How long was crypto graphics card demand ? Anyone remember ?
GPU prices have gone up and up since RTX 2000 era, basically.
My 3090 is about as expensive than it was at launch, 6 years ago.
There were crypto slumps, many moons ago, but I don’t think that’s happening again.
After the last one they never really lowered prices back, we were still buying near crypto era priced GPUs before the AI boom.
No problem. So much enshittification on the web these days, don’t need so much RAM, I might sell some.
Don’t worry all orders will be canceled when the bubble pops.
My prediction:
- Investors continue to fuel demand, choking supply
- Plans to monetize LLMs don’t pan out
- Eventually, lenders call their margins
- Investors see the called margins and get spooked; demand fizzles
- AI companies desperately try to liquidate their positions. Seeing the price of components, there’s a rush to sell while prices are high.
- PC components flood the market
- Record high supply, record low prices
- PC gaming golden age
I want to believe
Just like how GPUs became cheap and plentiful after mining crypto became passé, right?
…They actually did.
I got an AMD 7950 is a crypto bust. Then a Nvidia 980 TI in the next one, dirt cheap.
Nvidia figured it out in the most recent bust though, and controlled supply fairly well.
After the last crypto bust, prices stayed up though. They’re still way higher than before the last crypto boom.
And nVidia, AMD, Intel all know better now than to overproduce consumer GPUs.
PC components flood the market
*Server-grade purpose built AI components flood the market
SXM to PCIe adapters already exist hehe
Not to mention my dumbass would chuck money at a whole server just to throw it in my home lab.
Its not trivial.
Check the level1 forums, where some electronics gurus are trying to get a single adapter MI250X to even boot. The problem isn’t the adapters, but firmware that makes it difficult.
I see no problem if I can snag some generic Instinct accelerators, and a server to go. I’ve already got half a rack full.
6 will actually be “RAM suppliers buy the stock in bulk and sell as refurbs, keeping prices up”
Eventually, lenders call their margins
Lending to companies doesn’t work like that.
Softbank’s $40bn loan will be called on March 25, 2027. If they can’t roll this over it’s the banking equivalent of a margin call.
They aren’t PC components. They’re server GPUs.
They don’t even have ROPs anymore. There was a time when the Nvidia P100/A100 could technically game, but that’s generations ago.
FYI, during the crypto busts, Nvidia bought back GPUs and basically incinerated them to keep demand for new stuff up.
…I’m pretty sure that’s what they will do again.
yeah, but chip manufactures are reducing output for consumer class products, so yeah, there are still PC components choked by AI’s current supremacy in the market.
It’s frustrating to see the tech industry take this path, because there were alternative paths that didn’t lead us here. We are in a selfish and shortsighted time.
HBM and Vera Rubin are not useful for PC Gaming, at least not for current architectures. Main issue is that DDR5 production simply got cut to make more HBM. But there will be a huge amount of LPDDR, NVME / SSD drives and refurbished Hard Drives available, and hardware for running inference at home for pretty powerful models will become cheap as fuck, liberating us all from the hyperscalers; the open weights models from china are available for everyone.
At least i pray it will be this way.
But aren’t the wafers uncut? It could be used for anything if they’re just in storage.
If the memory wafers are largely HBM, so they can’t be used for gaming hardware.
If you’re thinking of HBM gaming GPUs, the chip design pipeline is too long. If Nvidia/AMD/Intel started making one right now, and rushed it, it would still be many years away.
…However, a lot of LPDDR5 is being made too.
Hence the rumors of AMD hedging their bets and using LPDDR for some future GPUs.
According to Tom’s Hardware, it includes “DDR5 and LPDDR4/LPDDR5 as well as premium HBM.”
Could gpus use HBM? I thought AMD used it 10 years ago in theirs.
I’m not sure I worded this right before, but it takes ~half a decade to plan and produce a GPU.
They can’t just switch out the bus or memory on an existing design.
So they could make a HBM gaming GPU. But they would have had to have started ~2 years ago.
…That’s extremely unlikely.
It’s unlikely they’ll even plan one now.
AMD did indeed try making HBM gaming GPUs, and it turns out it’s supremely uneconomical; they immediately pivoted to the opposite extreme (narrow bus + larger cache).
However, LPDDR GPUs appear to already be in the pipe, so that will provide some GPU market relief.
I’ve never heard of uncut wavers hitting storage or the market. These things are made to be used for whatever is needed at the moment. If you hear this news regarding “Ram cap sold out”, it means all available wavers they expect to be produced for that year have been allocated to whoever paid for them. If 50% of the wavers go into HBM, 30% into DDR including LPDDR and 20% go into NAND storage (i do not have the exact numbers, but i know that HBM needs around 3x the amount of silicon compared to DDR), then you are done selling your capacity - whatever happens afterwards does not make DDR5 appear magically on the market.
