In re: semiconductor shortage and its length
Most people (except maybe my pal Koog) are unaware of how semiconductor fabs operate. They use very, very expensive equipment that is very, very sensitive and deals with harsh, extremely corrosive chemicals and highly reactive plasmas that wear them ~OUTT quickly. These tools need to be maintained and calibrated regularly - and replaced. And you cant just order them on Amazon. There are only a small number of vendors for this equipment, which are mostly in Silicon Valley/Boston or Japan - Applied Materials, Lam Research, KLA-Tencor, Kulicke & Soffa, Novellus, and Varian in the USA - ASML in the Nederlands - and Tokyo Electron, Nikon, Hitachi, and Teradyne in Japan are the key major players.
And fabs are set up in production lines. A fab may have several lines, each optimized and specialized for specific processes that are for specific chip types.
Adding a fab line often means first adding space by building a new building(s) to house the new equipment. Once the equipment is ordered, it has to be adapted for the specific process that is used, optimized for that process by endless testing, and then when a process and its equipment is yielding test samples at a good yield, the who line needs to be validated before production runs can begin.
This typically takes 1-2 years, best case - you can kind of measure this by the time it takes one generation of DRAM chip density to go up a level - say the transition from 4M DRAM to 16M in the early 90s, which took several years and was a real problem resulting in DRAM shortages and very high DRAM prices.
The takehome message here is that there is no quick fix for a shortage of semicon manufacturing capacity. It has a lag time of years and will only gradually build. Just getting the equipment suppliers to ramp up is no small feat. They also have capacity and personnel constraints. PLUS, the harder they run the existing line equipment, the faster it wears ~OUTT and the more maintenance downtime they produce, and the more the semicon equipment makers have to use their capacity just to REPLACE work ~OUTT equipment, reducing their ability to supply new fabs and new fab lines.
So, the current semiconductor shortage is not going to go away any time soon. Maybe in 4-5 years the overall fab capacity will be able to meet demand for semiconductors - then again it may take even longer if semiconductor demand increases beyond current forecasts and backlogs.
It is kinda a grim situation unless you are a semicon equipment manufacturer or a semicon fab that has good processes with high yields and well-maintained recent equipment.
The fabs that make GPU chips like those of nVidia are already way behind the demand curve - largely because of the GPU usage in BitCoin mining! Those fab lines are maxxed ~OUTT and the price of certain GPU chips is now often over 5 times the MSRP in the aftermarket. A GPU chip that has a manufacturer suggested retail price of $900 often is now sold on places like eBay and gamer websites for $5,000.
Semicon manufacture is not like building cars, soup cans, or houses. There is a considerable amount of black magic involved - endless process tweaks to improve yields - even while production runs are going on. Endless process development and tweaking in dedicated testing fab lines that are only for process development and dont do production runs.
It is a complex biz with plenny of trade secrets and black magic that is kept ~OUTT of the public domain, and those that are made public are covered by a shit-ton of patents. So entering the semicon biz is hard. The barriers to entry are very high. A fab today will cost about $3-5 Billion US Dollars if you include equipment along with the buildings and infrastructure at the site.
This is not like pitching a tent and selling lemonade. There is no Moonshot Program that anyone - USA, China, Taiwan, Japan, Germany, etc. - even combined - can create that could meaningfully speed this up.
