How's TSMC doing these days?
In our semiconductor industry deck, you will find everything you need to understand the market
SUMMARY
How's TSMC doing these days? TSMC is doing extremely well right now, and the evidence is strong enough to say the company’s momentum is operational, not just stock-market excitement.
The clearest signal is growth quality. TSMC is still growing around 30% year over year while keeping margins at a level that most manufacturing-heavy businesses could not reach even in a perfect cycle.
AI is still the main growth engine, but the story has matured. The useful signal is no longer just “Nvidia needs more chips”; it is that Nvidia, Broadcom, AMD, Apple, and custom-silicon buyers are all negotiating for access to the same scarce manufacturing system.
TSMC’s 2nm transition now looks real rather than theoretical. AMD’s EPYC Venice ramp on N2 is especially important because it shows 2nm moving into server and AI infrastructure, not just premium consumer devices.
The competitive gap remains huge. Samsung and Intel are still relevant enough to matter strategically, but TSMC’s 72.3% foundry share versus Samsung’s 6.5% in Q1 2026 shows that the commercial default has not changed.
Advanced packaging may now be as important as leading-edge wafers. CoWoS is not just a bottleneck; it is one of the reasons TSMC stays in control of the AI hardware stack.
The Nvidia dependency risk is real, but it is not the whole story. Broadcom custom AI chips, AMD server CPUs, Apple 2nm demand, and mobile customers make TSMC’s scarcity broader than a single-customer cycle.
Arizona is becoming a real operating asset, but it still looks like geopolitical insurance rather than a replacement for Taiwan. The $20 billion capital injection and early profit signal matter, while labor, water, energy, and supplier localization remain hard constraints.
The under-discussed constraint is talent. TSMC can spend on fabs and equipment, but training thousands of engineers across Taiwan, Arizona, Japan, and Germany is slower and more fragile than ordering more machines.
Power and water are now part of the growth equation. The AI cycle does not only require more chips; it requires land, grid stability, clean power, water access, and political permission to keep scaling.
The most interesting shift is that TSMC is no longer just selling wafers. It is selling trusted access to the manufacturing system behind AI accelerators, Apple silicon, custom ASICs, server CPUs, and national semiconductor policy.
So the conclusion is pretty clear: TSMC looks stronger than ever today, but the challenge has changed. The question is no longer whether demand is there; it is whether TSMC can create enough TSMC.
This market map, featured in our semiconductor industry deck, highlights top companies and startups in the semiconductor industry
Is TSMC still growing fast right now?
TSMC is still growing very fast, and the recent numbers are strong enough that we can say this is not just market excitement around the stock.
The cleanest signal is TSMC’s own monthly revenue. In May 2026, the company reported NT$416.98 billion in revenue, up 30.1% year over year. For the first five months of 2026, revenue was up 30.0%. That is important because it is not one explosive month carried by one customer cycle but an actual five-month run-rate that says demand is still coming through the P&L.
The Q1 2026 earnings release made the same point in a more aggressive way. Revenue reached $35.9 billion, gross margin was 66.2%, and operating margin was 58.1%. For a manufacturing-heavy business, that margin profile is almost absurdly strong. It means TSMC is not buying growth by cutting prices or over-discounting capacity. It is growing while keeping the economics of scarcity.
There is also a market signal hidden in the reaction. Investor’s Business Daily reported in June 2026 that TSMC’s May revenue kept it on track for Q2 guidance, even though the stock sold off after a recent high. That tells us something useful: the operating business is still beating forward, but investors are already asking whether AI expectations have become too crowded.
So today, TSMC is not showing fatigue in the numbers. The company is still compounding hard, actually. The only fatigue is in how much perfection the market is already pricing in.
If you want more recent data on this point, please see our latest semiconductor industry report.
Is AI still carrying TSMC these days?
TSMC is still being carried by AI, but it is no longer a simple “Nvidia orders more chips” story.
