I’m a bit of an optimist. I think this will smack the hands of developers who don’t manage RAM well and future apps will necessarily be more memory-efficient.
Only if the software is optimised for either in the first place.
Ton of software out there where optimisation of both memory and cpu has been pushed to the side because development hours is more costly than a bit of extra resource usage.
Ok so Samsung, SK Hynix and Micron do not have the capacity to meet demand. Also, what little capacity they do have they are allocating to HBM over DRAM. Based on my limited knowledge HBM can not be easily repurposed for consumer electronics. Translation: main street is cooked for the next 3-4 years.
It doesn't stop there though. OpenAI is currently mired in a capital crunch. Their last round just about sucked all the dry powder out of the private markets. Folks are now starting to ask difficult questions about their burn rate and revenue. It is increasingly looking like they might not commit to the purchase order they made which kick-started this whole panic over RAM.
Soo ... how sure are we that the memory makers themselves are not going to be the ones holding the bag?
To add a more local hurdle as well, the Dutch power grid is at capacity and its managing company is now telling companies that planned to build a datacenter that they can't be connected to the grid until 2030, even though said companies already paid for and got guarantees about that connection.
That is, memory capacity is reserved for datacenters yet to be built, but this will do weird things if said datacenter construction is postponed or cancelled altogether.
> the Dutch power grid is at capacity and its managing company is now telling companies that planned to build a datacenter that they can't be connected to the grid until 2030, even though said companies already paid for and got guarantees about that connection.
Are the Netherlands a large proportion of global datacenters?
Amsterdam hosts a major internet exchange. It's not a bad place to build a datacenter and there are many. Northern latitude brings free air cooling, but also additional distance to clients. Lots of peers in AMS-IX, but not a lot of oceananic cable landings (one with two paths to the US, but most of the submarine cables land nearby in Europe)
Yes. Amsterdam has one of the largest IXPs (AMS-IX) in Europe and is also one of the largest European markets for Internet Infrastructure services (i.e. hosting, DNS provision, domain name registration, etc.)
What the grid looks like in different countries is very different. The Dutch power grid is already almost 50% renewables, which is an inconvenience for adding capacity because that's around where you have to start really dealing with storage in order to add more.
In most other places the percentage is significantly less than that and then you can easily add more of the cheap-but-intermittent stuff because a cloudy day only requires you to make up a 10% shortfall instead of a 50% one, which existing hydro or natural gas plants can handle without new storage when are more of them to begin with.
That guarantee is not as much of a guarantee as stated in the media. You get a guarantee it will be planned at a certain time (as in looked at), not that it will be build. The cost of doing business is taking risks and mitigating them. There is a reason the nuclear plant in Borsele was build: an aluminium smelter. Maybe you should arrange for something similar as a datacenter (no politician will fall on a sword for that but you can try). The (original) power draw is about the same 80-100MW.
The Radeon VII came out in 2019 as a $700 consumer GPU with an 1TB/s HBM2 memory subsystem which is more than any consumer GPU you can get today, including the high-end ones afaik. At that point in time, there was a whole lineup of AMD GPUs with HBM going down into the midrange.
If they could make this stuff and sell it to regular people a decade ago for very palatable prices, why do they come up with the idea that this is the technology of the gods, unaffordable by mere mortals?
I have been wondering this recently. It was the convention that if you wanted to keep costs down, try to keep the memory bus size down as low as possible. Still remember the awful Radeon 9200 SE - 64bit data bus that strangled an already slow GPU.
Heck, I have a phone with a 16bit memory bus for instance. The high(ish) clock rate only makes up the difference slightly.
But with general prices on all components going up, it might not be such a big factor any more.
HBM migght make sense for higher end products which can free up space for the lower end that will never use the tech.
The market is already stagnated. Even if OpenAI doesn’t buy what they reserved other players will do so. SK Hynix CEO said there is a 20% gap between supply and demand per year. And that doesn’t account the shock effect that will take place the moment prices normalize and everyone and their dog will go out and start buying inventory to avoid the next crisis. I for one would certainly buy more than I currently need just in case.
But wouldn’t you rather hbm prices come down first ? Memory makers will be fine. There is practically infinite demand.
Unless you get china style rationing of compute per person world wide.
The real issue is everyone wanting to upgrade to hbm, ddr5, and nvme5 at the same time.
I am betting the pendulum swings faster to the other side to excess capacity as all the construction lies of Altman fall through with financiers waking up the the fact they can't build the infrasctructure as fast nor make any profits on that infrastructure that will get built.
