Taiwan Semiconductor reported earnings last week, providing the first glimpse into the AI semiconductor industry in the second quarter. Riding strong AI chip demand, TSMC boosted its full year revenue growth guidance, yet there are still some lingering doubts about the chipmaker’s growth in Q4.
HPC revenue continues to accelerate, reaching a record at $18 billion in Q2. Net profit reached a record in Q2 as margins outperformed. With that said, TSMC signaled significant margin pressures in Q3 from FX and ramping overseas fabs.
TSMC boosts FY revenue by 5 points
Q2 revenue increased 44.4% YoY and 17.8% QoQ in USD to $30.07 billion, well above TSMC’s guidance for $28.4 billion to $29.2 billion in revenue. This was driven by strong demand for AI accelerators built on TSMC’s 3nm and 5nm nodes.
For Q3, TSMC guided revenue of $31.8 billion to $33 billion. At midpoint of $32.4 billion, this represents YoY growth of 37.8% and QoQ growth of 7.8%. FX is playing a role here in this high-37% guide, with TSMC noting that revenue growth in NT$ is expected to be negatively impacted by 6.6 points based on current exchange rates, or ~31.2% YoY.
For the full-year, TSMC boosted its revenue growth forecast from mid-20% YoY to close to 30% YoY, driven by robust AI and HPC demand.
As it stands, TSMC’s growth in the first three quarters is far above the close to 30% guide, suggesting Q4 could see growth in the single digits. Management said they remain more conservative on Q4 at the moment. This is because although Q4 is typically seasonally strong for consumer electronics and smartphones, there is risk that tariffs put a damper on growth.
Management added that they have not seen any changes in customer behavior, but they are well aware of uncertainties from tariffs on consumer and price-sensitive end-markets. They added that Chinese rebate programs are stimulating near-term demand upside, but this is expected to phase out rather quickly and only create a mild recovery in non-AI demand this year.
HPC revenue rises 14% QoQ and up 19.2% CC
TSMC continues to benefit from robust AI accelerator demand, with HPC now accounting for three-fifths of the chipmaker’s revenue. TSMC stated that HPC revenue rose 14% QoQ in NT$ in Q2, though this increase was more pronounced on a US$ basis due to FX. TSMC’s revenue is recognized in US$, so every 1% appreciation of the NT$ adversely impacts NT$ reported revenue by ~1%.
Therefore, on a US$ basis, HPC revenue rose 19.2% QoQ to ~$18 billion, or up nearly $3 billion from Q1 in constant currency — its largest growth on record. The pace of acceleration in HPC revenue has been astonishingly quick, as the segment is now 2.5x larger than it was two years ago at $6.9 billion. HPC’s share of revenue also has increased 8 points YoY to 60%.
Smartphone revenue increased 7% QoQ to account for 27% of revenue in Q2, notably the strongest sequential increase for the segment since 2022.
IoT revenue increased 14% sequentially to account for 5% of revenue, while automotive was flat and also accounted for 5% of revenue. DCE increased 30% sequentially to account for 1%, while other segments rose 14% sequentially to 2% of revenue.
Revenue by Technology
TSMC’s advanced nodes – 3nm, 5nm, 7nm, and soon, 2nm – contribute the majority of revenue at 74% in Q2, fueled by AI accelerators and smartphones. 3nm ticked back up to 24% of revenue in Q2, while 5nm held flat at 36% of revenue, supported by Nvidia’s Blackwell GPUs. 7nm saw its contribution shrink one point to 14%, while mature nodes also shrunk one point to 26%.
TSMC shared some more details about its upcoming advanced nodes, stating that it remained on track for volume production on its 2nm node beginning in the second half of 2025, with a ramp profile similar to the 3nm node. Management also stated that they “expect the number of new tape-outs for 2nm technology in the first 2 years to be higher than both 3nm and 5nm,” driven by HPC and smartphone.
For the A16 node (1.6nm), TSMC said that it remains on track for volume production in the second half of 2026, believing this node will be best suited for “HPC applications with complex signal routes and dense power delivery networks.”
Margins guided to decline due to FX headwinds in Q3
Despite some FX headwinds to gross margin, TSMC’s operating margin outperformed, driving profit to a record level in Q2. However, Q3’s guide showed increasing FX headwinds and sharper sequential impact on margins.
- Gross margin was 58.6% in Q2, at the high end of the guided range for 57% to 59%. Gross margin declined sequentially from 58.8%, with a 2.2 point headwind from FX and a 1 point headwind ramping overseas fabs offset by higher capacity utilization.
- Operating margin was 49.6%, increasing 1.1 points sequentially from operating leverage, above guidance for 47% to 49%. Operating margin was up more than 7 points YoY.
- Net margin was 42.7%, down slightly from 43.1% in the prior quarter but up nearly 6 points YoY.
For Q3, margins will decline due to FX:
- TSMC guided gross margin to decline sequentially to 55.5% to 57.5%, or 2.1 points to 56.5% at midpoint. This is again due to continued FX headwinds, with approx. 2.6 points from unfavorable FX and overseas fab ramp in Kumamoto and Arizona.
- Operating margin was guided to decline sequentially to 45.5% to 47.5%, or 3.1 points at midpoint. This would be the lowest level since Q2 2024.