At the June 2026 shareholder meeting, C.C. Wei said it would take a long time before TSMC could meet customer demand. That line matters because CEOs usually try to sound in control. Here, the message was closer to: demand is bigger than the system can absorb. The Verge also picked up the same signal, noting that TSMC is trying not to become the bottleneck in the AI supply chain.
Broadcom gave us a better second-source signal in March 2026. A TrendForce report, citing Reuters, said Broadcom was seeing TSMC capacity become a real bottleneck in 2026. That is more interesting than another TSMC quote because it comes from a customer feeling the constraint directly. When Broadcom says capacity is no longer “unlimited,” it means the shortage is showing up in planning conversations, not just in earnings-call optimism.
Nvidia adds the third signal. Around Computex 2026, Jensen Huang’s Taiwan trip was widely read through the Vera Rubin ramp and advanced-packaging needs. Separately, earlier 2026 coverage had Huang warning that TSMC may need to “work very hard” and potentially more than double capacity over the next decade to match Nvidia’s demand. That is not a casual supplier comment. That is the largest AI chip company telling the world its own growth still runs through TSMC.
Everything considered together, AI is still TSMC’s growth engine today. The difference is that the AI signal has moved from “orders are strong” to “the whole AI hardware system is negotiating for TSMC access.”
As this chart shows, and as featured in our semiconductor industry deck, search interest in semiconductors has been rising steadily
Is TSMC’s 2nm actually happening now?
TSMC’s 2nm is no longer just roadmap language; it is becoming a real customer platform.
The most concrete recent signal came from AMD in May 2026. AMD announced that its 6th Gen EPYC “Venice” processor had entered production ramp on TSMC’s N2 process. That matters because this is not a low-volume demo chip or a phone-only node transition. It is a server CPU platform for AI and HPC workloads, with AMD talking about up to 256 cores and a big performance jump versus Turin.
There is also a customer-count signal. Industry reporting from late 2025 said TSMC had around 15 customers lined up for 2nm-class technology, including Apple, AMD, Intel, and MediaTek. The exact customer list is not all confirmed by TSMC, so we should not treat every name with the same confidence. But the pattern is clear: early 2nm interest is broader than Apple.
Apple is still important here. Several supply-chain reports say Apple is taking a very large share of early 2nm capacity for 2026 devices, while MacRumors reported in February 2026 that Apple may use the base N2 process rather than the enhanced N2P for upcoming A20 and M6 chips. That detail is useful because it shows customers are not blindly chasing the fanciest variant. They are choosing the node version that best matches volume, cost, timing, and yield.
So yes, TSMC’s 2nm is real now. The more interesting point for us is that 2nm is already being split by customer type: Apple wants high-volume consumer efficiency, AMD wants server performance, and others are lining up behind them.
Is TSMC still way ahead of Samsung and Intel?
TSMC is still far ahead commercially, even if Samsung and Intel are making enough progress to stay relevant.
TrendForce’s Q1 2026 foundry-share data is the simplest way to see the gap. TSMC held 72.3% of the foundry market, while Samsung had 6.5%. That is not a normal competitive lead but a market structure where one company has become the default manufacturing platform for the most important chip designers.
Samsung’s signal is not zero, though. Samsung has reportedly been working through 2nm yield improvements and trying to win back confidence from customers. Intel is also not dead in this race. In 2026, Intel’s 18A-P moved into risk production, and there are signs that large customers want Intel to exist as a second source, even if only to reduce dependence on TSMC.
The hard part, though, is convincing Apple, Nvidia, AMD, Broadcom, Qualcomm, MediaTek, and hyperscaler ASIC teams that yield, packaging, libraries, IP, delivery dates, and crisis management will all hold together. TSMC’s lead is commercial trust as much as transistor density.
So TSMC is still the center of gravity. Competitors may win specific sockets, but they are not yet changing where the industry instinctively goes when a chip absolutely has to work.
