According to Duann, PC makers have to buy from SSD module makers because NAND vendors reduced allocation to the client/consumer PC market and redirected most NAND supply to data center products.
As a result, PC OEMs like Acer, Asus, Dell, and HP cannot get enough NAND or SSD supply directly from NAND manufacturers and have to turn to module makers for solid-state drives. The latter traditionally served end-users and had plenty of aftermarket products with enhanced performance and cooling, but now they increasingly serve PC makers instead.
It’s a gold rush which will have consequences a few years down the line. The data centre market will get saturated, and with a probable collapse in the AI market thats driving this (particularly given the “winner takes all” approach all the players are following) and associated massive duplication of data centres running different AI models for different companies, it’s likely to be a collapse, not a soft landing.
Hardware companies investing in expanding their output to service the data centres demand will be over producing once the market swings the other way. Expect prices to collapse and some of these memory producing companies to go bankrupt. This is another classic sign of a bubble: everyone thinks this will keep going and going, so they invest hard in having a chunk of it. But it will inevitably hit a wall - some AI companies will fail and their data centres become redundant, and the market overall will eventually swing away from endless expansion to consolidation. And thats best case scenario; more likely it a catastrophic collapse in which case the market is getting flooded with unneeded 2nd had product from data centres sold off during bankruptcy proceedings.
It’s not a question of if the party will end, it’s just a question of when. Even if people don’t think the AI market will pop, the economics of building more and more data centres by unprofitable competitors in this market is unsustainable and has to end at some point. And the evidence is we’re already well beyond the point of diminishing returns with current AI models in terms of scaling up.
So while times are hard right now for home PC users, I’d expect there to be period in the near future of oversupply and cheap components. This year? Next year? Hard to say exactly when but the writing is on the wall for the AI bubble imo.
From where? Data center equipment isn’t really suitable for the home. A used SXM module on an adapter might give you reasonably affordable compute but is completely useless for graphics. Data center memory is often HBM, which you can’t just transplant into your home PC. Getting an EDSFF SSD into your PC might work with an adapter but it’s also going to take up a lot of space.
Plus, there have been cases in the past of companies buying up and destroying the inventory of closing data centers specifically so it doesn’t end up on the market to compete with current products. That might very well repeat.
When the bubble bursts I expect to see a semi-decent supply of high-end hardware for specific use cases. If you have a space for a rack with noisy fans and a 3000 W power draw you’ll be able to build a kick-ass AI inference rig for like 2000 bucks. Or a really fast file server. But I don’t think there’s going to be much in it for people who just want 60 FPS in current games and an SSD those games fit onto. That’ll take another couple years.
I think it’ll be 2030 at the earliest until we see actually interesting consumer hardware from the usual companies. Maybe China will swoop in and deliver something worthwhile in the meantime but I’m not holding my breath.
I think a lot of people know it will end. But they also think, “I’m gonna get mine while I can, fuck everything else”
Look on the bright side: plenty of cheap compute and hardware will be available to research centers at that point.