Taiwan Semiconductor Manufacturing Company (TSMC) has crossed a symbolic and strategic threshold: more than $104 billion in revenue over the 12 months ending in June 2025, a jump of 34% year-over-year. This figure, compiled from historical series by MacroTrends, reflects the combined strength of two major drivers shaping the decade: high-performance computing (HPC) for artificial intelligence and the smartphone renewal cycle.
This is not an isolated milestone. The second quarter of 2025 closed with $30.07 billion in sales (+38.6% YoY) and a net profit that grew nearly 61%, both records for the global leader in the “fabless foundry”—as reported by Futurum Group and CNBC. For the third quarter, TSMC projects $31.8–33 billion in revenue (+38% YoY), with firms like Bernstein estimating that the company will surpass its +30% annual growth guidance to reach +33% in 2025, supported by higher-than-expected monthly sales (Investing.com).
With this momentum, the market message is clear: TSMC continues to lead the frontier of advanced manufacturing—3 nm / 5 nm / 7 nm nodes now accounting for 74% of wafer revenue—and it also absorbs an increasing share of AI-driven demand in data centers, where the Taiwanese manufacturer has become both a bottleneck… and a facilitator.
The year’s engines: AI drives growth, smartphones support
The business segment breakdown for Q2 2025 illustrates TSMC’s “mix”:
- HPC (AI/5G): 60% of quarterly revenue (+14% QoQ).
- Smartphones: 27% (+7% sequential).
- Advanced nodes (≤7 nm): 74% of total wafers, with 3 nm contributing 24%, 5 nm 36%, and 7 nm 14%.
This concentration on cutting-edge technology—highlighted by Futurum’s analysis—reflects a near-structural competitive advantage: TSMC remains the only player capable of mass-producing the GPU, CPU, and accelerator nodes demanded by the new wave of applications, with wafer yields and installed capacity that rivals cannot yet match.
Who benefits from the party: Apple and Nvidia share the throne; AMD consolidates
Within its customer base, Apple maintains around 25% of TSMC’s revenue, while Nvidia has climbed to an estimated 20–22% for 2025 (Yahoo Finance, FPT). AMD accounts for about 7%. Behind them are players like Qualcomm, MediaTek, and Broadcom. The de facto duopoly of Apple–Nvidia—covering mobile ecosystems and AI—explains much of the reserved capacity and the queues seen in CoWoS (advanced packaging) over the past two years.
Looking ahead, TSMC reports having secured around 15 early customers for the 2 nm node (N2)—including Apple, AMD, Nvidia, Intel, MediaTek—signaling the most extensive adoption at this stage of maturity (TechPowerUp, TechSpot). Translation: the high-value pipeline is more diversified than ever before reaching mass production.
Technological roadmap: N2 in H2 2025 and A16 (backside power) in H2 2026
The 2024 annual report from TSMC details the schedule:
- N2 (2 nm): volume expected in the second half of 2025.
- A16: separately offered with backside power delivery, expected in volume in the second half of 2026.
The “but” comes with the cost: N2 wafers are reported to be about 50% more expensive than N3, at roughly $30,000 per wafer (TechSpot). Parallelly, price increases of 5–10% in 3 nm / 5 nm nodes and 15–20% in CoWoS (advanced 2.5D/3D packaging with silicon interposers) are anticipated due to demand surges and costs of capacity expansion.
Implication: while power efficiency per transistor improves, the economics for dies become more expensive. Customers with high software/service margins (hyperscalers) are best positioned to absorb this; others will need to optimize architectures and segment products based on cost/benefit considerations.
Arizona, tariffs, and geopolitics: from risk to “factory exemption”
Amid political debates in the US, Trump announced his intent to impose a 100% tariff on imported semiconductors, with an exemption for companies manufacturing—or committed to manufacturing—within the US (Fortune, Al Jazeera). Taiwan confirmed that TSMC will be exempt thanks to its factories in the US (Fortune Asia).
The company has committed $165 billion in Arizona for six advanced factories deploying N2, A16, and subsequent nodes (Manufacturing Dive). The first began mass production at the end of 2024 with 4 nm. Message: market access and geopolitical resilience in exchange for huge capex and a multinational operation.
However, TSMC’s CEO, C.C. Wei, warned in the latest earnings call that while he expects a temporary impact, there are uncertainties related to potential tariffs—especially in price-sensitive segments (Manufacturing Dive). Ultimately, the client and its elasticity will determine how the risk is distributed: not everyone can bear the same tax.
Where the pinch points are: exchange rates, overseas launches, and margins
- FX: CFO Wendell Huang explains that nearly 100% of revenues are sensitive to the dollar–NT$ exchange rate; each +1% appreciation of the NT$ reduces reported sales by about ~1%, and the gross margin shifts by about 40 basis points.
