I'd come at this from a different angle: we still want this to be market system, so we need to make this priced into the market. How can we price this in?
I would try to solve this by making the market structure reflect the underlying difficulty: we have to decide what capacity to produce years in advance, to construct the memory fabs. So this should be a futures market, and a capacity crunch would affect short-term-futures, but leave full term futures at the same price. Because the companies supplying the memory can just construct more capacity to fill those futures at the same cost regardless of the AI demand.
ah yes, adding a futures market to a sector heavily invested in AI certainly won't lead to catastrophic over-speculation that will collapse the industry entire
the market can't fix this. the problem here is that the consumer market is dwarfed by big buyers. the consumer market is simply less profitable. traditionally when a market is no longer profitable enough for big businesses, it opens the room for new small businesses to step in. but that won't work here unless someone finds a way to make small fabs profitable.
I buried the lede in my suggestion. I was saying that a future would give enough time to build or expand the fab in order to fill it. This means the company can fill an arbitrarily large number of orders - both consumers and big players - because it will just build the capacity to meet it.
but if the AI RAM market is so much more profitable than the consumer RAM market, who is going to buy consumer RAM futures?
to buy consumer RAM futures i would have to predict the price the RAM can be sold at 5 or 10 years from now. that seems very risky. why would i do that instead of buying AI RAM instead?
I think the epic mistake was made by investors and executives who bought tales about imminent super-intelligence about to displace millions of knowledge workers from entire industries replacing them with AI datacentres.
These people are the main reason why AI companies have unlimited funding, and can afford to buy global RAM supply for years in the future despite their expenses exceed revenue by billions.
Google (which is diversified more and whose performance is less likely to track AI outcomes closely) aside, the two other members of the AI Big 3 are non-public, so, no, you really can’t.
If you could afford a computer at pre-AI prices (which is presumably what you're complaining about? If you couldn't afford a computer before, AI hasn't changed that for you), you have enough money to put where your mouth is.
You can make a lot of money with shorts if you are correct about when the market will move. If you think that the market is overestimating the real value you have not even half of the puzzle.
The same thing was said 4 years ago about NVIDIA on HN, that it's stock it's outrageously overpriced, given it's $20 bln revenue, that it should have at least 10 times more revenue to justify that stock price, which is fantasy, that there is no plausible way for such demand no matter what you think about GPT-2.
nVidia earning is way above zero by a healthy margin. This was also true 4 years ago. How much their shares should cost is debatable but still, nVidia is obviously a profitable business.
Anthropic net loss in 2025 was $42 bln, OpenAI $38.5 bln. Both are spending enormous amounts with no obvious path to profitability.
> How should the memory companies have acted differently?
Same way nVidia did through the crypto insanity - make sure they're supplying enough to the consumer market so it doesn't get completely destroyed and pulls down the other parts of the consumer market they're reliant for long term success.
> Should we blame this on memory companies or the AI companies bidding for memory?
Blame doesn't change the outcomes, neither does it improve the negative consequences. Think in terms of "what does destruction of our consumer market mean for my prosperity?" not "oh, how do defend poor companies again?"
Yes, but zoom out more. Wealth inequality and the k-shaped economy are the real culprit.
When a few people have so much wealth that they are no longer price-sensitive, they bid up the price of everything and anything of value be it stocks, real estate, computer hardware, fine arts, sports teams, etc.
The result is that a business which tries to make quality products at reasonable prices will fail. They aren’t luxurious enough for the people with money. They aren’t cheap enough for the people without. Customers in the middle hardly exist. Any valuable good that is genuinely scarce will inevitably become a luxury.
right, what to do is rather obvious. how to do it, less so.
personally i think it is a mix of education and and taxes, but i also have some other ideas like limiting the size of businesses and especially limiting mergers. limiting what businesses can do to fight competition. etc.
we know that monopolies are bad, but even without monopolies businesses buying the competition or merging have not always been good and more often serve to eliminate competition and make the owners/investors rich, rather than making the business sustainable.
for example once you have reached a certain size, or if you are among the top two or three in your market you should simply not be allowed to buy competing businesses. or you should not be allowed to make exclusive contracts like: "you can only sell coca-cola here if you do not also sell virgin cola"
basically, limit the ability for wealth to accumulate in one place and just a few people profiting from it.
