Earnings Preview | Is TSMC (TSM.US) on the Path to a $3 Trillion Market Cap?

According to Zhitong Finance APP, TSMC (TSM.US) will release its latest full quarterly results before the US market opens on Thursday. Despite several macroeconomic and industry headwinds, TSMC delivered excellent revenue performance in the third quarter, highlighting the strong resilience of demand for artificial intelligence infrastructure.

By 2027, increasing constraints on advanced process nodes and packaging technologies emphasize that capacity expansion is the key driver for TSMC’s growth and profitability improvement. While overseas expansion poses risks of profit margin dilution, higher utilization rates, operating leverage, and wafer price increases may offset the upcoming headwinds.

Revenue “Ahead of Schedule”: $46.7 Billion, Surpassing Guidance Upper Bound

TSMC has not yet released its complete quarterly earnings report, but the revenue figures have already been announced. According to TSMC’s revenue report published on October 8, consolidated revenue in September was NT$511.857 billion, down 0.6% MoM but up 54.6% YoY, a record high for the same period in history; August’s NT$514.806 billion remains the highest single-month record ever. Total consolidated revenue for the third quarter reached NT$1.494243 trillion, up 17.62% quarter-on-quarter, setting a new single-quarter high. Cumulative revenue for the first nine months was NT$3.898727 trillion, up 41.1% YoY.

Based on the company’s guidance using an assumed exchange rate of USD 1 to NT$32, third-quarter revenue is about $46.7 billion, not only surpassing the previous guidance range of $44.6 billion to $45.8 billion, but also exceeding the market consensus of $45.8 billion. Analysts point out that this record revenue provides a strong rebuttal to recent market concerns about the “sustainability of AI infrastructure spending”—worries that had been fueled by rising AI-related debt, the Fed’s resumption of interest rate hikes, and increased scrutiny of AI safety in the industry.

TSMC has scheduled its earnings conference call for October 15 (Thursday); as usual, the ten days prior to the earnings release (October 5 to 14) constitute a quiet period. Taiwanese investors expect, driven by the new Vera Rubin platform and 2nm demand, that fourth-quarter revenue will climb further. Key focuses include capital expenditure, progress in advanced processes and packaging, gross margin outlook, and the global capacity layout.

Concerns Over AI Infrastructure, and a Realized Contrarian Signal

In recent months, overall performance of the AI semiconductor sector has weakened amid industry headwinds. This sensitivity was starkly exposed in early September: Anthropic CEO Dario Amodei issued an open letter, advocating a slowdown in frontier model development for AI safety reasons, immediately triggering a sharp decline in semiconductor indicators, with both the VanEck Semiconductor ETF (SMH) and the 3x leveraged Semiconductor ETF (SOXL) significantly impacted. Related concerns have since eased—Jensen Huang of Nvidia (NVDA.US) and Lisa Su of AMD (AMD.US) both advocate continued but responsible advancement of AI; and industry expectations for incremental compute power for evaluation and monitoring workloads reinforce the continuity of infrastructure spending.

However, recent setbacks in ongoing projects have revived doubts: Oracle (ORCL.US) has issued force majeure notices to developers for its New Mexico 2.45GW data center campus Project Jupiter, seeking to defer some payment obligations if external factors like electricity and approvals delay delivery. Reportedly, the crux is that pipeline right-of-way was denied twice by the state, pushing the pipeline start date to February 1, 2027, and the project’s $18 billion syndicated construction loan now trades at under 90% of face value. This illustrates that constraints on AI compute deployment have shifted from chip supply to power and regulatory approvals.

Main Thread One: 2027 Capacity Crunch, and “Can Demand Be Monetized”

The driving logic on the demand side has shifted. AI infrastructure needs are increasingly determined by total token consumption, and agent-based AI, with its ongoing reasoning loops across task planning, execution, and evaluation, amplifies this consumption. Industry estimates show that post-training workloads may require compute capacity as much as 30 times that of initial foundation model training; and for high-intensity inference workloads, compute consumption may be over 100 times that of simple tasks like email summarization. Dell (DELL.US) expects inference tokens to grow 87-fold to 36 quadrillion by 2030, a projection later echoed by AMD.