If you want more recent data on this point, please see our latest semiconductor industry report.
This chart, featured in our semiconductor industry deck, shows annual venture capital investment in semiconductor startups
Is CoWoS still the real bottleneck at TSMC?
CoWoS is still one of TSMC’s most valuable bottlenecks, and the recent packaging signals are more interesting than the wafer signals.
At its 2026 technology symposium, TSMC put advanced packaging right next to leading-edge nodes in its core message: 3DFabric, SoIC, InFO, CoWoS, and system-level integration. That alone says packaging is no longer a back-office step. It is now part of the product.
Then, in June 2026, Kevin Zhang pushed back hard on the idea that panel-level packaging would soon replace CoWoS for giant AI processors. The point was not that panel packaging is useless. TSMC is developing CoPoS and expects a pilot line around mid-2026, with commercial use more likely around 2029–2030. Rather, the point is that wafer-level CoWoS still has better interconnect density and better tooling today. The future may get bigger, but the present still belongs to CoWoS.
There is also a capacity signal. Several industry estimates now put TSMC’s CoWoS capacity on a path toward roughly 120,000–130,000 wafer starts per month by the end of 2026, up massively from late 2023 levels. Even if those estimates vary by source, the direction is obvious: TSMC is throwing capacity at the problem, and customers still want more.
Finally, Broadcom’s March 2026 bottleneck comments and Nvidia’s Vera Rubin supply-chain pressure both point to the same conclusion. The AI chip shortage has moved from “can we make the chip?” to “can we package the full system fast enough?”
At the end of the day, CoWoS is not just a constraint for TSMC but rather one of the reasons TSMC stays in control.
Is TSMC too dependent on Nvidia now?
TSMC is very exposed to Nvidia, but the company is not a one-customer AI story anymore.
Nvidia is still the loudest signal. Jensen Huang’s repeated Taiwan visits, public praise for TSMC, and warnings about capacity all reinforce the same point: Nvidia’s AI roadmap depends heavily on TSMC. Some industry estimates even suggest Nvidia holds the majority of CoWoS allocation. Those figures are estimates, not official disclosure, but they fit the behavior we see around capacity talks.
The second signal is Broadcom. Broadcom’s AI accelerator and networking business has become one of the clearest non-Nvidia demand signals for TSMC. When Broadcom flags TSMC capacity as a bottleneck, it means the custom silicon wave is also fighting for the same manufacturing base.
AMD is the third signal, especially with EPYC Venice ramping on TSMC N2. Apple adds another layer with early 2nm demand for iPhone and Mac chips. MediaTek and Qualcomm are also tied to the 2nm mobile transition, based on supply-chain reporting. None of these companies has Nvidia’s AI accelerator scale, but together they show that TSMC’s scarcity is multi-customer.
If you want more recent data on this point, please see our latest semiconductor industry report.
This chart, featured in our semiconductor industry deck, looks at TSMC’s strategy in semiconductors
Is Arizona finally becoming real for TSMC?
TSMC Arizona is becoming real, but it still looks more like strategic insurance than a replacement for Taiwan.
The big headline is already known: TSMC has described a U.S. plan that could reach $165 billion, including multiple fabs, packaging facilities, and R&D. The less obvious recent signal came in May 2026, when TSMC approved another $20 billion capital injection into TSMC Arizona. That is a practical board-level signal, not just policy language.
There is also an early profitability signal. Tom’s Hardware reported, citing Taiwan officials and financial data, that Fab 21 generated about $514 million in profit in its first full operational year. New fabs are usually brutal economically at first, so that number matters. It suggests the Arizona site is not purely symbolic.
But the friction is still real. The same Arizona coverage pointed to water, labor, energy, visa, supplier-localization, and environmental constraints. Those are not side issues. A fab is only as good as the ecosystem around it, and Taiwan’s ecosystem took decades to build.
