Taiwan Semi beat revenue estimates by 2.2% & beat guidance by 3.4%. The bulk of the sequential growth was driven by higher revenue per wafer, mostly from mix shift toward pricier N3 and N2 wafers plus some price increases; wafer shipment growth was just 3.2% Q/Q and dipped below 10% Y/Y for the first time in a while. This is all tied to TSM being immensely capacity constrained. They’re selling everything they can make.

Advanced technology is now 79% of wafer revenue compared to 74% Y/Y. Their latest scaled 2N N2 technology soared to 8% of wafer revenue compared to 3% just a quarter ago and 0% Y/Y. The ramp looks to be going quite well.

By platform, high-performance compute (HPC) was again the star of the show. It rose 22% Q/Q to reach 68% of total revenue. Automotive was the other standout with 23% Q/Q growth, but that’s just 4% of total revenue. HPC carried the bulk of this outperformance. Smartphone growth was 10% Q/Q and represented 21% of total revenue.

For Q4, TSM revenue guidance beat estimates by 5.5%. 66% GPM guidance was a tad lighter than 66.3% margin estimates, but EBIT guidance was still 3.7% ahead thanks to stronger-than-expected OpEx leverage. For the full year, the company raised growth guidance from 40%+ to 44%, beating 42% Y/Y revenue growth estimates. Finally, it reiterated all previous CapEx commentary. It’ll come in around $62B for 2026. Furthermore, they remain confident in 2026-2028 CapEx being “significantly higher than the past 3 years.” As always, they spend more on CapEx when there’s more demand from customers. The heftier spend is a direct byproduct of expecting the demand backdrop to improve from here. And finally, revenue growth should outpace CapEx growth over the coming years “if they do their job correctly.”

TSM trades for 23x forward EPS. EPS is expected to grow by 64% this year and by 34% next year.

More on Margins & Efficiency:

As briefly mentioned above, GPM pressure is largely being driven by the successful scaling of their latest N2 platform. This is what always happens for them. TSM debuts a new technology, capacity utilization and economies of scale take time to build… and? They eventually do. Because the N2 ramp is so much more meaningful than previous technology platforms, the headwind is a bit more noticeable than it is for a typical manufacturing launch. On top of this current headwind, TSM continues to accept incremental margin dilution associated with building factories overseas. Its supply chain and economies of scale are superior in Taiwan vs. the USA. It’s not at all concerning to me for three reasons. First, this is an inevitable byproduct of TSM existing in Taiwan for a lot longer than they’ve been here. Second, they’re already making progress on closing the margin gap faster than they assumed. For context, the GPM headwind associated with this for 2026 is now set to be 2 points compared to 2-3 points previously. And finally, their margin profile is still best-in-class in the USA before the inevitable years of tech and utilization optimizations take place. This is an easy decision for them to make. They give customers a lot more supply with lower geopolitical risk and they’re showing clear signs of their USA business eventually being as profitable as Taiwan. Long way to go… but getting there and already leading the pack as they get there.

On operating margin, OpEx efficiency gains remain strong enough to more than offset any gross margin pressure. Fixed cost leverage is a wonderful thing and is turbocharged by the rapid compounding TSM is currently experiencing. Looking ahead, they do not think EBIT margin is anywhere near the long-term ceiling. At the same time, they’ll continue fixating on customer value and relationships over hiking prices as much as they possibly can. They will boost rates over time, but that will remain a slow and methodical process informed by inflation and total value creation.

AI Cycle & Demand:

All recent company signals point to the AI cycle remaining convincingly healthy. The explosion in consumer agents is providing another leg of massive compute demand while models continue to get more and more complex too. Additionally, through the popularization of open-source models and disruptive business strategies from Meta Muse (lots of free compute and monetize transactions and power users), AI is getting a lot cheaper. Not from a GPU perspective… but from a token perspective. That is fantastic for driving a needed broadening in overall AI demand beyond a few giant companies training frontier models. While the GenAI and chatbot era was historically good for business, this might be even more positive. More complex tasks… more use cases… cheaper access… It's a great setup for a lot more compute demand over the coming years. Furthermore, TSM doesn’t care about chip mix. They dominate across GPUs, custom accelerators and CPUs with margin profiles that are roughly similar across all three. As the world argues over which chip designer is best positioned, this manufacturer wins regardless.

