TSMC 2nm revenue starts a 3–4 point margin test

TSMC has moved 2-nanometer manufacturing from a roadmap slide to a revenue line, and the economics are arriving in the less comfortable order. The process contributed just 3% of second-quarter wafer revenue, yet management expects its ramp to subtract about 3 to 4 percentage points from gross margin in the second half of 2026. That mismatch—not the record profit headline—is the useful investor problem.

The quarter itself was formidable. Revenue reached $40.201 billion, up 12% from the first quarter and 33.7% from a year earlier in U.S. dollars, while gross margin rose to 67.7%. The official earnings release also put third-quarter revenue guidance at $44.6 billion to $45.8 billion. TSMC is not struggling to find demand. It is trying to turn a steep technology ramp into shipments without giving back too much of the profitability that makes its foundry model exceptional.

The market is treating that execution question seriously. At 9:54 a.m. ET on July 17, TSMC’s NYSE-listed ADR was down 4.18% at $392.63 in Nasdaq’s real-time feed. That is an article-time snapshot, not a final close, and the price move does not prove a single cause. The results, the capital plan and a weak early tape belong in the same analysis; they should not be collapsed into a convenient story about investors “rejecting” AI demand.

The expensive part of 2nm arrived before its scale

TSMC’s second-quarter management report provides the cleanest starting point. The new 2nm process accounted for 3% of wafer revenue, its first disclosed contribution. The 3nm, 5nm and 7nm nodes supplied 30%, 33% and 11%, respectively. Taken together, 7nm and more advanced nodes made up 77% of wafer revenue.

Those percentages describe the mix of wafer revenue, not customer applications. They also expose the awkward timing of a leading-edge ramp. A node can be commercially real while still being too small to absorb its initial manufacturing burden. Early batches run through fresh tools, new process steps and a still-developing cost curve. TSMC did not itemize the relative effect of yield, depreciation and utilization for N2, so assigning a precise amount to any one mechanism would be guesswork. Management did give the combined result: a 3-to-4-point gross-margin drag from the steep N2 ramp in the second half.

The two figures use different denominators and must not be cross-multiplied. Three percent of wafer revenue is not “equal” to a 3-point hit on company gross margin. The useful comparison is directional: N2 is still a small contributor to sales, but it is already large enough in production to influence the economics of the whole company.

Inventory carries a second clue. Days of inventory increased from 80 to 87, which TSMC attributed primarily to the N2 ramp. That does not, by itself, signal weak end demand. It shows that cash and material are moving into the manufacturing pipeline ahead of a fuller revenue contribution. TECHi’s Q1 TSMC report treated N2 as a coming ramp; the Q2 documents turn it into a measurable operating fact.

A 1.7-point guide decline masks a larger fight

At the midpoint, TSMC expects third-quarter revenue of $45.2 billion, roughly 12.4% above Q2. The gross-margin midpoint, however, is 66%, down 1.7 percentage points sequentially. Management said the N2 ramp alone is expected to dilute gross margin by about 3 to 4 points, partially offset by very strong leading-edge demand, cost improvement, productivity gains and better allocation of capacity across nodes.

That bridge matters. If the N2 headwind is 3 to 4 points while the guided net decline is 1.7 points, the other forces collectively imply roughly 1.3 to 2.3 points of offset, all else equal. TSMC did not publish an itemized reconciliation, so that range is an inference rather than company guidance. It still shows the scale of the work being done by utilization and manufacturing discipline.

The earnings-call transcript separates another cost that is easy to mix up with N2. Overseas fabs diluted Q2 margin even as cost improvements and higher overall utilization lifted it. Management expects overseas expansion to carry a 2-to-3-point margin drag in its early stages, widening to 3 to 4 points later as the footprint grows. That is a multi-year structural pressure, distinct from the near-term N2 ramp.

For investors, this is the stronger reading of the quarter: demand is buying TSMC room to absorb two expensive transitions at once. It is not making either transition free. A foundry moat is valuable precisely because customers keep filling advanced capacity while the operator works down the new-node cost curve.

HPC at 66% is not the same as 66% AI

High-performance computing represented 66% of Q2 net revenue, up from 61% in Q1 and 60% a year earlier. Revenue from the platform grew 20% sequentially. Smartphone revenue, by contrast, declined 4% and accounted for 22%.

The tempting shorthand is to call two-thirds of TSMC’s revenue “AI.” That would be wrong. HPC is a platform bucket that includes far more than AI accelerators: server CPUs, custom silicon, networking devices and other high-performance chips also live inside it. TSMC does not publish a cross-tab that tells readers how much 2nm revenue came from HPC or how much HPC revenue came from AI.