Only thing that could change that would be that the AI bubble crashes before 2027, then the allocation numbers will be changed.
Not sure what you mean with the wafers hitting the market. I was only referring to the specific deal that OpenAI made with Samsung and SK hynix to purchase a metric boatload of 900,000 undiced DRAM silicon wafers per month.
According to Tom’s Hardware it includes “DDR5 and LPDDR4/LPDDR5 as well as premium HBM.”
Honestly I’m not sufficiently familiar with how that works to understand if any of that is consumer grade stuff.
So you’re not 100% incorrect, but does it happen in 2027? 2028?
Logistically it’ll probably happen eventually, but when?
OpenAI is likely to be the first domino to fall that causes the whole bubble to pop. They’re burning over $10-15 billion a year in losses, and have shown that any attempts to scale only increase that burn rate. They’re running out of investor money to toss into the fire.
And they just announced a last ditch free unlimited for everyone plan to…I dunno drum up interest or hope it converts to paying customers or something. So their costs are about to skyrocket even more, accelerating their death spiral
I guess that I should use Klarna or the like to buy enough RDIMMs to let me use Qwen 3.8 122b, at the very least. The $606 I spent on buying 32gb of 5600 memory is looking like a bargain these days, and that was just a couple months ago.
Given that hoarding RAM for AI is the main culprit behind the shortage, you’re guaranteed to get pitchforks instead of compassion here.
Replace some of the labor you would use to pay for RAM and put it into learning a lighter weight operating system.
Most roundabout “install Linux” I’ve seen yet.
At least monitors are cheap now.
At least till they decide the AIs all need faces 🤣
Until we start sanctioning them, too
Dodge must be pretty happy.
We’re getting cloud computing/gaming in everything from now on, aren’t we?
Even in this hardware hellscape you can get some pretty decent handhelds for $200. I think the low end segment can stay pretty strong with devices like those and indie games.
But yeah I think high end gaming is going to be pretty stagnant for at least the next few years.
Indies4TheWindies
“High-end” gaming had no future either way, with how larger (in size and power need) GPUs slowly got and most of it being wasted on penny-pinching of developer time (e.g. atrocious performance) of AAA studios.
Ironically, browsers are so fucking bloated these days that any thin client that can run one can also enjoy most of gaming history.
Not long-term. No computers, phones, TVs? No consumers of cloud services. Including 90% of businesses. At some point, regulators have no choice (even if they are colluding), if they don’t want to strangle their economy.
The point with dissidents being on the cloud-rope still stands though. Especially on closed platforms like Apple, Android.
With imaginary money
Yeah but as long as the manufacturers keep selling all their production for real money, everybody else is screwed and prices are going up.
And people are trying to tax the imaginary money.
Can’t tax the imaginary money. It would destroy the imaginary economy.
Oh no, how will I get my flashing lights and data hoarding fix now!!??
Things don’t add up. I see articles about computer parts becoming unavailable along with articles about data center plans being cancelled or not getting built. I assume the data centers ARE getting built and those second articles are lies to calm people down.
It’s because it’s a Ponzi scheme. Huge amounts of investment dollars are being dumped into AI and anything related to AI. If you build any kind of chips that are used in AI datacenters, you suddenly have a bunch of money invested in your company. What do you do with it? You make a deal with OpenAI (or one of the other ones) that you’ll invest that money into them if they agree to buy a bunch of chips from you. Now you have revenue on books and are showing a reasonable return on that investment.
If you were a chip manufacturer and actually believed that there was a big future in AI, you’d be using those investment dollars to build some more chip fabs. And some of these companies are doing that. But overall, moving money to another company so they will order chips for datacenters that aren’t built yet (and may never be built) is just bubble ponzi shenanigans.
Computer hardware depreciates in value very quickly. You don’t buy hardware today so you can install that hardware in a datacenter that won’t be online for another year.
If you were a chip manufacturer and actually believed that there was a big future in AI, you’d be using those investment dollars to build some more chip fabs.
Major new semiconductor fabrication plant projects are expanding across the United States, led by a $16.8 billion SpaceX and Tesla “Terafab” project in Grimes County, Texas, massive expansions by TSMC in Arizona, and Samsung’s ongoing fab development in Taylor, Texas.
Computer hardware depreciates in value very quickly. You don’t buy hardware today so you can install that hardware in a datacenter that won’t be online for another year.
Hardly unusual for speculators to buy up a hot commodity the day of release and then resell it into an inflated secondary market a few days or weeks later.
SpaceX and Tesla “Terafab” project
Ah, so the only thing we know with confidence is that it’s definitely not happening.
Capitalists wet dream is to make more money without investing any new capital. Chip fabs are in a situation like that where they can make buttload more money without changing anything except charging more.
Nothing will change because the alternative is investing billions just to keep the price down. They’d rather pocket the billions now.