For a strong, tangible signal of AI demand continuing to rock and accelerate, consider their evolving N3 capacity plans. TSM does not usually build node capacity beyond company targets. These targets are created through borderline-obsessive work with their customers, their customer’s customers and even through hands-on data center visits in the USA to ensure things are progressing on schedule. These visits have been positive thus far and point to the power situation for the planned U.S. facilities being in reasonably good shape. TSM is generally pretty good at knowing how much to build for a given platform. This time… not so much. And that’s a very good thing. It shows the end markets are even stronger than TSM thought they’d be, and is leading to them building a few more 3N N3 technology facilities to meet the moment. This is happening as they continue to shift even more 5N N5 capacity to 3N N3 processes, as they do “everything they can” to serve customers as the supply/demand mismatch remains large.

TSM leadership wouldn’t go as far as concretely updating their mid-to-high 50% AI accelerator compounded annual growth guide for 2024-2029, but it did say that outperformance was likely.

Footprint & Roadmap:

Construction for its 2nd Arizona facility is now done. They expect volume production to begin during Q3 2027, which is about a quarter before their previous forecast as the firm continues to execute well on their challenging roadmap. We shouldn’t overlook this. Fabs are extremely hard to build. The USA doesn’t have the labor or built-in supply chain that TSM enjoys in Taiwan. The company moving forward faster than expected with an encouraging power supply situation is impressive. Their N3 Japan facility under construction is still on track for scaled 2028 deliveries.

The N2 business is developing quite well, with expectations that capacity will double in 2027. Looking ahead, the A14 technology is still set for volume production starting in 2028. Customer interest so far is outpacing N2, offering yet another sign of this cycle rolling on.

Consumer Hardware Market:

Memory inflation continues to be a persistent headwind for the lower end of the consumer hardware market. The resilience at the upper end of the market has helped offset some but not all of this. They were intentionally prudent in modeling this segment for next quarter as this cost obstacle is not going away any time soon.

IDM & Packaging Competition:

An Integrated Device Manufacturer (IDM) like Intel designs and manufactures its own chips. Leadership was asked about Intel’s current push to bring more of its outsourced chip production (much of it made by TSM today) back in-house and whether or not that potential competitive headwind is baked into guidance. It is. They were careful to say they don’t treat Intel’s business any differently than customers who aren’t trying to directly compete, but they did add that potential item to the forecast.

In terms of packaging, TSM is the only company shipping 5x reticle size packages. Reticle size refers to the biggest area a chipmaking machine can print in one shot, which caps how big a single chip can be. A 5x-reticle package gets around that cap by placing several of these maximum-size chips, plus memory, on one shared base about five times that size and wiring them together so they work like one much larger chip. TSM is developing 9x technology as we write/read. Intel’s competing packaging tech (“EMIB-T”) is quickly maturing into a needed second source of packaging capabilities across the globe. At the same time, research firms like SemiAnalysis point to TSM being comfortably ahead in this arena. Intel has a lot more catching up to do and, if they stub their toes on advancing the roadmap, TSM leadership hinted at being happy to step in and fill the void.

EUV Partnership:

Extreme ultraviolet lithography (EUV) prints patterns onto a silicon wafer based on the pre-set patterns of a photomask. High-NA EUV is the next-generation of this machine that can print even smaller, more detailed circuit designs. A photomask works like a stencil that holds a chip’s design. With High-NA machines, today’s 6-inch stencils are too small to fit a large chip in one go, so the machine has to print it in two halves and join them. The new initiative aims to move the industry to 12-inch stencils to remove that extra step (a pilot line is targeted for 2031, with production around 2033). ASML, the maker of the EUV machines, is partnering with Taiwan Semi to push and nurture this transformation, as it would eliminate a significant amount of need to print large chips in two halves and could lift throughput by 40% (according to ASML). Good to see these two complementary titans working more closely together.

Power Bottlenecks?

While the world worries about energy availability, Taiwan Semi thinks they’re in a good spot. Power constraints “have not reduced customer forecasts” and have not materially slowed down TSM’s business to date. That could eventually become a bottleneck for this business model like it has for so many others (seems intuitive), but it hasn’t yet.