- Overseas ramp-up: the start of factories in Arizona and Japan has diluted margins by over 100 basis points until these plants improve efficiency (Manufacturing Dive).
- Guidance for Q3 2025 margins: gross margin 55.5–57.5% and operating margin 45.5–47.5%, slightly below the 58.6% and 49.6% of Q2 due to combined effects of FX and ramp-up.
Despite these headwinds, Bernstein projects a +40% increase in EPS this year, with AI revenues in data centers rising from 6% of total in 2023 to 14% in 2024 and over 20% in 2025. The growth leverage—and its mix—appears intact.
Why TSMC is the “invisible giant” of AI
Because no other player transforms designs from Apple, Nvidia, AMD, Intel, MediaTek and others into leading-edge silicon at comparable pace and scale. Because its capacity for 3 nm / 5 nm / 7 nm nodes and its pipeline of 2 nm / A16 are today essential to executing the roadmap of GPUs, CPUs, and accelerators powering LLMs and services. And because its packaging (CoWoS) is the quiet bottleneck—either constraining or accelerating each wave of generative AI.
The logic of the decade confirms: those who control advanced manufacturing, control the pace of innovation. TSMC might not dictate Moore’s Law, but it determines how fast others can apply it.
Where are the risks?
- Customer concentration: reliance on Apple and Nvidia is both a strength and a risk. Changes in mix or cycles could impact results.
- Rising costs: transitioning to N2 (and A16) demands increased capex and complex fab constructions. TSMC’s pricing power remains, but it’s not unlimited.
- Geopolitical factors: tariffs, export controls, and tensions add uncertainty. The factory exemption in the US helps mitigate some risks but does not eliminate them entirely.
- Competition: Samsung and Intel Foundry are closing the gap with aggressive investments. Capacity and yield will be key metrics to watch.
Verdict: from “silicon arena” to “digital gold”
TSMC will close 2025 as the central supplier driving the AI revolution. With annual revenues exceeding $104 billion, margins still in the high range, and a roadmap aligned with the needs of designers and hyperscalers, the Taiwanese manufacturer is not just producing chips: it’s setting the pace for technological progress.
However, global success comes with costs: FX risks, overseas ramp-ups, multibillion-dollar capex, and geopolitical tensions. If N2 and A16 meet milestones—and CoWoS continues to scale—TSMC will cement a virtuous circle: leadership in tech → pricing power → reinvestment → more leadership. Should any link weaken, the actor that everyone currently relies on to push the frontier will face the challenge of managing the risk distribution.
Frequently Asked Questions
How much did TSMC earn in the last 12 months, and what percentage is AI (HPC)?
TSMC surpassed $104 billion in revenue over the 12 months ending in June 2025 (+34% YoY). In Q2 2025, HPC—including AI and 5G—made up 60% of the quarterly sales; data center AI is estimated to have grown from 6% of total in 2023 to 14% in 2024 and over 20% in 2025 (Bernstein estimates).
What is TSMC’s schedule for 2 nm (N2) and A16, and how does this impact costs?
N2 volume production is expected in the second half of 2025; A16 with backside power delivery in the second half of 2026. N2 wafers are about 50% more expensive than N3, around $30,000 per wafer. TSMC is likely planning price increases of 5–10% in 3 nm / 5 nm nodes and 15–20% in CoWoS driven by demand and cost factors.
Who are TSMC’s main customers, and what roles do Apple and Nvidia play?
Apple accounts for roughly 25% of TSMC’s revenue; Nvidia has grown to around 20–22% in 2025. AMD’s share is about 7%. For 2 nm, TSMC secures around 15 early customers, including Apple, AMD, Nvidia, Intel, MediaTek, marking the broadest early adoption for a cutting-edge node at this stage.
How do US tariffs and Arizona factories influence risks?
The 100% tariff announced by Trump excludes companies manufacturing within the US; TSMC is exempt due to its facilities in Arizona, where it has committed $165 billion for six factories (including a 4 nm mass production site from late 2024). Still, the company recognizes some short-term uncertainties stemming from price sensitivity in certain segments.
What pressures do FX and global expansion exert on TSMC’s margins?
Each +1% appreciation of the NT$ against the dollar reduces reported sales by about ~1% and the gross margin by approximately 40 basis points. The ramp-up in Arizona/Japan has compressed margins by over 100 bps. For Q3 2025, TSMC guides a gross margin of 55.5–57.5% and an operating margin of 45.5–47.5%, slightly below the 58.6% and 49.6% of Q2.
Note: This synthesis draws on data and forecasts from MacroTrends, CNBC, Futurum Group, Yahoo Finance, Manufacturing Dive, Investing.com, TechPowerUp, and TechSpot.