TSMC's 2nm Revenue Explosion Reshapes the Semiconductor Power Map

TSMC Just Surged Past Its Own Ceiling

TSMC reported its largest quarterly revenue in company history — 1.49 trillion New Taiwan dollars, or about 62 trillion won, with net profit hitting 763 billion New Taiwan dollars. The operating margin sat at a punishingly high 51 percent. But the headline number nobody outside the foundry business will care about is wrong. The real story lives in a different digit.

In the third quarter, TSMC’s 2-nanometer process revenue reached 8 percent of total revenue. In the second quarter — the first quarter the company even tracked it — that figure was 3 percent. One quarter. A 167 percent increase in six months.

This is not a gentle ramp. This is a cliff’s edge acceleration. The company went from barely registering 2nm revenue to commanding nearly one-twelfth of all its sales inside a single quarter. Something shifted.

The AI Chip Supply Chain Thesis Gets Confirmed

TSMC’s 2nm node is designed for one thing and one thing only: the most powerful AI semiconductors on Earth. Nvidia and Apple are the anchor customers. These are not companies that order capacity on speculation. They commit years in advance, and when they commit, they commit with enough volume to sustain an entire fabrication line for a generation.

The third quarter data confirms what anyone tracking the AI hardware boom has suspected: the demand curve is steeper than anyone estimated. HPC (high-performance computing) now represents 68 percent of TSMC’s revenue. Smartphones sit at 21 percent. IoT and automotive each command just 4 percent.

Let that distribution sink in. Nearly seven out of every ten dollars TSMC earns comes from chips that power artificial intelligence infrastructure. The company has effectively become the manufacturing spine of the global AI economy.

What 8 Percent Actually Means

Eight percent of 1.49 trillion New Taiwan dollars is approximately 119 billion New Taiwan dollars — roughly 4.9 billion US dollars in a single quarter. That is not pocket change. That is the annual revenue of several mid-tier semiconductor companies concentrated into one quarter, from one product node.

To put this in perspective: Samsung’s entire foundry revenue in the same quarter was estimated at around 8 trillion won, or roughly 6 billion dollars. TSMC’s 2nm alone generated nearly as much in one quarter as Samsung’s entire foundry business.

The math is brutal and it is unfolding in real time.

The Advanced Process Concentration Is Accelerating

Break down the process revenue further and the picture sharpens. 3nm accounted for 33 percent of total revenue. 5nm hit 28 percent. 7nm represented 10 percent. Advanced nodes below 7nm collectively captured 79 percent of all TSMC sales.

The company is effectively cannibalizing its own previous-generation products faster than any foundry in history. Every quarter, the proportion of revenue coming from the most advanced nodes increases, and the margin structure follows. 2nm wafers command a price premium of roughly 3 to 6 percent over 3nm, according to industry sources cited by DigiTimes.

This is a structural shift, not a cyclical one. The economics of advanced semiconductor manufacturing reward concentration at the top. TSMC understands this better than anyone and is executing on it aggressively.

Samsung’s Foundry Strategy Is Now Under Siege

Samsung Electronics entered the third quarter reporting foundry revenues of approximately 8 trillion won, with market share in the sub-5nm segment estimated at around 15 percent — down from roughly 20 percent two years ago. The gap between TSMC and Samsung in the most advanced nodes has widened from a factor of two to a factor of three or more.

This is not a temporary setback. Samsung’s 3nm process, launched in 2022, had known yield issues that delayed major customer onboarding. Apple pulled its order volume away from Samsung and directed it to TSMC. Nvidia made the same calculation. The customers who matter most chose the incumbent over the challenger.

Samsung’s internal response has been to pursue a two-track strategy: continuing to invest in custom HBM (high bandwidth memory) production while simultaneously pushing its own GAA (gate-all-around) transistor architecture. The timeline for meaningful GAA revenue is estimated at 2027 or later. By then, TSMC will have shipped two complete generations of 2nm-class products.

Intel’s Foundry Bet Faces a Different Kind of Risk

Intel’s foundry ambitions look dramatically different from Samsung’s problem. Intel’s issue is not execution — it is timing. The company has spent roughly 100 billion dollars on fabrication capacity expansion since 2021, targeting the US and European markets. The geopolitical thesis is sound: decoupling advanced chip manufacturing from Taiwan reduces supply chain risk for Western customers.

But the technical thesis is still unproven. Intel’s 18A process, the company’s most advanced node, has not yet shipped volume production for major AI customers. The first expected customer commitments are projected for late 2026 or early 2027. TSMC’s 2nm is already generating 8 percent of revenue.

The gap between promise and delivery is the single most important metric in the foundry business. Right now, TSMC is ahead by roughly two years and that gap is widening, not closing.

Who Wins, Who Loses, What Happens Next

The winners are clear: Nvidia, Apple, and the AI infrastructure builders who secure TSMC capacity at current pricing. The losers are equally clear: any company that assumed Samsung or Intel would close the process gap within a meaningful timeframe.

TSMC’s pricing power will increase. The company is already signaling 3 to 6 percent wafer price increases for next year, according to DigiTimes. Customers who need 2nm capacity have limited alternatives. They will pay.

Samsung’s foundry strategy needs a fundamental reassessment. The current two-track approach — HBM investment plus GAA development — addresses different problems. HBM is a memory play. GAA is a process play. Neither directly competes with TSMC’s 2nm volume production. The company needs a third track: something that captures value in the AI chip supply chain without requiring process leadership.

Intel’s foundry timeline remains the critical variable. If 18A shipments begin on schedule in late 2026 with major AI customer commitments, the narrative shifts. If delays continue, the geopolitical premium may not be enough to overcome a two-year technology gap.

The 8 percent number is a milestone. It is also a threshold. Once a product node reaches single-digit revenue share at a company this size, it ceases to be experimental and becomes infrastructure. TSMC’s 2nm is now infrastructure for the AI economy. Everything else is a question of who can catch up.

The answer, right now, is nobody.