Those are good ideas, but they are ways to prevent wealth accumulation, not to undo it. I think we also need to consider direct transfers of existing wealth to remedy the current situation.
fair point. it depends on how fast you want to get it done. if the rules i suggest are implemented than at least businesses could not grow any further. one thing that could be added is to break up companies that are to big. that has been done before even if not well (AT&T Bell went from a huge monopoly to many small/regional monopolies).
going to fast will produce to much resistance. instead of attacking the rich, let this generation die out, and make things harder for the next generation.
the important part is to change the system for a better future and less the status quo.
> fair point. it depends on how fast you want to get it done.
I want it yesterday. :-)
> going to fast will produce to much resistance. instead of attacking the rich, let this generation die out, and make things harder for the next generation.
But this generation will just pass their wealth, and their determination to protect it, to their heirs. There has to be some point at which the flow reverses and the accumulated wealth of the super-rich actually begins to decrease, not just increase at a slower rate.
People have the option of just not buying stuff. I sold my Tesla in 2021 for a stupid price, drove a beater for two years, and bought another new car a couple years later. Now, certainly not everyone has that luxury, but this is computers, not real estate. Prices will crash.
Way to miss the whole point of the article and fail to understand the point. Do you think if significant part of US industry goes bankrupt because they can't source parts for their products, work machines and consumers disappear... you'll still have luxury of running around in a Tesla?
I hope so. I feel so, so, so terribly guilty for not building my wife a computer for her in 2023. She is rocking a 1080ti, and I really wish I had at least upgraded her system, but now I just straight cannot afford to even consider it, given the price of rent, food, gas, energy, all on top of the actual hardware prices.
And I am a SWE.I am not making bad money. I can't imagine what others are going through.
I was so lucky to buy a high-end miniPC to replace my home PC for about $1,500 USD in 2025. The same model now sells for more than $8,000 USD, without disk or RAM. Triple that if you match my original disk and RAM buy. Everything got incredibly expensive, not only storage and RAM.
Unless you know, your main and only PC/notebook just died and need a replacement ASAP.
> Prices will crash.
Does your crystal ball also say when exactly? If my laptop broke then I need a replacement now in the overpriced market, not in X years when prices will crash.
If there were such a crystal ball, the same market doing the gouging would be pricing it in already ;) does suck that "wait and see" is the optimal answer to the average consumer but I guess that's the capitalist hellscape we have wrought
I think it's more likely the industry would rather rent us our compute and storage than let us own it ourselves.
We can have our Raspberry Pis and our ESP32s, but meaningful amounts of compute and especially storage will be rental-only.
Requiring proof-of-humanity using device-bound cryptographic attestations (which the "age verification" camp will rally behind) will insure the toy computers we are allowed to own won't be able to conduct meaningful online discourse. Only phones and other locked-down platforms will be trusted to make the attestations necessary to interact with banks, merchants, and governments.
2/ where is your capital and your means of production? Your machines? Your factories? Your raw materials, your tools? How long do you survive tomorrow when you quit your job and you have no other income?
You're not a capitalist. You're just a consumer for them to suck dry. The faster you learn it, the better off you will be.
You are right. People are misusing "capitalist" to mean someone who favours a capitalist system, where as it actually means someone who posses capital; someone whose relies on possession of assets rather than earned income.
This comes at a perfect time. The public doesn't need computers, they can just use their phones.
If the consumer had access to this RAM, they might all just run local or semi-local AI. It's important to outbid them so you can rent AI to them, and extract money from them in a million other ways while they use it.
Make RAM half the price it was before and you'd still have very few people trying to do local AI on it because it's just so damn slow for the task. Having 12 channels of DDR5 in an Epyc system with NVMe storage is still one of the slowest ways I can run LLMs locally.
The M4 (soldered wide LPDDR5X + unified architecture + GPU with neural cores) is also closer in design to a dedicated GPU than an Epyc server with DDR5 or the components in the article.
But yes, a dedicated consumer GPU will still typically outpace the M4 quite well (if things fit in VRAM) and both will have their socks blown clean off by a hyperscaler GPU cluster even when it's running a model 10x the size than either of those can fit.
No one is running cpu-only inference anymore. Deepseek flash / Qwen flash next et al only need ~12-24 gb of VRAM for the active experts, so anyone with a decent gpu + lots of RAM + server cpu can run them at okay speeds.