So yes, Arizona is now a real asset for TSMC. But if someone asks whether TSMC is “moving to America,” the answer is still no. The company is adding a geopolitical safety layer while keeping the deepest manufacturing muscle in Taiwan.
Is TSMC running into a talent problem lately?
TSMC’s talent constraint is becoming one of the most under-discussed risks in the story.
In March 2026, TSMC said it planned to hire 8,000 workers this year. It also said new master’s-level engineers would receive an average annual salary of about NT$2.2 million. That is not just a hiring headline but rather TSMC putting a price on the next layer of capacity.
The second signal came from C.C. Wei himself. Recent coverage of his shareholder-meeting comments said talent had become one of Taiwan’s urgent semiconductor constraints. That matters because talent is slower to fix than capital equipment. You can order more EUV tools, but you cannot instantly create thousands of process engineers who understand how to ramp a leading-edge fab without hurting yield.
The third signal is geographic. Arizona, Japan, Germany, and Taiwan expansion all need people at the same time. That creates a managerial problem: the same company has to train new teams, transfer tacit knowledge, localize suppliers, and keep the Taiwan core from being stretched too thin.
This chart, featured in our semiconductor industry deck, shows annual funding in semiconductor startups
Is geopolitics helping TSMC or making life harder now?
Geopolitics is helping TSMC win projects, while making the company much harder to run.
The U.S. wants advanced manufacturing onshore, customers want less Taiwan concentration, and governments want semiconductor sovereignty. TSMC is the one company everyone can agree they need. That is why the Arizona plan matters so much: it turns TSMC into a geopolitical tool as well as a supplier.
But Taiwan has no incentive to give away the whole crown jewel. Taiwan News reported in late 2025 on the “N-2” logic, under which overseas production would stay about two generations behind Taiwan’s most advanced domestic process. Whether the rule is formalized exactly that way or enforced more flexibly, the strategic message is clear: Taiwan wants TSMC global, but not fully portable.
Europe and Japan show the same pattern from another angle. Germany’s Dresden project, with Bosch, Infineon, and NXP, is about automotive and industrial resilience. Japan’s Kumamoto expansion is also more about regional supply security than winning the absolute bleeding edge. These projects deepen TSMC’s global relevance, but they do not erase Taiwan’s central role.
So geopolitics is a tailwind for TSMC’s demand and subsidies. It is also a permanent execution tax.
If you want more recent data on this point, please see our latest semiconductor industry report.
Are power and water still real problems for TSMC today?
Power and water are still real problems, and they are becoming more important because AI demand keeps raising the ceiling.
The power signal is the more urgent one. In late 2025, the TSMC-led Taiwan Semiconductor Industry Association warned Taiwan’s government about a “power siege.” The reported numbers were uncomfortable: only 14.1% of fab power came from renewables in 2024, far below the RE100 path many major tech companies have committed to. That is a real gap when TSMC’s customers increasingly care about carbon footprints.
Ars Technica added another useful signal in May 2026: TSMC is leaning into long-term wind power agreements, including a 30-year power purchase agreement tied to the Hai Long offshore wind project. That tells us TSMC is not treating energy as a PR line: it is trying to secure physical power for future growth.
There are also efficiency efforts. Recent reporting said TSMC’s EUV energy-saving program cut peak EUV tool power consumption by 44% and could save 190 million kWh by 2030. That sounds large, but it has to be compared with TSMC’s total electricity use, which is measured in tens of billions of kWh annually. The efficiency work helps, but it does not remove the country-level power problem.
So the actual risk is that every extra point of AI demand now needs land, water, grid stability, clean power, and political permission.
This chart, featured in our semiconductor industry deck, compares the main business model options for fabless semiconductor companies
Is TSMC getting more pricing power now?
TSMC has more pricing power today, but it is being careful because it wants customers locked in for years.
