TSMC Lifts 2026 Capex to $64 Billion, But Margin Caution Sends Mixed Signals

Taiwan Semiconductor Manufacturing Co. (TSM) delivered a second-quarter report on July 16 that landed squarely in the zone of elevated expectations—strong enough to validate the AI boom, but not explosive enough to silence the growing chorus of skeptics asking whether the chip trade has run its course. Revenue hit $40.2 billion, near the top of the company’s own guidance range, while gross margin came in at 67.7%, modestly above the forecast band. The headline numbers underscore a simple reality: TSMC remains the indispensable gatekeeper of advanced silicon, with nearly every major AI accelerator passing through its fabs. Yet the market’s tepid reaction suggests investors were primed for a blowout, not a beat.

Speaking at the company’s earnings call, Chairman and CEO C.C. Wei offered a characteristically blunt assessment of the supply-demand dynamic. “It will be a long time before we can meet customer demand,” he said, a line that has become something of a mantra as the company races to expand capacity. The statement encapsulates the central tension of the TSMC story: extraordinary growth is already priced in—Wei raised the full-year revenue growth target to “above 40%,” up from “at least 30%”—but the very constraints that guarantee pricing power also cap how quickly the company can convert AI hype into incremental earnings.

The quarter’s composition tells a story of structural transformation. High-performance computing, the segment that houses AI chip revenue from customers like Nvidia (NVDA), AMD (AMD), and Broadcom (AVGO), swelled to $26.5 billion, or 66% of total wafer revenue. Smartphone revenue, long the seasonal anchor of TSMC’s business, slipped to $8.8 billion, squeezed to roughly 20% of the mix. The shift is more than cyclical. With Nvidia’s Rubin platform, AMD’s MI350 series, and Google’s TPUv7 all migrating to the 3nm node, TSMC’s most advanced production lines are running at utilization rates above 100%. The 3nm node alone accounted for 30% of wafer revenue in the quarter, while the 2nm process—which entered commercial production in late 2025—contributed a material 3% for the first time.

“Revenue growth this quarter was driven by the migration of core customers’ AI chips to the 3nm platform,” the company noted in its earnings materials. The numbers bear that out. Wafer shipments rose 3.9% sequentially to 4,336 thousand 12-inch equivalent wafers, while average selling price per wafer jumped 7.8% to $9,271. The combination of higher volumes and richer mix is the hallmark of a company with structural pricing power, but it also raises the question of how much further margins can expand when the next-generation 2nm ramp is already introducing cost headwinds.

For the third quarter, TSMC guided revenue to $44.6 billion to $45.8 billion, ahead of the buy-side consensus of roughly $44 billion. But the gross margin outlook of 65% to 67% landed below the 67.5% that some analysts had penciled in. The culprit is the 2nm ramp, which TSMC has warned will dilute margins by 2 to 3 percentage points over the full year. Overseas fab expansions add another layer of margin pressure. The guidance, while directionally positive, lacked the upside surprise that might have reignited the broader AI trade after a bruising few weeks for semiconductor stocks.

The most consequential number in the report may not have been revenue or margin, but capital expenditure. TSMC raised its full-year capex target to $60 billion to $64 billion, a dramatic step up from the prior range of $52 billion to $56 billion and well above the $58 billion consensus. That implies second-half spending of $33.2 billion to $37.2 billion, a year-over-year surge of 56% to 75%. The increase is a direct response to the 2nm ramp and the insatiable demand for advanced packaging capacity, particularly the Chip-on-Wafer-on-Substrate technology that has become the industry’s most critical bottleneck.

CoWoS capacity has emerged as the true chokepoint in the AI supply chain. TSMC controls more than 90% of global CoWoS output, and current monthly capacity of roughly 90,000 wafers is expected to reach 120,000 by year-end, according to industry estimates cited by TrendForce. Nvidia alone has reportedly booked 60% of TSMC’s CoWoS capacity through 2026, plus more than half of the expansion planned for 2026-2027. The Economic Daily News in Taiwan has reported that the gap between CoWoS demand and available capacity could narrow from around 20% currently to roughly 10% by late 2026 as new production lines come online. If that timeline holds, hyperscale cloud providers could see reduced packaging delays in the second half of the year, accelerating the deployment of AI accelerators from Nvidia, AMD, and custom chip designers like Google, Amazon, and Microsoft.