Or one can also run Strix Halo / DGX spark / Apple. They handle MoEs well and use normal (not HBM) ram.
some benchmarks from locallama:
Deepseek V4:
"~25t/s decode at full quant, no speculative decoding, 2x3090 (only one being used for this model) + 9684x w/ 12 channel ddr5 4800, latest llama.cpp"
"Getting about 300 tk/sec pp and 23 tk/sec decode on Mac M3 Ultra 256 GB. Using original full precision model weights."
Yeah, great. Let me tell the hospital whose MRI needs maintenance to just not buy whatever it needs. Wome will die but that's a sacrifice I'm willing to make, right?
The AI companies must be regulated. They are already destroying the world and the economy before they can even make an attempt at destroying humanity.
Either AI will get cheap enough and blow up and destroy the economy by eating everyone's lunch and killing off sectors left and right. Or it will become an uncontrolable weapon destroying the internet and killing off people on the battlefield. Or it will tank the economy because there may not be a way to monopolize it making the billions of investments moot leading them to their only leverage of hoarding up all the resources pricing out the competition but also destroying the industry. And maybe all of those at once.
An aside, this is a poorly-structured article. It's mostly screenshots of citations from other articles with no alt or any other way to actually search through what they're citing.
nvidia investing in every aspect of the hype cycle is terrifying. as an ai skeptic i almost don’t want the bubble to burst anymore, i don’t know if the economy can recover at all. perhaps a new economic paradigm will emerge, but if so, the transition is going to be rough.
The economy always recovers because the value in society is create by people existing and wanting things. The people won't stop existing or wanting things and eventually they just will believe something new is valuable and the economy will recover and grow. Because nothing material was lost just numbers in the spreadsheet. The recovery will be rough though your right about that.
Building a gaming PC in the spring of 2025 was probably one of the luckiest things I've ever had happen. 32G DDR5 for 90 bucks, 4TB drive for about 200 bucks. Funny thing is I held off upgrading my GPU to wait for the price to come down from the crypto rush :)
81 comments
[ 6.0 ms ] story [ 72.1 ms ] threadI would try to solve this by making the market structure reflect the underlying difficulty: we have to decide what capacity to produce years in advance, to construct the memory fabs. So this should be a futures market, and a capacity crunch would affect short-term-futures, but leave full term futures at the same price. Because the companies supplying the memory can just construct more capacity to fill those futures at the same cost regardless of the AI demand.
Since almost everybody is a consumer of those companies, I do suggest we reuse the governance system we have that solves other "everybody problems".
to buy consumer RAM futures i would have to predict the price the RAM can be sold at 5 or 10 years from now. that seems very risky. why would i do that instead of buying AI RAM instead?
These people are the main reason why AI companies have unlimited funding, and can afford to buy global RAM supply for years in the future despite their expenses exceed revenue by billions.
Shorting doesn’t only require you to be right. It requires perfectly timing when the market will realize you’re right.
Can't afford thing? Gamble on stock market instead.
Ridiculous argument that already rich people make.
Anthropic net loss in 2025 was $42 bln, OpenAI $38.5 bln. Both are spending enormous amounts with no obvious path to profitability.
memory companies could have attempted to protect consumers, at least a little, but the AI money machine goes brrrr
Well if your thesis is that they should have acted differently, then we should blame the laissez-faire capitalists.
Same way nVidia did through the crypto insanity - make sure they're supplying enough to the consumer market so it doesn't get completely destroyed and pulls down the other parts of the consumer market they're reliant for long term success.
> Should we blame this on memory companies or the AI companies bidding for memory?
Blame doesn't change the outcomes, neither does it improve the negative consequences. Think in terms of "what does destruction of our consumer market mean for my prosperity?" not "oh, how do defend poor companies again?"
When a few people have so much wealth that they are no longer price-sensitive, they bid up the price of everything and anything of value be it stocks, real estate, computer hardware, fine arts, sports teams, etc.
