The Technology Letter

View Original

For ASML, AI increases everything

An interesting couple of notes appeared over the transom this week regarding Netherlands chip equipment titan ASML Holding, one of the TL20 stocks worth considering.

There was no immediate news event for these notes, just what appeared to be a spontaneous rush to endorse a good company’s stock.

The theme of the notes, you could say, was, Chip making is only getting harder, and that’s always good news for ASML.

In case you’re unfamiliar, ASML makes laser light sources for what’s known as photolithography, where a light is shown through a mask or screen, imprinting a pattern of circuits on the photosensitive layers of a semiconductor wafer such as silicon. That’s how circuits are carved into the wafer to become chips.

ASML makes equipment that projects the light from a laser light source and focuses it with various optics parts, and also moves the wafer around to get the right angle.

The company has sold the bulk of its machines, which are multi-million-dollar, as what are called “deep ultraviolet,” abbreviated as “DUV” by the company. Deep ultraviolet refers to the fact that the laser light source projects light at a wavelength of 193 nanometers, 193 billionths of a meter long, a wavelength shorter than visible light. Ultraviolet is, of course, present in the Sun’s rays that reach earth. The newer product from ASML shortens the wavelength of light dramatically, to just 13.5 nanometers, and is called “extreme ultraviolet,” abbreviated as “EUV” by the company. (Read all about ultraviolet at Wikipedia.)

So, why is it that chip making is getting harder and why is it good for ASML’s sales of these photolithography systems?

See this content in the original post

Jefferies & Co. chip equipment analyst Janardan Menon offers the view that a lot of it is about artificial intelligence. People are obsessed with OpenAI’s program called ChatGPT. “Since the advent of ChatGPT and other generative AI solutions, AI has become a major investor theme,” writes Menon.

This so-called generative AI stuff like ChatGPT is intense in use for chips, notes Menon. It takes tens of thousands of Nvidia’s GPU chips in a computer system to develop a program such as ChatGPT.

Those chips, notes Menon, are getting bigger and bigger as they get more complex.

“Alongside the rise in GPU and CPU volumes is the ongoing rise in GPU and CPU die sizes, as vendors like Nvidia and Intel increase their number of cores to increase performance,” notes Menon. For example, Intel’s latest Xeon server chip, “Sapphire Rapids,” which has to do a lot of the processing for the world’s queries to a system such as ChatGPT, is five times the size of the chip from 2017 it replaced, “Broadwell.”

And AI is being included as a function within many other, non-server chips, such as smartphone processors, which “is expected to drive an increase in the die size of smartphone and automotive processors as well.”

Now, as Taiwan Semiconductor Manufacturing, the world’s largest contract manufacturer of chips has said, those kinds of chips demand the most cutting-edge circuits. AI is one of the things that Taiwan Semi’s CEO CC Wei has told the Street is driving higher the most complicated circuits, things that require the EUV tools ASML makes.

Said Wei in October, Taiwan Semi’s most cutting-edge chips, those measuring three billionths of a meter, or three nanometer, at their critical dimensions, are in greater demand than the company can currently supply. Revenue from three-nanometer, said Wei, will be higher than was revenue from the prior cutting edge, five-nanometer, when that generation was introduced in 2020. Those “three nano” chips, as they’re called, are being used for high-performance computing, probably things such as Nvidia’s latest and greatest chips, for AI and such, and for smartphones, probably Apple’s next thing.

Menon offers a chart of how those most-advanced circuits have become a bigger and bigger part of Taiwan Semi’s business.

Menon notes that if every company has a big computer like the ten-thousand GPU machine, because every company wants a ChatGPT kind of function, then “The expected growth in the adoption of these AI solutions across enterprises and hyperscalers is, therefore, expected to result in a sharp increase in GPU, CPU, memory, and high-speed connectivity chip demand.”

As Menon sees it, “we expect AI to be the major driver of advanced logic capex [capital investment] investment growth.”

Menon cites Gartner for the obligatory market stat: “In May 2022 Gartner forecasted AI-related chip revenue to grow at 20% to 2026, a forecast which we believe could be revised upwards after the recent launch of multiple generative AI applications led by ChatGPT.”

The bottom line for ASML’s EUV systems is that “This combination of higher chip volume and bigger die sizes should translate into growing demand for leading-edge wafer capacity at the 3nm, 2nm, 1.4nm, and subsequent nodes,” writes Menon.

Menon predicts ASML’s revenue will rise by twenty-five percent per annum, compounded annually, though 2025.

The AI link here is not surprising to me. A little over two years ago, I wrote that “ASML is a leading indicator for AI.”

The angle I took in that article is that AI is good for ASML because it requires a lot of DRAM memory chips made by Micron Technology and Samsung Electronics and others. DRAM uses the older equipment, the DUV. The DUV stuff has a higher profit margin than EUV because older equipment is now far enough along a cost curve that it tends to be more efficient for ASML to produce. That means DUV sales can boost ASML’s profit margin over time.

Right now, DRAM is in a slump because of the general semiconductor cycle, but it will come roaring back because of memory-hungry programs such as ChatGPT. When it does, it’ll be a lift to ASML’s margins.

The other endorsement last week came from Krish Sankar of TD Cowan, who has ASML stock on his “best ideas” for European stocks list. “ASML is our top pick given the businesses' resiliency to the current WFE cycle,” writes Sankar, using the same word, “resilience,” of which ASML CEO Peter Wennink is fond of speaking.

Sankar is confident that with the “lead times,” the time a customer must wait to get an order for ASML’s EUV equipment filled, the company is set to “outgrow WFE near-term,” referring to total industry sales, known as “wafer fab equipment,” or WFE.

Investors, writes Sankar, underestimate how the profit margin for the EUV equipment is set to expand over time, though he doesn’t go into details on the matter.

After the company’s analyst day meeting this past November, writes Sankar, he came away convinced that “the company's long term roadmap is well articulated,” adding, “we see €25 in earnings power in CY25 and €40-€65 by 2030.” That compares to €14.14 in 2022.

The stock, moreover, deserves its premium, which tends to be 1.9 times to 2.5 times other chip-equipment makers’ stocks, writes Sankar. “We value ASML on a P/E basis assigning a 44x multiple to our CY23 EPS estimate of €17.18 to arrive at our €750 price target.”

ASML shares, at Friday’s close of $647.53, are up thirty-six percent since being picked in the inaugural TL20 back in mid-July, making it the third-best performing name behind Nvidia and Arista Networks. The stock is up eighteen percent this year.

One other quick thought on the company. Jefferies & Co.’s Menon offers a separate note concerning Nvidia’s announcements this past week of new products and technologies at its GCT conference, which I discussed on Tuesday.

I mentioned one of the less-appreciated items brought up by Nvidia CEO Jensen Huang was the company’s new software that would let Taiwan Semi and ASML improve the chip-making process. Menon explains that Nvidia’s software improves on two key algorithms used in chip-making, “Optical proximity correction (OPC) and inverse lithography (ILT).”

Writes Menon, the Nvidia software, called “cuLitho,” could be a boon to ASML. “We see this as a positive development for ASML and the chip industry, as the faster and potentially cheaper creation of complex ILT masks and OPC algorithms will make it easier for chip manufacturers to move to more advanced nodes.”

It should fuel more of those EUV machine sales for those more-advanced circuits, in other words.


Subscribe Now to The Technology letter

For $30 a month, you’ll get access to daily news in the world of tech companies and stocks, lengthy interviews with tech CEOs and CFOs, podcasts, and updates to The Technology Letter 20 group of stocks to consider — and future great features. Choose to pay the year in full and get 10% off.


See this form in the original post