As token consumption rises, TSMC faces stronger wafer demand, further boosted by two factors: first, the shift to next-generation commercial processors—Nvidia’s Vera Rubin system, AMD’s Instinct MI450X accelerator, and Zen 6 EPYC Venice server CPUs are all designed for increasingly complex agent inference workloads; second, adoption of custom ASICs is broadening, as hyperscale operators increasingly tailor architectures by workload to optimize unit economics for inference at scale.

This tightens the supply side further. TSMC management previously stated that demand will remain “very strong” in 2029 and 2030; at the last quarterly call, C.C. Wei stressed that clients and their customers (i.e., AI data center operators) persistently offer strong outlooks, advanced process capacity is still outstripped by demand, and the company is not concerned about competition from Intel or Samsung.

The bottleneck is focused on the highly popular and constrained N3 node. Reports indicate that N3 accounts for roughly 30% of TSMC’s wafer revenue mix and is the technology underpinning next-gen AI processors like Nvidia’s Vera Rubin architecture, with incremental capacity still notably limited. According to last quarter’s disclosures, advanced processes at 7nm and below collectively accounted for 77% of wafer revenue (3nm at 30%, 5nm at 33%, 7nm at 11%), while 2nm contributed 3% for the first time. The company is accelerating N3 capacity expansion in Taiwan, Arizona, and Japan, as well as repurposing N5 equipment to supplement capacity.

N2’s ramp-up is another catalyst, expanding from smartphones into high-performance computing, with demand drivers including Apple’s (AAPL.US) new A20 Pro and M6 chips (for the latest iPhone 18 Pro and Mac mini, respectively), as well as AMD’s next-gen EPYC Venice server CPU ramp. Taiwanese institutional estimates suggest that if 10 million AI GPUs are equipped with HBM4e custom base dies, they will consume 45,000 to 50,000 3nm wafers per month—accounting for 20% to 25% of 3nm output; even if monthly 3nm capacity grows from 160,000 wafers at end-2026 to 220,000 by end-2028, 2027 utilization will still be over 120%. This is precisely why TSMC is making an unprecedented investment this year in three new 3nm fabs.

Therefore, the execution speed of capacity expansion becomes the key variable for this earnings call—it determines whether TSMC can convert strong end demand into incremental revenue beyond what is already reflected in the current share price.

Main Thread Two: Overseas Margin Dilution, and Impending Price Hikes

Another looming issue is profit margin dilution caused by overseas expansion. TSMC’s third-quarter guidance for gross margin is 65% to 67%. Even at the upper bound, this reflects at least a 70-basis-point compression from the second quarter’s 67.7%. Management has repeatedly reminded investors that, in the early stages of overseas capacity ramp-up, company-wide gross margins could be diluted by 2 to 3 percentage points per year, expanding to 3 to 4 points later. The main reason is the initially higher costs of 2nm mass production, with the rapid ramp diluting margins by about 3 to 4 points. Capital expenditure expansion further exacerbates these concerns. In July, TSMC increased its Arizona investment by $10 billion, with total commitments reaching $26.5 billion; its 2026 capex outlook was also raised from $52–56 billion to $60–64 billion. There are also reports of a potential manufacturing site in Texas, and of partially leasing future Terafab plant capacity from Tesla (TSLA.US) and SpaceXAI. Amidst rising capital costs, this more capital-intensive expansion cycle introduces risks of incremental return compression.

However, persistently better-than-expected revenue is reinforcing confidence that operating leverage from advanced capacity scaling may outperform overseas dilution, forming an earnings inflection point beyond market expectations. Evidence comes from the second quarter: revenue reached $40.2 billion at the high end of guidance, gross margin at 67.7% and operating margin at 60.3% exceeded guidance upper bounds, after-tax net profit was NT$706.6 billion, up 77% YoY, and EPS was NT$27.25 (about $4.31 per ADR), with annualized ROE at 45.9%.

