I just want some HDDs for my server, that’s all. Is that too much to ask? They’re fire now. $300 for a 4tb HDD? That’s a crime.

cheap RAM already next year.

Unlikely because ram has been retooled for hbm server gpus. It will be another year before it's back to ddr5.

The only thing more exciting than prices correcting is having cheap ram subsidized by Nvidia shareholders.

The chips made for data centers aren’t consumer grade. I doubt there will be a sudden decrease in RAM prices even with a few bankruptcies.

And without much complications to the "market". Since the 3 yrs supply they "bought" in advance was never paid for. Find a way to void the contracts and start making stuff people actually want to buy. It's not like nvidia is going to actually pay them at this point. E:typo

I've said it before, but there's a reason SanDisk was happy to sign multi year contracts to provide storage at fixed prices. That is not a bet on hardware staying expensive. That's a tech company that's been around thinking the prices in 2026 are as good as they were going to get moving forward.

My guess is that we will have cheap RAM already next year. 😀

Maybe. Taiwan amount for 2-3% of the global RAM production, only. When - some say if, but it doesn't look like it - China decides to start a war with Taiwan and it turns out they can't invade the island in just a few weeks, it could have an effect on their RAM production. China covers 10%. Other RAM manufacturers: South Korea 63%, Japan, also would be affected, I guess?

The companies switching to in-house chips will still need other companies to do the manufacturing. If Nvidia cut its reserved production capacity it would just go to those other companies. Demand for Nvidia's data center stuff is down, but that doesn't necessarily mean demand for data center stuff is down overall.

The AI bubble bursting might take us into yet another "once in a generation" recession, but at least we'd finally be able to actually buy computer parts again!

I mean the orange turd's war with Iran already skyrocketed the cost of living crisis in the west, so we might as well get a recession to pair with it like a nice Chianti.

I just dont understand the "in-house" chips part. What in-house chips?

Later in the article they mention "alternative" chip manufacturers. What alternative manufacturers? AMD?

@Buffalox @nothrone that would be #awesome and the end of #ramshortage ? :D

DeBeers used this strategy for decades and it's worked out for them.

Y’all in the West seriously need to bring back guillotines and public execution.

Wouldn't help—there's no deterrent effect, because people like this have too much ego to even consider that they might be caught. We need to start catching them before they make a mess.

I'm trying to think of anywhere in the non-West that has guillotines or public executions, that I'd want to emulate as a good policy model.

..

Nope, nowhere comes to mind.

“mainstream news outlets” that are supposed to be for the people

There's your problem. If you start with a false premise, all your subsequent logic is invalid.

Nvidia has done supply shenanigans like this in the past, too. Repeatedly. Oldschool hardware outlets covered it extensively.

HN posted an article yesterday about employees trying to get him back. https://news.ycombinator.com/item?id=50477231

May he fall in a hole and never get out.

You can also imply some grisly possibilites without stating it directly.. or who knows, maybe he just likes it in there and wants to stay. Could be full of chinese honey or something, idk.

May he lose all sense of smell and taste, lose his hair, and forever have itchy skin.

An issue with LLM inference is that the software architecture changes significantly, very quickly, and the combination of Nvidia’s software support + some architectural flexibility makes GPUs quite advantageous for that.

Just as an example, everyone saw KV cache as this existential limiting factor for LLMs. Now, it’s tiny. Another is the old dense model paradigm vs. the almost comically sparse ones coming out. Even matrix-multiply compute could be rendered irrelevant if Bitnet gains traction.

And that’s not to speak of training.