Hacker News Reader

> I know I might sound like a conspiracy theorist, but there is either some cartel deal between huge companies (intel, amd, nvidia, qualcomm,...), to keep prices artificially high and blame parts shortage, or they're all incompetent.

So here's what actually happened. Car companies cancelled their orders because there was this thing called COVID. Those slots on fabs got sold to other people in the entertainment industry because at the time it was believed people would just stay at home and the entertainment sector would need the extra fabs.

In a sudden twist of what could be called "irony", it turns out people are scared of getting on public transit during and after covid so everyone who relied on that or uber/lyft started buying cars so demand for cars went up. Car companies tried buying their slots back kicking off a giant demand hike which basically had people buying fab slots for far more than they are worth... the issue is that those ICs aren't like a graphics card.

If you don't have 5nm chips, well that's fine, you can't produce a graphics card per IC, for a car, there could be dozens of different chips. If your 30nm chips are out of stock, it doesn't matter how many 105nm chips you have in stock, you can't sell half of a car.

The photolithography machines that you need in order to pattern silicon for etching features smaller than ~20 nm or so are made by one company -- ASML. One machine is about the size of a house, costs >$100 million, and their order book is full for at least the next two years.

A semiconductor fab typically needs several of these machines (maybe only one or two of the best machines for the smallest layers, but more of the coarser machines for the less dense layers) in order to produce chips competitively.

Me think you grossly under-estimate the complexity of creating a facility that processes basic inputs, a few chemicals, and lots of optics/lights to make state of the art nanometer scale 3D features at enormous volume and very low defect rates.

World cup in korea was in 2002... between then and now, many generations (sizes) of semicondctors have been 'invented' (..the method to produce them) and replaced with smaller ones.

https://en.wikipedia.org/wiki/Semiconductor_device_fabricati...

There is a table on the right, and there are rougly 2-3 years between 'generations' - so 2-3 years between starting a new production line of top-of-the-line flagship (eg.) cpus, to making a new production line for even newer, "smaller" (transitors) cpus, and the old line making other, cheaper ones.

I'm not saying that Johnny Sixpack can do it in his garage, but companies like Intel, AMD, samsung, etc. surely could do it in that timeframe, because they're already doing it now.

I think the problem may be best addressed by spinning up ASML alternatives and other critical singular fail points/capabilities. Some things like this do rise to a national/global security threat.

Fabs are multi-year projects, and take quite a while to run a profit (since you first have to invest 10s of billions into building one). So "will be able to sell whatever" right now could still mean that the overall thing ends not very profitable once the competition recovers from their outages and finishes their expansion plans. And similarly, the supply chains for the equipment you need to actually build a fab don't magically scale, and are wary of doing so - because again, just because everyone wants their high-end fab equipment now doesn't mean they will in a few years, and they've been bitten by market contractions in the past, so they have little motivation to scale extremely and then struggle because they've satisfied demand. Their order books are basically full already, so you can't just go and equip your shiny new fab quickly. Even if they now wanted to expand more than they expected, them tooling up for faster production would take a while.

In the end, making chips just isn't that high on the overall list of things to be able to opt.out of normal market forces. If it looks like customers will pay enough to make rapid expansion and the changes necessary to enable such rapid expansion worth it, it'll happen.

But demand is not actually that much higher than normal, and thus the investment has limits. You can be sure that everyone is working to finish their expansions as fast as possible (hundreds of billions of investment over the next few years) and get their broken-down fabs back online ASAP, but its not like they suddenly can spend massively more resources on it. Compared to big challenges like fighting COVID its overall a minor concern, and as you point out even that did not go as all-out as it could have.

The main pinch-point is lithography machines, from what I have read. Companies like ASML [1] produce these room-sized machines to order, and the turnaround approaches 2 years from PO to delivery, depending on process node and wafer size.

[1]: https://en.wikipedia.org/wiki/ASML_Holding