Not an expert in the area, but I think the highest of the high-end chips is a big market, but not the biggest market as revenue for fabs. It is just the most profitable part of the market.

Maybe this changed with the AI race but there are plenty of people buying older chips by the millions for all sorts of products.

▲

mandevil 6 days ago | parent [-]

The key for getting (financial) value out of fabs is their time after they are the overtaken by the next node. The ability to keep the order book full after you have a better node is what pays off the fab. So its all the other chips- the chips for cars, for low-power internet connected devices, etc. that make the fab profitable. That is where TSMC's ability to work with different customers enables them to extract value from a fab that pure-play CPU makers struggle with.

▲

DanielHB 6 days ago | parent [-]

Ah, that makes sense. I guess Intel is stuck with making x86 CPUs for datacenters even on their old-node factories so they need to retool them for newer nodes more often/earlier because they don't have a foundry business.

	▲	mandevil 5 days ago \| parent [-]
		Oh, the integrated players (now pretty much just Intel and Samsung, but in the past people like AMD as well) have other things they make on these older fabs- modems, OOB managers, USB controllers, all the other chips that go on a motherboard, but these are lower margin than a pure-play fab can get selling to all of those customers who don't need the latest node. This is one reason Intel and Samsung are both hesitant about going to the next node- Intel has put out official statements that they are only going to 14A if they can get Foundry up with a significant partner, and Samsung is hedging their bets and being cagey about their own 1.4nm node (at least in English, I haven't seen any direct demand for a major foundry customer from Samsung, just statements saying that they were going to be delaying and might not be building it at all).