No, but it’s specifically cutting-edge chips that are being sought here. Intel doesn’t manufacture bulk chips, neither for itself nor for Foundry customers. These new Intel plants are consistently geared toward future process nodes.
Of course, a large part of the factory buildings are reusable for decades if maintained, but the production lines themselves require renewal quite frequently if the goal is for the plant to produce state-of-the-art chips. The kind that are needed to reduce the world’s total dependence on Taiwan.
This, in turn, will only happen if the plant can produce chips profitably. It’s great that subsidies are available for putting up the structures, but in the long run, these are only economically sound investments if chips can be manufactured competitively for decades. For some reason, this previously proved difficult, and almost all manufacturing began shifting toward the China-Taiwan-Korea axis…
Since I’ve been looking for cigar butts and cash cows, I recently had some small change tied up in Texas Instruments and Marvell Technology. I then went and sold them. Might have been a mistake.
But I wonder how bright the future is for these not-so-leading-edge companies? Or for Infineon and STMicroelectronics?
My working hypothesis is that there are enough devices designed long ago that people are reluctant to redesign due to costs, meaning these companies would have relatively steady markets that even the Chinese wouldn’t find particularly worth bothering with, as the growth is elsewhere.
They grind out money on thin margins, fluctuating slightly according to chip sector cycles. The problem with bulk stuff is that the margins aren’t anything special. Dividend machines without the hockey sticks.
The AI boom is predicted to be a massive demand driver for the coming years, and chip companies are also racing to invest heavily to meet this demand. Additionally, the historical trend where chips always get faster while becoming cheaper is shifting. Chips are still getting faster, but at the same time, they are getting more expensive because the latest technology is so complex and costly to manufacture.
This effect partially explains why the prices of professional graphics cards and flagship smartphones are skyrocketing.
The Netherlands had already previously banned the export of EUV (extreme ultraviolet) equipment to China, but now restrictions are also coming for the export of DUV (deep ultraviolet) equipment, with China’s pride SMIC among those affected. By the way, the Reuters article speaks of “ultraviolent” equipment. I wouldn’t want to get more closely acquainted with those.
As an investor and just an ordinary citizen, I receive the tightening restrictions around China with somewhat mixed feelings. I won’t start unpacking all my thoughts on the matter in this context. Overall, I consider it a healthy development that export restrictions are being introduced and especially that high-tech manufacturing is shifting massively—at least formally—to democratic countries. I hope that in the longer term, this will also improve the competitiveness of some globally operating Finnish companies against unfair Chinese competition. Wearing only an investor’s hat, one can also see negative aspects in the development, and as far as stock prices are concerned—ASML’s share price will hardly be thankful for the new restrictions, and the situation will surely be reflected more broadly in the stock prices of semiconductor manufacturers.
Reuters →
The planned U.S. rule, which sources said may be published by late July, will require licenses to export equipment to about a half dozen Chinese facilities, including a fab operated by SMIC, China’s largest chipmaker, the person familiar with the U.S. plans, said. Licenses to ship the equipment to those facilities will likely be denied, the person said.
It is reported that the Netherlands will announce further restrictions on ASML and other companies on Friday, requiring them to It is necessary to obtain a license before exporting deep ultraviolet light exposure machine (DUV) to China. It is estimated that models such as TWINSCAN NXT:2000i of Esmer will be affected, and its older models such as TWINSCAN NXT:1980Di may also be affected. Prohibited for use in specific Chinese fabs.
As I understand it, the Chinese are about 4-5 years behind the West in these machines and have decided that they must have their own production. In a 10-15 year timeframe, it could still turn out that China completely detaches from Western technology and creates its own ecosystem, which is then sold throughout the developing world.
This could then result in more of a hit for ASML. Signed, “I used to own ASML, but the valuation multiples became too steep for my taste.”
There is one problem with that comparison – it compares the A100, which is two-year-old technology. NVIDIA sells a newer H100-based accelerator these days. For some reason, it’s not being compared to that…
Sure, you can crunch AI with MI250 cards as well, especially if better ones aren’t available anywhere, but AMD also needs the new MI300 series cards on the market (they’ve been launched, but deliveries seem to still be going only to major customers).
Related to the same topic — and also a similar take from this investor — though from a slightly different angle.