My take is more of a:
1000 DCs were planned
200 are getting cancelled
800 are yet to be built, built, supplied with productsand of the 800 remaining DCs, some are buying the stock that is now free or the orders are rerouted by the ordering company to other locations that also need more hardware.
Production is usually sold a year in advance. This includes orders from retail outlets.
OpenAI has a deal with Samsung and SK Hynix to purchase 900,000 DRAM wafers per month, or 40% of global DRAM output. These are just uncut wafers, accumulating in warehouses that serve two purposes:
- Prepares them for the distribution to data centers,
- Manipulates the supply chain to squeeze their competitors, who are left with slim pickings.
You and I are the victims of this until the bubble finally pops and the deal falls through.
You and I are the victims
We’re way downstream. The “victims” are all the SaaS and smaller market cap firms that need business compute for accounting and commerce.
Case in point, I got a request for a new server delayed three months because we blew through our Q1 budget for new hardware faster than expected.
That’s not really hurting me, personally. But it means our server teams need to do more with less even as the business expands.
Local small businesses are being priced out of self hosted server options since ~October (Source: MSP)
What did that Michigan politician say “we’re used to ignoring people and just doing it anyway” something along those lines…
Why are AI companies buying consumer grade RAM?
So that you can’t.
You can’t organize against billionaires online if walled garden censored posting to the cloud is the only thing available to anyone. You can’t pirate media if you can’t store media locally. You can’t run a ad blocker if they’re not allowed on their app store. You can’t install linux if your tablet can’t be jailbroken.
You have too much freedom. They must have control back. We allowed them to have enough money to take it, they just had to fool the rubes with the promise of utopian robotic labor via addiction.
This is it. Add in that AI is a 1-way filter of information, substituting web search engines. The AI companies will get all the questions and queries from people instead of search engines, you will not find information unless through the black box service of an AI company. The death of the open web, for the billionaires benefit.
They aren’t?
They are buying up capacity from manufacturers who are stopping production of DRAM and pivot to HBM which is more profitable. Abandoning DRAM which affects a lot more than just “consumer grade RAM”
They’re publicly traded companies. Their investor meetings point out that HBM is not more profitable.
It is in theory more guaranteed profit because it’s a smaller number of bigger contracts. But that’s the gamble they’re making.
But isn’t it kind of a game of chicken then for the consumer markets? Whoever gambles the other way, can win in both volume and availability, being able to cut the others off market just by pricing, enabled by the prior two points, get brand points by the ladle while at it, and kind of become a staple of the consumer market. When the bubble pops and others return, the market isn’t the same and everyone else sits at a disadvantage?
Yes.
Which is why it’s bad that the only 3 companies with capacity currently are colluding.
Good. Time for software to be optimized again.
Truth.
I am astounded that games like Crash Bandicoot, Tomb Raider and Abe’s Oddysee with their massive sprawling worlds were crammed into such relatively tiny file sizes.
Necessity gets results
how will the vibe coders survive this?
please, will somebody please think about the feelings of the poor vibe coders :(
Vibe coders will ask the AI to write everything in Rust, memory usage will be reasonably low
The people writing everything in Electron are gonna have a fun time with their users though.
nah some dude got recently covered in press b/c he used an LLM to derive Assembly from C.
Huh? C compiles to machine code.
Yes, but why write C when we can let the LLM write assembly directly? Compilers are unnecessary bloat we don’t have a need for with modern technology.
Putting aside assembly code gets compiled as well, the bottle neck is always validation of these systems. Making the systems more opaque makes the already difficult task of ensuring it works even harder.
Assembly code isn’t ”compiled”. To quote wikipedia:
Assembly code is converted into executable machine code by a utility program referred to as an assembler. […] The conversion process is referred to as assembly, as in assembling the source code. The computational step when an assembler is processing a program is called assembly time.
My point is that people are already trying to replace compilers with LLMs, and yes, it is stupid as shit, but still real.
Also, maybe see if an LLM can help you detect sarcasm without explicit content tagging.
low-level compilers can output very ugly-looking assembly. he probably did this and then used LLM to super-optimize it. may be perfomant, but id guess that theres a risk that its unsafe.
Who cares about ugly-looking assembly ? No one’s gonna actually look at it, that’s, like, the whole point of C. And I really doubt the plagiarism machine can somehow do better than a compiler. I’m gonna look this up
Edit : is it this one ? https://arxiv.org/html/2505.11480v1
I was about to get insulted but I almost closed the lid on the laptop by accident and now have to make sure my cloude still operational on the task I gave it, so I can’t be rerouting my attention span resources
Half the joke of the AI bubble is how American tech firms prioritized brute force over efficient engineering.
Chinese to the rescue?
It will take ages for them to build capacity and they will probably be wary of growing too fast since demand and prices could drop quickly when the AI boom goes bust. Likely CXMT is already supplying a lot of the consumer market under the stickers. China will be the big winners in the long term. This will be a gift for them.