The financials already show the pricing power. A 66.2% gross margin in Q1 2026, followed by a Q2 guide of 65.5% to 67.5%, is not what you see when a supplier is weak. It is what you see when the supplier has scarce capacity and customers have limited alternatives.
But C.C. Wei’s June 2026 comments add the more interesting layer. He said TSMC would avoid sudden price hikes even while demand remains hard to satisfy. That sounds almost modest, but it is actually strategic. If TSMC pushes too hard on price, customers have more reason to fund Intel, Samsung, internal silicon alternatives, or political pressure for second sourcing.
There is a nice contrast with memory makers too. Recent coverage noted Wei’s comments about RAM makers raising prices sharply during the AI boom. TSMC could probably push harder, but it seems to prefer long-term dependence over short-term extraction.
So yes, TSMC has pricing power.
Is there any founder or insider-style weak signal around TSMC lately?
There is no useful “founder tweeting every week” signal for TSMC, and that absence is itself important.
Some companies give you a live emotional feed through founder tweets, VC posts, or messy LinkedIn debates. TSMC is the opposite. Morris Chang is not running the company through social media. C.C. Wei’s most important signals come through shareholder meetings, earnings calls, customer interactions, symposiums, hiring plans, and government-linked announcements.
That makes TSMC harder to read in the “startup gossip” sense, but easier to read in the operational sense. The weak signals are not spicy tweets but rather Broadcom saying capacity is no longer unlimited, Nvidia flying senior attention into Taiwan, AMD moving a real server chip onto N2, TSMC hiring 8,000 people, Arizona getting another $20 billion, and Taiwan’s chip association warning about power.
So the insider signal is boring on the surface but strong underneath. TSMC behaves like a company that does not need to talk loudly because customers, governments, and competitors are already revealing how central it has become.
This chart, featured in our semiconductor industry deck, shows the revenue mix across customer segments in the semiconductor industry
So, how’s TSMC doing these days?
TSMC is doing very well right now. The company is growing around 30%, margins are still exceptional, 2nm is moving into real customer ramps, CoWoS remains one of the most valuable bottlenecks in AI, and competitors are still far behind commercially.
The more interesting conclusion is that TSMC’s product has changed. It is not just selling wafers anymore. It is selling access to the trusted manufacturing system behind AI, Apple silicon, custom ASICs, server CPUs, and national semiconductor policy.
The risk is also clear now. TSMC has to scale almost everything at once: 2nm, CoWoS, Arizona, Japan, Germany, Taiwan talent, clean power, and customer allocation. So if we take the temperature today, the company looks stronger than ever, but the story has moved from “can TSMC win demand?” to “can TSMC create enough TSMC?”
If you want more recent data on this point, please see our latest semiconductor industry report.
| Question | Short answer | Signals we checked |
|---|---|---|
| Is TSMC still growing fast now? | Yes. The growth is still showing up in recent revenue and margins. | May 2026 revenue +30.1% YoY; Jan-May revenue +30.0%; Q1 gross margin 66.2%; Q2 guide still very strong. |
| Is AI still carrying TSMC these days? | Yes. AI demand is now turning into system-wide capacity pressure. | C.C. Wei’s June 2026 shortage comments; Broadcom capacity warning; Nvidia Vera Rubin and long-term capacity pressure. |
| Is TSMC’s 2nm actually happening now? | Yes. N2 has moved from roadmap to customer production. | AMD EPYC Venice ramp on N2; reported 15 early N2 customers; Apple expected to use base N2 for A20/M6. |
| Is TSMC still ahead of Samsung and Intel? | Yes. The commercial gap is still huge. | TrendForce Q1 2026 foundry share; Samsung 2nm yield efforts; Intel 18A-P risk production still early. |