EPS increased 67% YoY, up from 54%
TSMC delivered record profit in Q2, rising 61% YoY to NT$398.3 billion, or ~$12.8 billion. Adjusted EPS of $2.47 beat estimates for $2.31 and increased nearly 67% YoY, accelerating from recent growth in the 50% range. However, EPS growth is forecast to decelerate rather rapidly through Q4, with TSMC barely maintaining double-digit growth.
For 2025, adjusted EPS is expected to increase 35.2% YoY to $9.52, up from 31.5% growth two months ago.
Cash Flows dip yet Balance Sheet is Healthy
Cash flow margins dipped by a larger margin sequentially, and TSMC’s balance sheet remained healthy.
- Operating cash flow was $16.2 billion for a 53.8% margin, down from a 74.5% margin in Q1 and a 56.1% margin in the year ago quarter.
- Free cash flow was $6.5 billion for a 21.7% margin, down from a 35.1% margin in Q1 and a 25.5% margin in the year ago quarter.
- Cash, equivalents and marketable securities totaled $90.4 billion, while debt totaled $32.3 billion.
- Capex rose more than 51% YoY to $9.6 billion, slowing from a 74% pace in Q1. TSMC maintained its full year capex guide at $38 billion to $42 billion.
- Inventories were $10.43 billion, up from $8.83 billion in Q1; however, the sequential increase looks to have been impacted by FX, as inventories in NT$ were up less than 4% QoQ.
Earnings call Q&A
While management offered little to no clarity on long-term AI growth or CoWoS capacity, they discussed long-term diversification of advanced node manufacturing to the US. Management also offered insights into advanced node capacity that signal Nvidia’s growth could remain strong come 2H.
Arizona Expansion
TSMC provided an update on its global expansion plans, which is important considering onshoring US manufacturing helps reduce geopolitical risk from China for the AI server supply chain. Management shared that they are expecting to bring 30% of their 2nm manufacturing to the US in Arizona.
As a result, TSMC is accelerating and expanding its presence in Arizona with its recent $165 billion investment plan, for six fabs, two advanced packaging fabs, and a major R&D facility to meet high multi-year demand from customers. The second fab in Arizona, utilizing 3nm tech, has finished construction with TSMC aiming to speed up volume production by several quarters. The third fab, offering 2nm and A16 advanced nodes, is under construction.
Management added that “despite the higher cost of overseas fabs, we will leverage our increasing size in Arizona and work on our operations to improve the cost structure,” to help minimize gross margin dilution impacts. This is important considering TSMC is forecasting increasing margin dilution from its growing overseas presence, widening from “2% to 3% every year in the early stages and widen to 3% to 4% in the later stages” over the next five years.
Nvidia H20 Impact
Morgan Stanley’s Charlie Chan questioned about Nvidia’s approval to resume shipping its H20 GPU to China, and if unlocking the Chinese market again would help TSMC raise its AI accelerator growth CAGR upwards from the mid-40% range.
C.C. Wei was rather tight-lipped about the potential impact, given that shipments have (likely) not yet resumed, saying that it is too early to provide an estimate on how this would impact growth. Wei explained that TSMC is not yet ready to increase its forecast, and “another quarter probably will be more appropriate to answer your question,” hinting that the AI accelerator CAGR may be updated in Q3.
However, it is expected that a majority of the H20s to be sold will be from existing inventory in the Taiwanese supply chain, meaning chips already built and revenue already booked. Therefore, it’s hard to see how TSMC could meaningfully increase its CAGR for the next four years based on just the H20.
In terms of China, however, the bigger opportunity here for TSMC may stem from Nvidia’s China-specific Blackwell B30 GPU, which is estimated to see shipments of up to 1.2 million units, or ~20% more than the total estimated H20 inventory. The B30 is expected to hit the market in Q4, following the H20’s resumption largely in Q3.
Advanced Node Capacity
Goldman Sachs’ Bruce Lu asked management about advanced node capacity, and supply-demand imbalances as more AI chips begin to shift to the 3nm node. While TSMC did not offer much beyond capacity being tight, one comment suggested Nvidia’s demand remains very strong.
C.C Wei explained that TSMC’s 5nm capacity is “very tight,” while 3nm capacity is even tighter and will continue to remain tight for a couple of years. Wei explained that TSMC can quickly retool advanced node fabs, such as 7nm to 5nm, 5nm to 3nm, etc, and keep utilization high to help meet demand. He would not commit to saying that demand would outpace supply, but that was implied given his comments of trying to “narrow the gap” between supply and demand.
However, one of the more important comments here was Wei stating that TSMC is using 7nm capacity to support 5nm demand, which provides another piece of evidence alongside surging HPC revenue that Nvidia’s Blackwell ramp is accelerating rapidly. Blackwell is built on TSMC’s N4P process, a subfamily of its 5nm node, offering higher performance, better power efficiency and higher transistor density.
Nvidia CEO Jensen Huang had stated that Nvidia was shipping nearly 1,000 racks per week to hyperscalers in May and expecting to ramp further, and this comment from TSMC supports lasting 5nm demand, likely from Nvidia given its high share of CoWoS (and manufacturing) capacity.
Conclusion
TSMC’s earnings provided more evidence that AI GPU demand remains strong, particularly for Nvidia, with HPC revenue rising to a fresh record with its largest QoQ increase, and commentary for 7nm capacity helping meet high 5nm demand. Supported by this robust AI and HPC demand, TSMC boosted its full-year guidance from mid-20% revenue growth to close to 30% growth, despite lingering concerns of tariff-related weakness come Q4.
Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.
Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.
Recommended Reading: