Earnings: TSMC, ASML, AEHR, Telecom
It's probably peak lithography for ASML. TSMC flexes pricing power, and AEHR.
I don’t think it’s been any secret that I’ve been a peak litho bear for quite some time. I think we are finally seeing reality hit the company and the stock. Let’s discuss earnings, and then I’ll talk about peak lithography and TSMC.
ASML accidentally leaked earnings this week, and the leak wasn’t good. They beat Q3, but the issue was the outlook for next year and orders coming way below expectations. It’s funny because Q4 revenue is expected to accelerate, but no one is paying attention to that.
ASML Holding reports Q3 net income €2.08B vs consensus €1.91B
Reports Q3:
Revenue €7.47B vs consensus €7.15B and guidance €6.7B-€7.3B
Guidance 2024
ASML expects Q4 2024 total net sales between €8.8B and €9.2B
Guidance 2025
ASML expects 2025 total net sales to be between €30B and €35B vs consensus €35.8B
Memory was half of the measly 2.6 billion-order book and is likely driven more by DUV for HBM than logic High NA.
China still accounts for the majority of revenue. The problem is that it’s about to become the minority of revenue. Shares have dropped meaningfully on this result, about ~20% since the results came out. The 20 percent 2-day drop is maybe the worst return in the last ten years.
So, let’s talk about what went wrong because ASML is almost in a perfect storm to have caused this drop. To summarize, ASML guided down revenue due to leading-edge logic orders from Samsung and Intel. Meanwhile, Chinese revenue, the other side of the business, is about to normalize to 20% of its total revenue in 2025. That’s compared to 45% of sales last year. Both sides of the business are faltering.
Leading Edge Issues
When the news leaked, the speculation was that 18A’s foundry book was dead. Reality seems more mixed, and Reuters reported that Samsung in Taylor, Texas, is starting to see delays. Because Samsung is not getting any customers, they are holding off orders, and if the shell isn’t finished, you can’t deliver tools.
Meanwhile, the noise (and probably truth) around 18A is similar. Broadcom was happy to say bad things about the 18A process, and the 20A node was canceled to “move forward” 18A. But an alternative reading is that there isn’t enough demand on the internal node, so let’s kick it to 18A. Pretty much 18A’s success is in the fog of war. ASML’s comment on “competitive dynamics on logic” creates more fog.
I think we also indeed mentioned as well some competitive dynamic on Logic. And this also contributed to some of the pushout. So I think those two things are really the two dynamics that have come up in the last few months to a level where our customers basically started to really make, I would say, a decision that were in line with their expectations, both on the total market but also maybe, in some cases, on the share they may end up having in some of the Logic market.
I think that this is a bit of a hardball at TSMC. Samsung and Intel have ordered High NA machines, but I believe TSMC is holding out because they are trying to negotiate harder. They are the only customer who can convert the orders of High NA machines or Low NA machines into revenue, so they have leverage. I believe they will adopt it eventually, but only after price breaks.
This is the same company that complained incessantly about Arizona’s progress, and then, after the CHIPs Act approval, the fab magically got back on track. TSMC plays hardball in public, which is what’s happening today.
Let’s move on to the call.
ASML thinks that ex-AI demand is sluggish.
The relatively low order intake is a reflection of the slow recovery in the traditional end markets as customers remained cautious in the current environment.
So we are still quite optimistic about AI. I think today without AI, the market will be very sad, if you ask me. But for the rest, I think our customer continues to confirm that when it comes to mobile, when it comes to PC, when it comes to automotive, the recovery is not what I think everyone had wished for. And that affect, I would say, a large part of our customer on all segments and application.
The initial leak made it seem like it was all leading edge, but the big news was that China would revert to 20% of its revenue. This is in anticipation of further export restrictions and the understanding that the Chinese WFE gorge can’t continue forever.
We also now expect our China sales to be around 20% of our total revenue next year, trending back towards our historical China percentage and in line with its share of the backlog.
Meanwhile, non-China DUV will ramp up with new memory and logic capacity. 2nm ramps will require a lot of EUV but is going to grow at the same rate as DUV tools.
I mean with the China business going down, as we've indicated, that means that the non-China part of the deep UV business will go up. And the main reason is that if you look at the more advanced nodes and if you do the math on the growth that you might anticipate in the EUV business for next year, you will see that there is quite some growth there.
And I think what you're going to see is that the non-China part of the deep UV business is sort of following that trend. So the attach rate, if you want to call it like that, of the deep UV business to the EUV business will be kind of similar. So the growth rate you will see on EUV, you will sort of see back also in the growth rate for the non-China part of the deep UV business.
Another way to interrupt this statement is that EUV intensity is peaked if they think the DUV attach rate is growing as fast as EUV. This has been a long time coming, and Applied Materials puts it well in this slide - 3nm is the apex of lithography market share gains. This is not good for ASML and the EUV story.
