driven by 3nm ramps in Apple’s latest devices, data center parts and ongoing packaging throughput gains.
2026 growth decelerates to the mid‑teens as 2nm (N2) ramps late 2025/early 2026 and packaging expansion normalizes bottlenecks and gross margin hovers mid‑50s as FX headwinds and overseas dilution are countered by pricing and mix.
On those assumptions, a forward P/E in the mid‑20s and EV/EBITDA ~15x is consistent with a high‑moat, mid‑teens revenue compounding business with dominant share at the cutting edge. The PEG on multi‑year EPS growth (teens to high‑teens) remains reasonable.
If things go well, CoWoS/SoIC pricing and volumes will surprise to the upside thanks to Nvidia/Rubin and other AI accelerators consume larger share of 3nm and shift to 2nm quicker than modeled.
TSMC’s “earn our value” stance then shows up as blended ASP uplift across leading nodes and packaging in 2026 (we’ve already seen credible reports of packaging and 3nm price hikes).
Plus, high‑NA EUV deferral keeps depreciation lower than peers, supporting higher free cash flow conversion into the back half of the decade.
In this scenario, multiples don’t need to expand much; the stock works as EPS outgrows consensus.
But don’t overlook the downside that a 2026 AI capex pause pushes utilization down at exactly the moment overseas fabs ramp.
Even then, two offsets matter, the first being node leadership with N2 ramping and the second being pricing. Management has been explicit it will seek to recover value in price when FX and cost swing against it.
TSMC’s economic moat is as a monopolist‑adjacent at advanced nodes and by many third‑party estimates, it controls the overwhelming share of 7nm and below, and, at 3nm and 2nm, practically all high‑volume capacity for leading edge AI, CPU, and modem parts.
What If Hyperscalers Stop?
Hyperscalers on a tear in spending but a pause means packaging utilization would dip. That risk is real, but the pipeline of 3nm/2nm designs, increasing HBM attach, and the shift to larger GPU/accelerator packages insulates leading‑edge demand better than legacy cycles.
Export‑control/license regimes for advanced AI chips create friction in China and nearby markets. Taiwan risk never fully leaves the narrative, though TSMC and Taiwan’s government have been clear that cutting‑edge R&D and the newest nodes remain in Taiwan even as overseas capacity grows.
Apple and Nvidia are likely the two largest customers. That grants leverage and also fortifies TSMC’s “must have” status at the exact nodes those customers need.
Is The Multiple Fair?
At a glance, TSMC’s mid‑20s forward multiple can feel “full.” In context, it looks reasonable, and arguably conservative, for a company holding near‑monopoly share at the leading edge and solving the AI industry’s tightest bottleneck (advanced packaging) while raising price.
TSMC isn’t “overvalued” so much as it is fairly valued for a scarce asset. With AI capex growing roughly along the current path, the stock can compound into today’s multiple. And with packaging upside materializing alongside a clean N2 ramp, consensus EPS likely has room to move higher, and today’s multiple will look undemanding in hindsight.
The author has no position in any of the stocks mentioned. Financhill has a disclosure policy. This post may contain affiliate links or links from our sponsors.