TSMC is capturing more revenue from each unit of wafer output. The harder question is whether its next generation of capacity can earn the same returns.
TSMC's third-quarter results reveal something more important than another strong AI-driven revenue number.
Revenue rose 50.9% year over year to NT$1.49 trillion. Shipments, measured in 12-inch-equivalent wafers, increased just 9.6%.
In US dollars, revenue grew 41.2%.
That gap matters. TSMC is generating substantially more revenue relative to its wafer shipments, reflecting a richer technology mix, pricing, advanced packaging and other factors.
This is not a direct measure of wafer ASP. Consolidated revenue includes businesses beyond wafer fabrication, and equivalent wafer shipments do not capture differences in process complexity or packaging content.
But the direction is unmistakable: revenue is growing much faster than physical wafer output.
The premium is reaching operating profit
Third-quarter operating income reached NT$876.45 billion, up from NT$500.69 billion a year earlier.
Operating margin expanded from 50.6% to 58.7%.
The incremental operating margin on the additional NT$504.32 billion of revenue was approximately 74.5%.
That is an unusually strong result for a capital-intensive manufacturer.
It suggests TSMC is capturing substantial operating leverage as customers shift toward more advanced technologies.
The cash flow is equally notable.
TSMC generated approximately NT$920 billion in operating cash flow during the quarter, against NT$443 billion in capital expenditure. The difference was roughly NT$477 billion.
The business is not merely reporting higher accounting profits. It is generating considerable cash while funding one of the industry's largest manufacturing expansion programs.
Yet these figures describe the returns of the existing business and its current mix of assets. They do not establish the returns on capacity that has yet to ramp.
That distinction is becoming increasingly important.
N2 changes the investment equation
Two-nanometer technology already accounted for 8% of wafer revenue in the third quarter, while three-nanometer contributed 33%.
The transition is moving quickly. But its near-term economics are not yet as favorable as the headline revenue growth might suggest.
Management attributed the quarter's 1.3-percentage-point sequential decline in gross margin primarily to dilution from the initial N2 ramp.
That is not evidence of a failed technology transition. New manufacturing capacity carries depreciation, startup costs and yield-learning expenses before it reaches mature utilization.
It does, however, expose a fundamental difference between technological leadership and capital efficiency.
A more advanced wafer can command a higher price and still generate a lower return on invested capital if the capacity required to produce it is sufficiently expensive.
N2's long-term contribution will depend on how quickly utilization, yields and unit costs improve—not simply how much customers are willing to pay for the technology.
The older node may matter just as much
N3 is not disappearing as N2 expands.
TSMC continues to invest in N3 capacity and optimize equipment across technology generations. Customers do not all move to the newest node simultaneously. Their decisions depend on performance, power, chip area, product cycles and cost.
This creates a potentially valuable advantage: TSMC can serve several generations of high-performance products at once.
It also creates a capital allocation test.
If N3 remains well utilized while N2 ramps, TSMC can earn returns on a broader installed asset base.
If customer migration accelerates faster than replacement demand, older advanced-node capacity may become less productive before its economic value has been fully recovered.
The outcome cannot be inferred from technology revenue shares alone.
Overseas expansion adds another cost layer
The next investment cycle is not limited to Taiwan.
TSMC expects its overseas fabs to dilute consolidated gross margin by approximately two percentage points in 2026, with dilution potentially widening to three to four points over the following years as more US capacity ramps.
These facilities may offer strategic benefits to customers, including supply-chain diversification and geographic resilience.
But those benefits are not automatically equivalent to higher manufacturing returns for TSMC shareholders.
The company must either recover the additional costs through pricing, improve its overseas cost structure, or accept lower returns on some of that capital.
This is where the AI manufacturing premium faces its most important test.
The next dollar matters more than the last
TSMC has already demonstrated extraordinary earnings power.
The unresolved question is whether the next dollar invested in N2, advanced packaging and overseas fabrication will earn returns comparable to the assets producing today's cash flow.
Three developments would strengthen the case:
First, revenue growth continues to outpace wafer shipment growth without sustained margin deterioration.
Second, N2 ramps while N3 retains enough demand to support high utilization across both generations.
Third, operating cash flow grows fast enough to absorb rising capital expenditure without a persistent decline in free cash flow conversion.
The opposing case is equally clear.
If capital intensity rises faster than the earnings power of new capacity, TSMC could continue delivering impressive revenue growth while its incremental return on capital weakens.
Investors should not confuse those two outcomes.
TSMC's technological moat is well established. The investment question is how much of that moat will translate into returns on the next generation of manufacturing capital.
That is the real test of its AI premium.