And how are we that surprised? Most innovations have been explicitly focused on reducing the EUV layer count. I’ve said this repeatedly, but you can argue that Sculpta, reduction of EUV layers at TSMC, and even BSPDN are about removing EUV layers. When the entire industry plans to use less EUV, lithography is peaking.
The last decade has been the decade of Lithography, and I can almost assure you that Lithography spending will grow slower in the next decade than WFE. And let’s not forget that in the decade before EUV, lithography was only 15-20% of the expenditure. I believe that lithography will revert to the mid-20s of WFE.
I think it’s time for investors to accept the decade of lithography is over. The premium ASML has slowly gained for itself for the last decade should end as the decade of litho ends. And the recent price move has been a collapse of that premium. I believe it should trade in line with other semi-cap companies, as I’m sure that Litho will not outgrow WFE but will be valued for it’s strong monopoly.
It’s a monopoly and a good company, but so are all semicap companies. From a financial returns perspective, KLAC is much better on an ROE basis, and most companies are more capital-light than ASML. Lam and Applied have lowered the share count faster. Lowered growth but high quality should demand a multiple in line with competitors. ASML’s premium should collapse.
But amid China revenue going down 30% year on year and leading-edge logic pushouts on EUV, things are not as bad as they seem. ASML guided for a 17% revenue growth year and already has most of the year booked in the backlog. Meanwhile, WFE should grow faster than litho alone in the future.
From the headlines, you’d think that the world is over, yet most of the things that should hurt them are happening. China is going to 20% revenue, and leading-edge logic is collapsing to just TSMC, but they are still growing 17% next year!
The rational fear is that the order book implies a much worse 2026, but that can change quickly if AI and traditional markets grow. To hit the high end of next year’s forecast (35 billion), they need $3.8 billion in bookings. I think that the fear of backlogs is a genuine concern.
However, if you step back and take a longer-term view, ASML’s issues are not based on technology or growth but on expectations (30%+ growth) and multiples (50% premium to the group). It’s been a good few years for semicap, and I think if you lower return expectations, the entire group is acceptable.
The premium collapse should happen at ASML. I think it will continue to collapse in line with other semi-cap peers. I believe it’s an "average” time to buy semicaps, and even if derates happen, the stocks are acceptable. They are not a screaming buy nor an obvious short, and I appreciate the headline volatility.
I believe there could be another puke on export controls, but the risk-reward is fine at around a 30% drawdown. Multiples on this drawdown are higher than the previous corrections, so that’s the flip side.
If you’re a long-term investor, it’ll probably be okay - but landmines and fear are the game's name for now. Export controls could lead to another leg down, but Semicap has been 20%+ TSR CAGRs for a long time, and if it drops again, I think this could be a solid time to pick up semicap for the long term.
These are good businesses and will be just fine. Speaking of which, let’s turn to TSMC, the biggest spender on litho and on WFE.
Taiwan Semiconductor Manufacturing (TSMC)
Sometimes, it’s tiring to write about how good TSMC is and how mediocre the rest of the industry is. As Intel’s order book for 18A becomes more uncertain and the Samsung foundry continues to implode, TSMC is the company that cannot stop winning.
This quarter’s form of winning came in the form of a 57% gross margin. Everyone expected the revenue to beat because of monthly sales, but gross margins were much hotter than expected. This is almost an all-time high in margins, and PC, consumer, and smartphones have not even recovered. Let’s review the results.
Taiwan Semiconductor reports Q3 EPS NT$12.54 vs FactSet NT$11.49
Reports Q3:
Revenue NT$759.69B ($23.50B) vs FactSet NT$749.27B and prior guidance $22.4-23.2B
Operating income NT$360.77B vs FactSet NT$330.18B
Gross margin 57.8% vs FactSet 54.7% and prior guidance 53.5-55.5%
Operating margin (OPM) 47.5% vs FactSet 43.9% and prior guidance 42.5-44.5%
Let’s take the gross margin result and see what that implies for utilization.
Gross margin increased by 4.6 percentage points sequentially to 57.8%, mainly reflecting a higher capacity utilization rate and cost improvement efforts.
A good rule of thumb is about 40-50 bps of gross margin for each 1% in utilization. That implies that utilization ripped up 10% this quarter. While lagging edge utilization is weak, I believe that the leading edge is now 90%+ utilization.
If you look at this by platform, it’s evident that it is driven by HPC revenue. And HPC recently has been a proxy for Nvidia revenue and AMD CPUs.
Of course, one of the call's questions was whether AI is sustainable. TSMC’s response was reassuring, especially from the conservative Taiwanese company.
And why I say it's real? Because we have our real experience. We have using the AI and machine learning in our fab and R&D operations. By using AI, we are able to create more value by driving greater productivity, efficiency, speed, qualities. And think about it, let me use, 1% productivity gain, that was almost equal to about TWD 1 billion to TSMC. And this is a tangible ROI benefit.