More importantly, the YoY growth rate for monthly revenue in the third quarter accelerated each month—reaching 54.6% in September. Taiwanese analysts therefore estimate that, with initial dilution from 2nm offset by product mix optimization, TSMC’s third-quarter gross margin could reach 68%, better than the guidance target of 65%–67%. Goldman Sachs and Morgan Stanley are relatively conservative, expecting gross margin to reach 67.5% in the third quarter.

The variable that could truly change the earnings trajectory is price hikes. Reports indicate that TSMC is expected to launch across-the-board price adjustments in 2027: base prices for advanced processes at 7nm and below will rise 5% to 10%, advanced packaging 10% to 15%, and mature processes 3% to 5%; customers requesting additional high-performance computing capacity on top of existing orders will pay an extra 10% to 15% premium, making some additional orders up to 25% higher. There are also reports that 2nm wafers may get an additional 6% to 8% increase, and some mature nodes up to 10%. The price adjustments are intended to “offset rising costs of materials, manufacturing equipment, and overseas plant construction”; while TSMC has not confirmed this, it aligns with their “higher where possible” pricing philosophy, provided it doesn’t undermine client success or “squeeze” customers out of the market.

Morgan Stanley, based on this, estimates that 7nm and below advanced processes will see price hikes of 5% to 10% in 2027, with mature processes increasing by no more than 3%, pushing TSMC’s 2027 gross margin close to 70%. The sector-level pricing environment supports this: earlier reports noted that Samsung Electronics has raised some advanced process new orders by up to 15% (4nm up by 10–15% for US and mainland Chinese clients, 5nm by the same amount, and 8nm by nearly 10%); Counterpoint Research data shows the global pure-play foundry market grew 29% YoY in Q2 2026, with TSMC holding a 73% share—ten times that of runner-up Samsung (7%).

Thus, further explanations on pricing strategy at this earnings call will be a core highlight: average selling prices driven by the supply side can provide TSMC’s earnings trajectory with a critical safety cushion to offset overseas dilution and the costs of 2nm ramp and A14 development.

Valuation: Divergence in Institutional Target Prices

According to Summit Research’s valuation framework: TSMC is expected to achieve a five-year revenue CAGR of 28% through 2030, driven by a higher share of advanced processes supporting higher unit-value AI high-performance computing opportunities; platform-related revenue CAGR is estimated at 35%. Using a DCF model with a 3.5% perpetual growth rate for 2030 EBITDA terminal value and a 10.3% WACC, the baseline target price is $498 per share; if capacity monetization is stronger, the optimistic scenario sees $535 per share, making a $3 trillion market cap “within reach.”

Sell-side analysts have launched a wave of target price upgrades ahead of the earnings call. JPMorgan (J.P. Morgan) Taiwan Research Head Gokul Hariharan raised his target from NT$3,200 to NT$3,300, rating “Overweight.” Morgan Stanley (MS) Taiwan Research Head Chi-Hung Chan increased his target from NT$2,988 to NT$3,088, also “Overweight.” Goldman Sachs maintained a “Buy” rating with a target of NT$3,300, and UBS kept its target at NT$3,650, highlighting significant “re-rating” potential for the stock.

Goldman Sachs semiconductor analyst Bruce Lu expects TSMC’s 2026 and 2027 dollar revenue to grow by 42% and 36.9% YoY, with gross margins at 67.2% and 67.5%, and raised capex forecasts for 2027 and 2028 to $85 billion and $98 billion, respectively. Morgan Stanley’s Chi-Hung Chan expects the company to raise its annual revenue growth guidance to a lower range of 40%–45%, fourth-quarter revenue to rise 10%–15% quarter-on-quarter, 2026 capex to stay at $54 billion, and 2027 to expand to $75–80 billion; he also listed three key topics for the call: US capacity expansion (a possible Texas site within five years), Intel outsourcing trends, and MediaTek’s large TPU chips with big-die CoWoS or CoPoS packaging support.

Ultimately, investors want to hear two essential things from this earnings call: First, the visibility of capacity expansion execution—whether end demand can be monetized; second, the resilience of gross margins under overseas dilution pressure, which determines if free cash flow can continue to expand.