| Is CoWoS still the TSMC bottleneck? | Yes. Packaging is now a core part of the moat. | TSMC 2026 symposium focus; Kevin Zhang’s June 2026 CoWoS comments; CoWoS capacity expansion estimates; Nvidia and Broadcom pressure. |
| Is TSMC too dependent on Nvidia now? | Exposed, yes. One-customer story, no. | Nvidia CoWoS dependence; Broadcom custom AI chips; AMD N2 server ramp; Apple and mobile 2nm demand. |
| Is Arizona becoming real for TSMC? | Yes, but it is still a hedge. | $20B May 2026 capital injection; reported Fab 21 profit; $165B U.S. plan; water, labor, energy and supplier constraints. |
| Is TSMC short of people lately? | Yes. Talent is becoming a real scaling bottleneck. | 8,000-worker 2026 hiring plan; NT$2.2M new engineer pay; CEO comments on Taiwan talent constraints. |
| Is geopolitics helping TSMC now? | Yes commercially, but it makes execution harder. | U.S. onshoring push; Taiwan N-2 logic; Germany and Japan fabs focused on regional resilience. |
| Are power and water still risks? | Yes. The issue is now permission to scale. | TSIA “power siege” warning; renewable share gap; Hai Long wind PPA; EUV energy-saving program. |
| Does TSMC have more pricing power? | Yes. It has pricing power but is using it carefully. | Q1 2026 margin strength; Q2 margin guide; C.C. Wei avoiding sudden price hikes; contrast with RAM price increases. |
| Is there founder-style weak signal? | Not really. The useful signals are operational, not social. | No meaningful founder-tweet trail; customer comments, hiring, capex, supplier pressure and policy moves reveal more. |
This chart, featured in our semiconductor industry deck, shows how advanced foundry node manufacturing technology has evolved over time
OUR METHODOLOGY
The question behind this page is simple, but not obvious: how strong is TSMC right now? Instead of answering from market mood, stock momentum, or general AI enthusiasm, we broke the question into the main dimensions that actually shape TSMC’s current position: growth, margins, AI demand, 2nm, advanced packaging, competition, customer concentration, overseas expansion, talent, power, water, geopolitics, and pricing power.
For each dimension, we looked at recent signals first: company-reported financials, customer announcements, supplier and capacity comments, technology updates, hiring plans, government-linked developments, and credible industry data. We then aggregated the strongest signals, separated official evidence from directional reporting, and used that combined picture to reach a clearer judgment than any single metric would allow.
This page is independent editorial analysis. It should not be read as investment advice or as a recommendation to buy, sell, or hold any security.
Key sources used for this analysis include: TSMC monthly revenue, including May 2026 and year-to-date revenue, TSMC Q1 2026 quarterly results, TSMC Q1 2026 earnings release, TSMC Q1 2026 earnings presentation, TSMC 2026 Technology Symposium, TSMC 3DFabric page, TSMC CoWoS technology page, TSMC HPC 3DFabric and advanced packaging page, AMD’s announcement on EPYC “Venice” ramping production on TSMC 2nm, AMD’s TSMC N2 product silicon milestone release, TrendForce wafer foundry research, TrendForce Q1 2026 foundry datasheet page, TSMC Arizona official page, TSMC press release on expanding U.S. investment to $165B, TSMC 2024 Sustainability Report, TSMC EUV Dynamic Power Saving article, Northland Power on the Hai Long offshore wind CPPA with TSMC, Focus Taiwan / CNA on TSMC’s 8,000-worker 2026 hiring plan and NT$2.2M engineer salary signal, The Verge on C.C. Wei’s June 2026 AI-demand and bottleneck comments, TrendForce on Broadcom flagging TSMC capacity as a 2026 bottleneck, and SCMP on Jensen Huang urging TSMC to expand capacity amid AI chip demand.
In our semiconductor industry deck, we identify pain points entrepreneurs should prioritize
Related blog posts
- How strong is fundraising in the semiconductor industry right now?
- Which companies have raised the most funding in the semiconductor industry?
- Which companies are the most valued in the semiconductor industry?
- Is the semiconductor industry growing now?