The guy makes a good point, as AMD is ahead of its competitors in chiplets. On the other hand, both Nvidia and Intel are already following suit with this technology, but AMD has been producing chiplet solutions for almost a decade. As for Nvidia, it might be difficult for them to move quickly from their overflowing coffers to cannibalizing their own business.
I ended up selling my Nvidia shares. The position wasn’t very large. The reason for selling wasn’t the emerging competition. I simply didn’t have the nerve for these valuations. I’ve already missed good selling opportunities in semiconductors a couple of times, and I made my last two purchases way too early. I made sure that this time I wouldn’t be left with regrets. There’s nothing wrong with the company, but as the AI fever subsides, the market might cool down a bit.
$AMD is set up to take a lot of market share from $NVDA for 2 reasons:
Much of of $NVDA ´s moat stems from CUDA, which enables developers to interact with $NVDA GPUs seamlessly when coding. However, Pytorch has become the dominant framework for deep learning (AI) and now, $AMD GPUs work out of the box on Pytorch.
This opens the veil for $AMD to disrupt $NVDA ´s software moat and levels the playing field.
$NVDA ´s chips are monolithic, while $AMD ´s chips are chiplet based. Chiplets have enabled $AMD to disrupt $INTC by yielding high performing but cheap chips.
This is because when you make a chip out of chiplets, if one of the components goes wrong you don´t have to throw the entire thing away. Hence, yields are much higher than in monolithic architectures.
I believe the same is going to happen on the GPU side.
Chiplets have lower margins and so pivoting to this would hurt $NVDA ´s financials: a clear case of the Innovator´s Dilemma. $NVDA has a fantastic business and will hardly find the incentives to disrupt itself moving towards chiplets.
A plausible scenario, but personally, I don’t believe it will happen in the very near future (i.e., the next year or two). Instead, AMD will likely gain market share simply because NVIDIA is selling everything they can produce, and since that still doesn’t meet demand, every remotely competitive device from AMD will also sell—even if at a slightly lower margin and only to customers willing to put in a bit of extra effort to get more hardware up and running.
Is it possible that AMD could suddenly pull ahead of NVIDIA’s products with chiplet GPUs, especially on the professional side? Perhaps, but likely not for the next generation at least. It should also be noted that rumors suggest NVIDIA is also working on a chiplet-based product, though it’s still unclear when it might be commercialized. Just waiting for more solid info on the next-gen AI/compute cards…
The global topic of the day is China’s export restrictions on two important (semi)metals, gallium and germanium, which are used in semiconductor components, solar panels, LEDs, infrared and fiber optics, etc. In addition to these, China announced it is imposing restrictions on more than 30 other related metals and materials.
This is not the first time China has used raw materials as a means of pressure, and it comes as no surprise to anyone—except in Germany, where the country’s critical infrastructure is preferably built on equipment supplied by authoritarian countries because, for some reason, they always get a dirt-cheap deal, or “Handeln”. Many other countries understand that if China has a means of leverage, it will certainly use it to achieve its own goals.
This time, it’s about the next chapter of the US-China tussle. China is responding to US export restrictions in its own way, which it considers equivalent to the restrictions imposed on itself. China has leverage particularly regarding gallium (image from the WSJ article) →
The consequence of this is, of course, rising costs, which is unfortunately a bad thing right now—in many ways and for many companies.
WSJ:
“This measure will have an immediate ripple effect on the semiconductor industry, especially with regards to high-performance chips,” said Alastair Neill, board member of the Critical Mineral Institute who has nearly 30 years of experience with China’s metals industry.
Below is CGPT’s “management summary” of the good WSJ article.
China has announced export restrictions on gallium, germanium, and other related materials essential for manufacturing semiconductors, missile systems, and solar cells.
The move is seen as a show of strength ahead of economic talks between China and the United States.
The restrictions, effective August 1, aim to protect national security and require future export applications to be reviewed by the country’s government.
The United States has also imposed export restrictions on advanced semiconductor equipment and pressured its allies to do the same.
Both countries consider these measures necessary to protect national security interests.
Restrictions on gallium and germanium could significantly impact the semiconductor industry, particularly high-performance chips.
China’s efforts to develop its semiconductor industry have faced obstacles as the United States has imposed restrictions and warned of military implications.
