Q4 FY2025 Earnings Recap

Taiwan Semiconductor Manufacturing Company Limited (TSM)

Reported Jan 15th 2026

News Summary

TSMC raised its 5-year AI accelerator revenue CAGR forecast to 'mid-to-high 50s percent' through 2029, with AI accelerators already representing high-teens percent of total 2025 revenue and HPC growing 48% year-over-year to 58% of full-year revenue, signaling that AI has structurally reshaped demand for leading-edge silicon.
TSMC's wafer supply—not power or other infrastructure—is the binding bottleneck for AI data center buildout, a posture reinforced by $52–56B in 2026 CapEx (a ~30% step-up) with 70–80% targeting advanced nodes and capacity explicitly sized for 2028–2029 demand, implying the AI chip supply-demand gap persists for years.
N2 entered high-volume manufacturing in Q4 2025 with strong yield at multiple fabs, with a fast ramp expected across HPC and AI applications in 2026; CFO Wendell Huang noted N2 will be larger in revenue than 3nm from the start, with N2P and A16 derivatives entering volume production in H2 2026.
C.C. Wei personally vouched for AI demand durability after cloud service providers showed him financial evidence that AI is growing their businesses, directly addressing bubble concerns and reinforcing that hyperscaler AI infrastructure spending is grounded in measurable returns.
TSM stock jumped on what was characterized as blowout Q4 results, with Q1 2026 revenue guided to $34.6–35.8B (38% year-over-year growth at midpoint) and full-year 2026 revenue expected to grow approximately 30%, reflecting broad analyst and market confidence in sustained AI-driven demand.

Financial Highlights

Metric	Q1 '24	Q2 '24	Q3 '24	Q4 '24	Q1 '25	Q2 '25	Q3 '25	Q4 '25
Revenue	$592.6B	$673.5B	$759.7B	$868.5B	$839.3B	$933.8B	$989.9B	$1056.0B
Rev Growth (YoY)	16.5%	40.1%	39.0%	38.8%	41.6%	38.6%	30.3%	▼ 21.6%
Gross Margin	53.1%	53.2%	57.8%	59.0%	58.8%	58.6%	59.5%	▲ 62.3%
Operating Margin	42.0%	42.5%	47.5%	49.0%	48.5%	49.6%	50.6%	▲ 53.9%
EV/Sales	38.5x	42.0x	36.3x	37.5x	32.4x	35.2x	43.0x	45.2x

Earnings Call Highlights

TSMC's Q4 2025 call was defined by an unmistakably bullish posture on AI demand, with C.C. Wei personally vouching for the reality of hyperscaler spending after direct conversations with cloud service providers who showed him financial evidence that AI is growing their businesses. Management raised its 5-year AI accelerator revenue CAGR guidance to mid-to-high 50s percent and projected overall company revenue CAGR of ~25% through 2029, signaling a structural conviction that goes well beyond typical cycle optimism. Analysts probed hard on bubble risk, Intel foundry competition, and supply constraints, but management deflected each concern with confidence rooted in technology lead times and customer lock-in dynamics. The dominant theme was a supply-constrained world where TSMC's wafer output remains the binding bottleneck for AI infrastructure, and the $52–56B CapEx commitment for 2026 is explicitly building for 2028–2029 demand, not the near term.

TSMC raised its 5-year AI accelerator revenue CAGR forecast to 'mid-to-high 50s percent' from 2024 to 2029, and C.C. Wei stated flatly that 'the capacity is very tight' and that TSMC's wafer supply—not power or other infrastructure—remains the primary bottleneck for AI data center buildout, a direct signal of sustained demand for leading-edge silicon.
CapEx guidance of $52–56B for 2026 represents a ~30% step-up from 2025's $40.9B, with 70–80% allocated to advanced process technologies; management clarified this spending targets 2028–2029 capacity, meaning the supply-demand gap for AI chips will persist through at least 2027.
AI accelerator revenue accounted for 'high teens percent' of TSMC's total revenue in 2025, and HPC as a platform grew 48% year-over-year to represent 58% of full-year revenue, underscoring how thoroughly AI has reshaped TSMC's revenue mix.
N2 entered high-volume manufacturing in Q4 2025 with 'good yield' at both Hsinchu and Kaohsiung, with a fast ramp expected in 2026 across smartphone and HPC/AI applications; CFO Wendell Huang noted N2 will be 'a bigger node than 3-nanometer from the start' in revenue dollars, and N2P and A16 derivatives are scheduled for H2 2026 volume production.
On Intel foundry competition, C.C. Wei was dismissive, noting that designing for and ramping a new advanced node takes 3–5 years in total, and that TSMC has 'competed for 30-some years' without losing confidence—a posture that suggests TSMC does not view near-term share risk as credible given technology complexity and customer switching costs.