And I believe we cannot be the only one company that have benefited from this AI application. So I believe a lot of companies right now are using AI and -- for their own improving productivity, efficiency and everything. So I think it's real.
Because of the demand for AI and general semiconductors, TSMC believes the market will grow meaningfully over the years. This is “just beginning.” TSMC is not one to chase hype and is more than happy to call Sam Altman a podcasting bro. But the demand will continue for a while.
The demand is real and I believe it's just the beginning of this demand, all right? So one of my key customers said, the demand right now is insane, that it's just the beginning. It's [ a form of scientific ] to be engineering, okay? And it will continue for many years.
AI’s growth is TSMC's opportunity. Server AI revenue will triple this year, a bit faster than Nvidia's data center revenue from FY24 to FY25. A mix of volume and pricing increases drives this faster than Nvidia's growth. AI is now in the mid-teens of revenue, and as it grows faster than the rest of revenue, it should contribute up to 30% in the future (by my estimates).
Because of this growth, they also raised their total revenue for the year to grow 30%.
We now forecast the revenue contribution from server AI processors to more than triple this year and account for mid-teens percentage of our total revenue in 2024. Supported by our technology leadership and broader customer base, we are well positioned to capture the industry's growth opportunities. We now forecast our full year revenue to increase by close to 30% in U.S. dollar terms.
That brings me to my next point—spending on capex and future demand for 2nm.
3nm has grown faster than 5nm and faster than 7nm. Part of this is ASP, but leading edge growth outperforms each previous node and the lack of foundry competition matters. So when TSMC was asked about progress on their 2nm node, the answer was clear.
Actually, we have many, many customers interested in the 2-nanometer. And today, with the activity with TSMC, we actually see more demand than we ever dreamed about it as compared with N3. So we are to prepare more capacity in N2 than in N3.
And following by A16, again, A16 is very, very attractive for the AI servers' chips. And so actually, the demand is also very high. So we are working very hard to prepare both 2-nanometer, A16's capacity.
This brings me to my next point. Capital intensity is increasing in leading-edge nodes, and capex will accelerate even for the rest of the year. If you look closely at TSMC’s capex guidance, it implies massive quarter-over-quarter acceleration for leading-edge capex.
Next, let me talk about our 2024 CapEx. Every year, our CapEx is spent in anticipation of the growth that will follow the future years. And our CapEx and capacity planning is always based on the long-term market demand profile. As the strong structural AI-related demand continues, we continue to invest to support our customers' growth.
We now expect our 2024 CapEx to be slightly higher than USD 30 billion. Between 70% and 80% of the capital budget will be allocated for advanced process technologies. About 10% to 20% will be spent for specialty technologies. And about 10% will be spent for advanced packaging, testing, mask making and others.
But if you do the math, they have only spent ~18.4 billion years to date, implying almost 100% quarter-over-quarter capex growth. I believe this is to support the 2nm process and the strong demand we are seeing at the leading edge driven by AI. And while most are freaking out about China demand, the leading edge is fully utilized, and TSMC will likely spend more than 40 billion dollars next year on Capex to support GAA intensity and higher production.
Now as C. C. said, next year looks to be healthier. So it is very likely that our CapEx next year will be higher than this year. And we will provide you more updates in January conference.
Despite the fear of Chinese WFE demand dropping, I believe the leading edge will be enough. They tell you they will invest in the capital-intensive 2nm node at higher capacity additions than 3nm. WFE will be up 10% or greater next year, even with the implosion of Chinese demand. Don’t forget that HBM is capital-intensive, too!
Right now, we are at the trough of ex-US WFE, and while China will go down by 30-50%, people are seriously discounting the rest of the world’s WFE spending. The number will be lighter than the previously expected mid-teens growth, but WFE should grow by double the digits. I understand the fear, but AI demand will require new wafers.
TSMC continues to invest heavily because the future is probably just as bright as the past. I believe that revenue growth from AI will matter and will become a larger and larger part of their total wafers with much higher margins. Their negotiating leverage is improving, and Intel is not exactly a competitive foundry. It’s hard to imagine a real risk other than China aggressively expanding outward. Pretty much TSMC is the only game in town. And what does a company that’s a virtual monopoly trade at?
The company trades at 22x forward earnings and can likely grow earnings north of 20% for a long time. There are no real threats on the horizon (well, except the one across the sea), and I think gross margins will likely see new cycle highs. TSMC’s position is better than ever, and it’s cheap as a stock.
Even after the almost 100% year-to-date run, shares look well-priced for future outperformance. Most of that was a re-rating, but I don’t think shares should de-rate here. If they can grow EPS by 30% forward, that should be close to the return expectation.
Meanwhile, the rest of their business, which is not AI, is stabilizing. I want to discuss that next behind the paywall. This contradicts what ASML is saying, and I think TSMC likely has a better view of demand than ASML.
Overall semiconductor demand, except the AI, I think is everything stabilized and start to improve.
Please read my thoughts below for more information about Telecom and AEHR.