According to the U.S. Geological Survey, gallium and germanium are categorized as critical minerals due to their importance to the U.S. economy and national security.
Gallium is used in semiconductor manufacturing, and gallium arsenide is widely used in high-performance chips.
Germanium is used in fiber-optic systems and solar cells.
Export restrictions on these commodities resemble previous restrictions on rare earth metals, which China also primarily produces.
China’s export restrictions allow for targeted actions against individual companies and sectors with geopolitical considerations in mind.
The move could give China leverage in future discussions with the United States regarding export restrictions.
Restrictions on gallium and germanium are seen as a demonstration of China’s control over supply chains and its strength in high-tech industries.
If true, it means that Intel’s troubles with manufacturing processes continue.
This was supposed to be the first “New Intel” processor generation manufactured on a state-of-the-art node for laptops and desktops, coming in late 2024. Finally, Intel was competitively taking on TSMC on comparable terms.
Except that the goalposts have apparently moved again, and plan-B is to have the chip manufactured by TSMC.
Previously, the intention was for the Compute tile to be Intel 20A, the GPU to be TSMC 3nm, and additionally, there would be other tiles on Intel’s older nodes, but now that Intel 20A has apparently disappeared from internal roadmaps for this product.
The change also comes quite late if this is intended to be in stores by Q4 2024, so it increases the risk that the entire product will be delayed. However, we cannot know at what stage this was decided; we only know that the information has now leaked out unofficially. If that was already decided 6 months ago, the situation is not as critical.
And the year when Intel might be competitive in manufacturing with TSMC has moved again into the indefinite future. 2025? 2026?
I wonder how TSMC will have enough 3nm manufacturing capacity if NVIDIA, Intel, and AMD are all competing for it in 2024…
This would mean that Arrow Lake would be entirely manufactured by TSMC (Intel 3 TSMC Compute / TSMC 3nm GPU / TSMC 3nm IO & SOC) and not just 2/3. This 2/3 solution had previously been openly questioned as well. The article, by the way, mentions that there is no information yet on whether this would affect just one segment or all of them.
Why do you think that this rumor, if it comes true, would have anything to do with overall competitive manufacturing and its postponement?
If 20A were a viable process, why would this product be moved entirely to TSMC?
Okay, in theory, there could be other reasons why this is the solution, but this was supposed to be the first end product made using the 20A process, which suggests that the commercial use of said process is being postponed or has been canceled entirely. I believe it was 2025 when products using the 18A process were already supposed to arrive…
We are still relying on rumors for now, but these have a nasty tendency to be confirmed later.
Today’s vague Intel rumors: The Battlemage architecture, which was originally supposed to arrive at the end of 2023 (Intel’s new GPU generation), has been pushed back to at least the second half of 2024. As a side effect, the iGPU in the Meteor Lake laptop processors coming at the end of this year will not be based on the Battlemage architecture, but will instead use a slightly refined version of the current Alchemist-generation graphics cores.
In other words, GPU design remains difficult, and this also fuels persistent rumors that Intel might be considering giving up on the dGPU front. Then again, looking at the AI hype train, this would seem very short-sighted, but…
This isn’t necessarily related to potential difficulties on the manufacturing side, as Battlemage was originally announced to use TSMC’s process—4nm for dGPU cards and 3nm for the Meteor Lake GPU tile.
How reliable do you consider those rumors at this stage? You mentioned earlier that regarding Intel, they have tended to turn out to be true. But how much of this is the “once a failure, always a failure” mentality, or are the rumors based on leaked information from inside the company? The company certainly doesn’t benefit from these kinds of rumors starting to circulate.
[quote=“Lexus, post:1272, topic:2300, full:true”]But how much of this is that “once a failure, always a failure” mentality?
[/quote]
I would say that it is very much about exactly this. Market skepticism will not fade until the company A) beats consensus for several consecutive quarters, B) gets its most advanced process nodes (viivaleveys) in order, C) separates the manufacturing and design departments from each other at the corporate level.
Personally, I do believe in and trust Intel, because the company now also has the support of Washington behind it. But Intel’s stock is unlikely to be a “To the moon” stock comparable to Nvidia in the future either, but stable and growing in the long term nonetheless. In the AI era, Intel cannot really afford to lose share in the GPU chip sector either. Let’s wait and see what Battlemage looks like.