Management’s own discussion makes the distinction useful rather than pedantic. Chief Executive C.C. Wei said agentic AI is increasing demand for CPUs alongside accelerators because data centers need more general-purpose compute around AI workloads. The company works across x86, Arm and RISC-V customer programs. AI is broadening the silicon stack, but the quarterly platform number cannot be treated as a pure AI sales disclosure.

This is where the 2nm ramp becomes more interesting than another “AI boom” headline. The product mix gives TSMC multiple ways to fill leading-edge wafers. The node economics determine whether that volume converts into the margins investors have come to expect.

The record profit needs an operating split

Net income attributable to shareholders reached NT$706.56 billion, up 77.4% year over year, and diluted earnings were NT$27.25 per share. Those are reported results, but they include a material non-operating gain.

TSMC recorded NT$63.20 billion in disposal and mark-to-market gains on shares of Vanguard International Semiconductor. The company said the gain contributed NT$2.24 to Q2 earnings per share, about 8.2% of the reported EPS. It is real accounting income, but it is not evidence that wafer economics improved by the same amount.

The operating business hardly needs embellishment. Gross margin was 67.7%, operating margin was 60.3%, and operating income increased 16.3% sequentially. Separating the Vanguard gain from the foundry result does not diminish the quarter; it prevents a one-time portfolio event from being mistaken for a repeatable manufacturing margin.

Cash flow adds useful texture. Operations generated NT$783.36 billion, while capital spending consumed NT$496 billion. Free cash flow was NT$287.36 billion, down NT$60.85 billion from Q1 because capital expenditure grew faster than operating cash flow. TSMC can fund the ramp, but the ramp is visibly consuming cash before all of its output reaches customers.

Inventory is the early cost of the roadmap

The balance sheet recorded a NT$74.08 billion sequential increase in inventories. Combined with the seven-day rise in inventory days, that is the clearest near-term measure of how N2 is moving from qualification into volume.

Inventory should not be read in isolation. A weak order book can create unwanted stock, while a healthy node ramp can build work in process ahead of shipments. TSMC explicitly tied the increase in days to N2, and its Q3 revenue midpoint calls for another 12% sequential increase. The more revealing test comes over the next two reports: whether inventory days stabilize as N2’s revenue share expands.

Capital allocation points in the same direction. TSMC raised its 2026 capital budget to $60 billion to $64 billion. About 70% to 80% is earmarked for advanced process technology, with 10% to 20% for advanced packaging, testing, mask making and related capacity. The company spent $26.8 billion in the first half.

That budget is a statement of customer visibility, but it is also a higher hurdle. Tool purchases arrive before depreciation is fully absorbed, and new capacity has to be qualified, loaded and kept productive. The relevant question is not whether TSMC can spend $64 billion. It is whether the spending produces enough saleable leading-edge output to keep gross margin in the mid-60s while overseas fabs scale.

The tape is testing execution, not disproving demand

TSM traded between $386.03 and $394.37 through 9:54 a.m. ET, below the prior close of $409.74. Alpha Vantage’s available quote was limited to that July 16 close, so the live snapshot in this article uses Nasdaq and names the provider rather than presenting an end-of-day value as current.

The chart offers context, not a verdict. TECHi’s TSM technicals page showed a 50-day moving average of $425.09 and a 200-day moving average of $349.70 using July 16 end-of-day data. The ADR was therefore trading below its medium-term trend line but still above its long-term one at the article-time print. That bracket helps explain why the post-earnings move feels consequential without pretending one morning has settled the investment case.

Readers tracking new disclosures can use TECHi’s TSM news feed, but the immediate price action still needs disciplined language. The official results establish rapid revenue growth, the first 2nm contribution and a higher capital plan. They do not establish that every seller acted because of margin guidance, capex, valuation or a broader semiconductor move. “No verified single catalyst” is more accurate than an invented explanation.

What would settle the margin question

Q3 gives investors a compact scorecard. Revenue near the $45.2 billion midpoint would confirm that leading-edge demand is converting into shipments. Gross margin within the 65% to 67% range would show that utilization, cost improvement and capacity optimization are offsetting much of the N2 burden. A rising 2nm share alongside stable or declining inventory days would indicate that the ramp is moving through the factory rather than accumulating inside it.

The invalidator is equally concrete. If gross margin falls below guidance while N2 revenue expands, inventory days keep climbing and overseas-fab dilution worsens, the current offset story is too optimistic. If margin holds near the midpoint as revenue rises, TSMC will have shown that its manufacturing system can absorb an unusually steep node ramp without surrendering the economics of the franchise.

TSMC has already proved that customers want the output. Q2 did not prove that the first wave of 2nm output is mature. The next stage of the AI foundry trade depends less on another demand superlative and more on a quieter factory question: how quickly 3% of wafer revenue can stop carrying a company-wide margin penalty.