Can TSMC's Overseas Fabs Improve and Still Dilute Its Margins?

The cost difference hidden inside an earnings call

TSMC's third-quarter results, released on 15 October, showed US-dollar revenue of $46.72 billion and a gross margin of 66.4%. The company also confirmed that the cost of manufacturing overseas remains higher. Management expects overseas fabrication facilities to dilute the group's gross margin by about two percentage points during 2026, with dilution of three to four percentage points in the latter part of its five-year outlook. [1][2]

That forecast looks counterintuitive. TSMC is expanding its Arizona manufacturing system, creating opportunities to spread facilities and support costs across more output. Management also expects productivity and cost improvements. Why, then, could the company's aggregate margin penalty increase?

The answer is that the cost penalty per unit and the total cost penalty are different quantities. Efficiency can improve at each factory while the company puts more of its production through factories that are still more expensive than its Taiwanese operations.

What two percentage points mean financially

TSMC reported $35.90 billion in first-quarter US-dollar revenue, $40.20 billion in the second quarter and $46.72 billion in the third. For the fourth quarter, management guided to $52.3 billion to $54.3 billion. Adding the midpoint of that guidance to the first three reported figures produces an indicative 2026 revenue base of $176.12 billion. [1][3][4]

Multiplying that base by two percentage points gives $3.52 billion of annual gross-profit equivalent. At the same unchanged revenue base, three to four percentage points correspond to $5.28 billion to $7.04 billion.

Source: Rokin Research calculations using TSMC's reported January-September 2026 US-dollar revenue and the midpoint of its Q4 guidance. The 3-4 point rows hold 2026 revenue fixed solely for comparison; they are not future earnings forecasts or disclosed fab losses.

The denominator matters. TSMC has not disclosed the overseas share of comparable wafer revenue or a Taiwan-versus-Arizona cost difference on matched chips. Its margin-dilution guidance is an estimated effect on the consolidated group, not a cash bill for its US factories. It also includes overseas operations beyond Arizona.

Why lower unit costs can coexist with greater dilution

Consider an illustrative overseas production system with a total cost premium of two units relative to producing equivalent output elsewhere. Suppose overseas output doubles, while its extra cost per unit falls by 12.5%.

The new total premium is 2 x (1 - 0.125), or 1.75 times the old premium. On a fixed company revenue base, an illustrative two-percentage-point dilution would rise to 3.5 points. Nothing in that example requires factory-level productivity to deteriorate.

To offset a doubling of overseas output fully, the additional cost per unit would have to fall by at least 50%, with company revenue, product mix, pricing and the underlying counterfactual held constant. The model is a conditional arithmetic experiment, not an estimate of TSMC's actual volume growth or achievable cost reduction.

This distinction is important during expansion. Newly commissioned equipment absorbs capital before it reaches mature utilisation. Improving yield also raises the amount of usable output from a given number of wafer starts. These changes alter unit economics, while opening further factories increases the volume exposed to the overseas cost structure.

What an Arizona factory benchmark does, and does not, show

TSMC's original December 2022 Arizona announcement envisaged roughly $40 billion of investment for two fabs and more than 600,000 wafers of annual output when complete. Those are historical project figures, not the scope or cost of the much larger expansion announced subsequently. [5]

They can still illustrate capital intensity. Treating $40 billion and 600,000 annual wafers as a stylised historical planning case, and spreading the investment evenly over 15 years, produces an annualised capital burden of about $7,407 per started wafer at 60% utilisation. At 90% utilisation it falls to about $4,938. These are not TSMC's depreciation charges or wafer manufacturing costs; no labour, energy, materials, equipment replacements or yield adjustments are included.

The comparison shows the direction of the operating leverage, not the size of any present-day US-Taiwan gap. Without matched-node facility data, the difference cannot be quantified responsibly.

What the result means for semiconductor reshoring

TSMC has said its customer pricing considers production location as well as technology and market segment. That matters because manufacturing competitiveness and commercial profitability need not be identical. A customer may pay for geographic production flexibility, even if the fab carrying out the work remains more costly to run. [2]

For investors, the relevant observation is not simply whether overseas gross-margin dilution rises or falls. It is how that dilution changes relative to overseas output, operating maturity and the pricing customers accept. For industrial policy, the distinction is between financing the construction of a factory and establishing a production system that can operate competitively at scale.

The current evidence supports a measured conclusion: TSMC's guidance puts a significant group-level figure on the cost of geographic diversification, but does not identify an American cost premium per wafer. A rising aggregate margin effect can coexist with improving overseas factory efficiency. Judging whether the strategy earns its cost requires information on site-level output, pricing, depreciation and yield that TSMC has not publicly separated.

Research note

This article combines TSMC's October 2026 disclosures with transparent arithmetic and illustrative operating scenarios. The revenue-based equivalents are not reported losses, and the historical Arizona exercise does not estimate current US production costs. The accompanying report develops the model boundaries, sensitivities and outstanding evidence requirements; the workbook makes the scenario assumptions editable.

Sources & References

[1] TSMC, Third Quarter 2026 Earnings Release, 15 October 2026. Consolidated results, US-dollar revenue, gross margin. investor.tsmc.com/english

[2] TSMC, Q3 2026 earnings conference transcript, call of 15 October 2026, pp. 3, 14, 18 (LSEG edited transcript distributed by TSMC). Margin guidance, later-period clarification and geographic pricing. Open source

[3] TSMC, First Quarter 2026 Results, 16 April 2026. US-dollar revenue. investor.tsmc.com/english

[4] TSMC, Second Quarter 2026 Results, 16 July 2026. US-dollar revenue. investor.tsmc.com/english

[5] TSMC, TSMC Announces Updates for TSMC Arizona, 6 December 2022. Original two-fab capital and output intentions (historical). pr.tsmc.com/english