The result is that a business which tries to make quality products at reasonable prices will fail. They aren’t luxurious enough for the people with money. They aren’t cheap enough for the people without. Customers in the middle hardly exist. Any valuable good that is genuinely scarce will inevitably become a luxury.
personally i think it is a mix of education and and taxes, but i also have some other ideas like limiting the size of businesses and especially limiting mergers. limiting what businesses can do to fight competition. etc.
we know that monopolies are bad, but even without monopolies businesses buying the competition or merging have not always been good and more often serve to eliminate competition and make the owners/investors rich, rather than making the business sustainable.
for example once you have reached a certain size, or if you are among the top two or three in your market you should simply not be allowed to buy competing businesses. or you should not be allowed to make exclusive contracts like: "you can only sell coca-cola here if you do not also sell virgin cola"
basically, limit the ability for wealth to accumulate in one place and just a few people profiting from it.
going to fast will produce to much resistance. instead of attacking the rich, let this generation die out, and make things harder for the next generation.
the important part is to change the system for a better future and less the status quo.
I want it yesterday. :-)
> going to fast will produce to much resistance. instead of attacking the rich, let this generation die out, and make things harder for the next generation.
But this generation will just pass their wealth, and their determination to protect it, to their heirs. There has to be some point at which the flow reverses and the accumulated wealth of the super-rich actually begins to decrease, not just increase at a slower rate.
And I am a SWE.I am not making bad money. I can't imagine what others are going through.
Unless you know, your main and only PC/notebook just died and need a replacement ASAP.
> Prices will crash.
Does your crystal ball also say when exactly? If my laptop broke then I need a replacement now in the overpriced market, not in X years when prices will crash.
last news I heard was end of 2027, when several new production lines from existing producers start operations
I think we're seeing the end of the PC as we know it, personally.
/s if you need it
We can have our Raspberry Pis and our ESP32s, but meaningful amounts of compute and especially storage will be rental-only.
Requiring proof-of-humanity using device-bound cryptographic attestations (which the "age verification" camp will rally behind) will insure the toy computers we are allowed to own won't be able to conduct meaningful online discourse. Only phones and other locked-down platforms will be trusted to make the attestations necessary to interact with banks, merchants, and governments.
Yes, it's the manufacturers fault, not the unbelievable market demand.
We are all Capitalists, until the Market comes after the stuff we love.
1/ speak for yourself
2/ where is your capital and your means of production? Your machines? Your factories? Your raw materials, your tools? How long do you survive tomorrow when you quit your job and you have no other income?
You're not a capitalist. You're just a consumer for them to suck dry. The faster you learn it, the better off you will be.
If the consumer had access to this RAM, they might all just run local or semi-local AI. It's important to outbid them so you can rent AI to them, and extract money from them in a million other ways while they use it.
But yes, a dedicated consumer GPU will still typically outpace the M4 quite well (if things fit in VRAM) and both will have their socks blown clean off by a hyperscaler GPU cluster even when it's running a model 10x the size than either of those can fit.
No one is running cpu-only inference anymore. Deepseek flash / Qwen flash next et al only need ~12-24 gb of VRAM for the active experts, so anyone with a decent gpu + lots of RAM + server cpu can run them at okay speeds.
Or one can also run Strix Halo / DGX spark / Apple. They handle MoEs well and use normal (not HBM) ram.
some benchmarks from locallama:
Deepseek V4: "~25t/s decode at full quant, no speculative decoding, 2x3090 (only one being used for this model) + 9684x w/ 12 channel ddr5 4800, latest llama.cpp"
"Getting about 300 tk/sec pp and 23 tk/sec decode on Mac M3 Ultra 256 GB. Using original full precision model weights."
Qwen-Flash-Next: https://www.reddit.com/r/LocalLLaMA/comments/1wom3fe/qwen_38...
https://www.reddit.com/r/LocalLLaMA/comments/1vcaztx/what_sp...
Someone misspelled "Altman", https://www.mooreslawisdead.com/post/sam-altman-s-dirty-dram...
> Enhanced memory performance, system stability, and compatibility with CXMT memory chips for improved low-latency operation.
A year ago I'd hardly know what CXMT is.
Consumer market is not you and your console only.
Either AI will get cheap enough and blow up and destroy the economy by eating everyone's lunch and killing off sectors left and right. Or it will become an uncontrolable weapon destroying the internet and killing off people on the battlefield. Or it will tank the economy because there may not be a way to monopolize it making the billions of investments moot leading them to their only leverage of hoarding up all the resources pricing out the competition but also destroying the industry. And maybe all of those at once.