For twenty years, we and everyone we know in the semis industry toiled in relative obscurity. Our families did not really understand what we worked on, did not know what Moore’s law meant, had no idea what a nanometer was, or what a fab looked like. Then suddenly in the past two years, our industry found itself at the center of the global spotlight. There are 200,000 semis experts on Twitter now (many of whom may recently have moved onto other areas of instant expertise like anti-tank doctrine), and our families all have so many questions.
The title of this piece is by far the most commonly asked question we seem to get. Below are out thoughts on the state of the supply chain. This is largely anecdotal, in that we have not gone through and spread the financials of 100 semis companies, nor have we conducted 100’s of interviews. That being said, we have spoken to many people, and we have closely monitored the tech press both here and in Asia. So consider this a qualitative best guess.
To set the background – the pandemic clearly roiled the semis supply chain. China shut down in early 2020, and just as they started to get factories back on line in late 2020, the rest of the world was shutting down. This was then followed by a massive surge in demand. Starting a factory up is not a simple process, especially when half the work force disappears during the closure. For much of 2021 the industry had to play catch up. Everything was in short supply, with some components completely unavailable. This was hard to understand, in part because some of the lowest value, most commoditized inputs were the ones that briefly became the hardest to find (looking at you substrates). Put simply, the recovery was uneven, and remains a work in progress.
The next phase of the supply crunch was essentially an industry exercise in prioritization. Big customers got priority, with everyone else going on to some form of allocation – as in they got whatever was left over. This seems to remain the guiding principal for what is going on today.
For example, by mid 2021 Apple was largely able to procure almost all of its components. The other big chip companies – Qualcomm, Nvidia, AMD, etc. – are in a similar position. Management teams still talk about unsatisfied demand due to shortages, but that is starting to look a bit long in the tooth. The dog ate my components…
We can see this in the press. Xiaomi expects chip shortages to ease in the second half of this year. Qualcomm and Mediatek are expected to cut prices, which is normal for this stage of the 5G roll out. Fab utilization for GaAs power amplifiers (PAs) is falling. Chip companies generally are seeing pricing pressure. And everybody’s favorite horseman of the apocalypse – memory companies are worried about demand. Big customers are getting the product they need, and by big we mean companies that buy a lot of chips.
So then how do we explain all the opposing headlines that shortages will continue into the far distance? Here is one about foundries and tool makers seeing their orders “piling up”. In a similar vein, Nikkei reports that the big foundries warn of parts shortages delaying their expansion plans (h/t to Eric Jhonsa for that one). Or reports that the Volkswagen expects parts shortages to last until 2024. Or this report about car shortages killing demand.
We think there are two things going on here.
First, half the stories we found (and linked to) on this topic deal specifically with autos. The reality is that the auto industry is a small consumer of semis, especially compared to mobile phones, PCs and data center. The big customers are getting their parts, the small ones, even some very large companies, are not getting their parts. This is exacerbated by the fact that much of the semis for cars are produced on trailing-edge, older manufacturing processes. Those fabs have gone through a decade or two of under-investment. No one saw much point in adding old capacity when Moore’s Law was pushing ever onwards. The pandemic and the slowdown in Moore’s Law broke that math and created a shortage where no one ever expected to see one again. It is unclear to us when this gets sorted out, but we would be surprised if it really does last into 2024.
The other factor here is that many of the companies reporting continued shortages are the foundries. Those articles we linked above are about companies delaying fab expansion because they cannot get enough tools. Part of the issue here is that wafer fabrication equipment (WFE) is comprised of very complicated machines that take a long time to build (and can cost $100 million and up). There is just no way to speed up that production.
However, there is another force at work. The semiconductor industry is terrible at forecasting. The average chip design company has essentially no idea what their customers will demand three months from now. We know this first hand, we have built many of these forecast models over the years, and their accuracy is not something we would brag about. We actually wrote about this a long time ago, back in 2015 we compared the semis supply chain to a bull whip, with every segment on a cracking whip sloping in a different direction. The WFE companies are at the far end of the whip and thus have the least insight into end-demand. It now seems very likely that those companies are moving in one direction (more capacity) just as the rest of the industry is moving in the other (price cuts, slacking demand).
This is totally normal. We have seen this many, many times in the past. The semis industry is cyclical and will eventually reset. The big question this time around is who will be left holding the proverbial bag – with mountains of inventory and ensuing write-downs. It seems possible that this cycle may play out differently. The foundries, for one, used the supply shortage to lock down firm commitments from their customers – with Apple, Nvidia, AMD and the rest paying significant upfront payments. If the foundries can collect on those (and they largely have already) they are likely to continue their expansion plans, dampening the hit on the WFE companies (although they are not going to be totally immune). Our best guess is that the ones to get hardest hit by the someday-downturn will be smaller chip companies who got pushed and pulled around, stuck between their big foundry suppliers and their big customers. In many cases, those are the same companies that are today seeing the biggest backlog of orders piling up, orders that may not actually materialize down the road.