Comment by Animats
11 hours ago
The DRAM shortage and lack of fab capacity have also caused the Playstation 6 to slip to 2029 or so.[1] Game consoles are vulnerable. They need a lot of RAM and have to sell at a moderate price.
The IDC article says that DRAM prices are not expected to come down again. "While memory prices are projected to stabilize by mid-2027, they are unlikely to return to previous level — making the sub-$100 segment (171 million devices) permanently uneconomical." Before, they always came back down in the next RAM glut, when everybody built too much capacity. Why is that not going to happen next time?
[1] https://www.heise.de/en/news/Storage-crisis-Playstation-6-co...
You’re asking why a market that has had 3 price fixing lawsuits in less than 2 decades (criminal convictions in 1998, civil in 2006 and 2018) isn’t going to follow market dynamics?
One reason we end up with excess capacity is process improvements; adding new fabs to get more density or performance doesn't make old fabs go away, and so we go through cycles of excess capacity. Demand has been relatively constant.
Here we're facing different forces-- unprecedented demand for DRAM that may be durable. But it also looks like the pace of supply changes may be decreased as process improvements get smaller and the industry stops moving so much in lockstep.
It still matters what happens to the demand function, though. If enough AI startups blow up that there's a lot of secondhand SDRAM in the market, and demand for new SDRAM is impacted, too, that will push things down.
Sort of like what happened with the glut of telecom equipment after
> Why is that not going to happen next time?
Because this shortage isn't natural, it's the result of OpenAI flexing monopsony power to deprive everyone else for its strategic gain. Unlike an organic shortage, there is no compelling reason for otherwise excess capacity to be built, since this artificial shortage can end as arbitrarily as it started.
The datacenters are still going to be built, and their usage won't suddenly fall just because the companies behind some of the products on them suddenly lose value. The demand is not tied to their profits, so I find it unlikely for the shortage to just end.
These data center projects are losing hundreds of billions of dollars which they don't have, and some evidence is starring to come out they're just money laundering schemes to get money from the government to contractors. I wouldn't bet on them all being built.
There far too many railways, amusement parks, housing developments and other bubble ventures that were either never even completed after wasting a lot of money or went bust soon after opening.
No reason the same can't happen now - especially for something as expensive and faily easily re-sellable as a datacenter & the hardware insite. Just rip it all out and sell it for parts where they are actually needed.
3 replies →
> The IDC article says that DRAM prices are not expected to come down again
Sure thing. I'd take a look at IDC & similar firms' forecasting history before worrying too much about what they say.
There is an AI boom right now. There will be a consolidation cycle at some point. When that happens half the players, if not more, will disappear. The huge hardware budgets will go with them.
We also can't be certain that the DRAM makers aren't capitalizing on this opportunity because they can. Remember: all of them are convicted monopolists. As in actual prison time convicted. And fined. And lost civil lawsuits. Multiple times.
I just can't see AI paying enough of a premium on HBM to justify the DRAM spikes. Frankly I can't see the volume either. Wafer starts on DRAM are dramatically bigger than you are probably imagining. DRAM is in practically everything these days. AI servers is but a drop in the bucket. 10% of the market? Yeah right, if its 4% I'd be shocked. And you are telling me a shift of 4% of wafers to HBM is driving these prices and shortages?
I humbly suggest if you look at the numbers something smells funny.
Disclaimer: none of us has access to the actual data, a lot of it is inferred by industry players. Some are well connected and usually accurate but that is not evidence. Therefore it is possible this is a genuine market action and nothing nefarious is going on.
HBM is not normal memory. It uses a lot more area per bit and has lower yield too. So a Gb of baseline DRAM and a Gb of HBM are very different measurements, the latter equates to so much more in terms of volume.
We don’t know when the PS6 is going to be released, as of now that is just a rumor.
I'll bet you $5000 it doesn't release before 2029.