The competitive landscape, while still heavily tilted in TSMC’s favor, is worth watching. Samsung Foundry and Intel (INTC) have entered their own advanced nodes—SF2 (2nm) and 18A (1.8nm), respectively—but both lag TSMC on transistor density and yield. Samsung and Intel’s latest nodes have transistor densities below 250 million transistors per square millimeter, compared to 294 MTr/mm² for TSMC’s previous-generation N3P. Both rivals are primarily producing their own chips, whereas TSMC serves a broad external customer base. Still, with AI chip capacity so tight, even modest improvements in advanced packaging and yield at Samsung or Intel could capture spillover orders, particularly given the political tailwind behind Intel’s U.S.-based manufacturing footprint.

The following table summarizes TSMC’s key second-quarter metrics against expectations:

Metric	Q2 2026 Actual	Guidance / Consensus	Notes
Revenue	$40.2 billion	$39.0–$40.2 billion guidance; ~$40 billion consensus	Near top of guidance range
Gross Margin	67.7%	65.5–67.5% guidance; 67–69% buy-side consensus	Modest beat; below some 69%+ expectations
Q3 Revenue Guidance	$44.6–$45.8 billion	~$44 billion buy-side consensus	Ahead of expectations
Q3 Gross Margin Guidance	65–67%	~67.5% buy-side consensus	Below elevated expectations
Full-Year Revenue Growth	“Above 40%”	“At least 30%” prior; ~35%+ consensus	Modest upgrade
Full-Year Capex	$60–$64 billion	$52–$56 billion prior; ~$58 billion consensus	Significant increase
2nm Revenue Contribution	3% of wafer revenue	N/A	First material contribution

Note: Buy-side consensus figures are based on analyst estimates compiled ahead of the earnings release.

Broader industry sentiment supports the view that AI demand remains robust, even if near-term stock performance has faltered. A survey of 151 semiconductor executives conducted by KPMG and the Global Semiconductor Alliance found that 93% expect higher industry revenue in 2026, with nearly three-quarters identifying AI as the most significant growth driver. The confidence index rose to 63, the third-highest reading in two decades. For the first time in the survey’s 21-year history, however, executives ranked tariffs and trade regulations as the biggest challenge facing the industry, a reminder that the geopolitical landscape is growing more complex even as demand fundamentals strengthen.

TSMC’s geographic concentration remains both a strategic asset and a vulnerability. North American customers—Apple, Nvidia, AMD, Qualcomm (QCOM), and others—accounted for 78% of revenue in the quarter. Revenue from mainland China held steady at roughly $2.4 billion, but its share of the total slipped below 10% as U.S. client growth outstripped all other regions. The company is expanding its Arizona fabs with U.S. government support, and it is accelerating the construction of two new packaging facilities in Taiwan, according to CNBC. These investments are designed to diversify manufacturing geography and ease CoWoS constraints, but they will take years to fully materialize.

The market’s reaction to the report reflects a reset in expectations rather than a rejection of the fundamentals. TSMC shares have risen more than 40% year-to-date, and the stock was trading at roughly $418 ahead of the earnings release, according to Yahoo Finance. The trailing price-to-earnings multiple has compressed below 20x as the broader AI sector has sold off, a level that historically has marked a floor for the stock. The issue is not whether TSMC is a well-run company with a near-monopoly on advanced manufacturing—it clearly is. The issue is whether the AI investment cycle can continue to accelerate at a pace that justifies the massive capital outlays TSMC and its customers are making.

Wei’s comments on pricing strategy offer a window into how TSMC is managing that tension. Asked whether the company could raise prices in the current supply-constrained environment, he acknowledged the temptation but drew a distinction with memory chip makers. “I’d like to do that. We still need to make money,” he said, according to a Seeking Alpha report. But he emphasized that TSMC is “focused on long-term, sustainable operations” and not interested in sudden price spikes that could damage customer relationships. That strategic restraint—leaving pricing upside on the table to protect partnerships with Nvidia, Apple, and Broadcom—is part of why the gross margin, while healthy, has not surged to the levels some bulls had projected.

Looking ahead, the second half of 2026 will test whether the AI supply chain can begin to ease its most acute constraints. If CoWoS expansion stays on track and 2nm yields continue to improve, TSMC is positioned to capture another leg of growth as hyperscalers deploy next-generation infrastructure. If packaging bottlenecks persist, the industry will remain supply-constrained regardless of how much wafer capacity comes online. Either way, TSMC remains the single most important company in the semiconductor ecosystem—a position that makes its earnings reports required reading not just for chip investors, but for anyone trying to gauge the trajectory of the AI revolution.