Your problem here is a belief that fabs are all the same where sand goes in, chips come out.FarleyCZ wrote: Wed Jan 14, 2026 12:07 pm [Well the problem is in the supply. There's only a finite amount of fabs. It's too volatile for TSMC to build new plants. Until the bubble bursts, big tech will buy all the chips it can get it's hands on, so they can build more data centers. Households always shared the same chip pool with datacenters, but so far the supply was enough to satisfy both demands. Now, with the bubble growing, big tech always outbids households.
The memory devices for HBM come from a different collection of fab equipment to those used for DDR5 āĀ because you have to drill big holes through them to make the stacks.
And the fab lines for DDR4 are based on older gear which can't be reallocated to the high-end stuff. Big tech ain't buying up all the capacity. It's just put a stranglehold on the very top-end stuff because they're willing to pay more. Except they're not.
Intel doesn't have a business left unless it bids the same rate for the bits of TSMC's output that it needs. Ditto Apple. And I can guarantee you, TSMC has no plans to piss off Tim Apple. Intel is also mostly not making chips for AI servers, which is why everyone thinks Intel is toast (at least right now), so has very little interest in trying to bid up prices on that stuff.
Basically, your whole thesis is bunk and arguing the toss over the minutiae isn't going to help it.