> Soo ... how sure are we that the memory makers themselves are not going to be the ones holding the bag?
We aren't. The remaining memory manufacturers fear getting caught in a "pork cycle" yet again - that is why there's only the three large ones left anyway.
If they don't expand capacity much, the only negative consequences I foresee happening for them is that they might lose spending discipline, and that systems will be set up to make do with a little less memory. Apart from that, it's just very high profits followed by more or less regular profits.
They could wind up losing all their business to China though.
China has memory makers who are creeping up through the stages of production maturity, and once they hit then there's no going back.
If the existing makers can't meet supply such that Chinese exports get their foot in the door, they may find they never get ahead again due to volume - that domestic market is huge so they have scale, and the gaming market isn't going to care because they get anything at the moment, which is all you'll need for enterprise to say "are we really afraid of memory in this business?"
Memory makers did get themselves into this situation by selling all wafers for empty promises and alienating everyone but OpenAI tbh. I do hope they end up holding the bag once again, cause after covid and the cartel thing they don't seem to ever learn their lesson on how to have the tiniest amount of integrity.
> Memory makers did get themselves into this situation by selling all wafers for empty promises and alienating everyone but OpenAI tbh.
Wasn't the problem here that OpenAI was negotiating with Samsung and SK Hynix at the same time without the other one knowing about it? People only realized the implications when they announced both deals at once.
Permanent public ownership of (very large stakes in) these companies doesn't seem like such a bad idea anymore, does it? It's what we used to have for most of the 20th century at least in Europe.
It's a business with huge up-front capital expenses and typically very low margins. Supply is scaling up slowly because it's hard, and if you overshoot, you go out of business.
Nobody is "allowing" this. It's a natural property of being both advanced technology and a commodity at the same time.
The strange deals on the entire future output are what was allowed. Try to do the same thing with onions and the government understands you are a criminal.
Allowed? We live in a neoliberal world where corporate monopolies / oligopolies aren’t even remotely regulated. If you try to do even the gentlest regulation of companies people scream about communism and totalitarianism. Unless the regulation serves the monopolies by making it harder to enter the market.
It started with raegan, and even parties on the “left” in the west believe in it with very few exceptions.
> We live in a neoliberal world where corporate monopolies / oligopolies aren’t even remotely regulated. If you try to do even the gentlest regulation of companies people scream about communism and totalitarianism. Unless the regulation serves the monopolies by making it harder to enter the market.
The thing that enables this is pretty obvious. The population is divided into two camps, the first of which holds the heuristic that regulations are "communism and totalitarianism" and this camp is used to prevent e.g. antitrust rules/enforcement. The second camp holds the heuristic that companies need to be aggressively "regulated" and this camp is used to create/sustain rules making it harder to enter the market.
The problem is that ordinary people don't have the resources to dive into the details of any given proposal but the companies do. So what we need is a simple heuristic for ordinary people to distinguish them: Make the majority of "regulations" apply only to companies with more than 20% market share. No one is allowed to dump industrial waste in the river but only dominant companies have bureaucratic reporting requirements etc. Allow private lawsuits against dominant companies for certain offenses but only government-initiated prosecutions against smaller ones, the latter preventing incumbents from miring new challengers in litigation and requiring proof beyond a reasonable doubt.
This even makes logical sense, because most of the rules are attempts to mitigate an uncompetitive market, so applying them to new entrants or markets with >5 competitors is more likely to be deleterious, i.e. drive further consolidation. Whereas if the market is already consolidated then the thicket of rules constrains the incumbents from abusing their dominance in the uncompetitive market while encouraging new entrants who are below the threshold.
> Soo ... how sure are we that the memory makers themselves are not going to be the ones holding the bag?
I hope they do, they did not have to agree to sell so much RAM to one customer. They’ve been caught colluding and price fixing more than once, I hope they take it in the shorts and new competitors arise or they go bankrupt and new management takes over the existing plants.
Don’t put all your eggs in the one basket is how the old saying goes.
This will result in demand destruction which will starve the enterprise which will starve the hyperscaler. theres no situation where people not being able to afford hardware for 4 years results in the bubble not popping
They won't be, prices are high because they are refusing to build capacity for demand that may evaporated by the time they are done. They are holding back and building only enough so when the bubble pops they will be fine.
So the ML hate is weaponized in the form of memory demand collapse FUD, and the public at large has to pay through their nose for it... thanks party poopers!
I don't think its from the ML collapse FUD, its most likely from the multiple time's in the past when they overbuilt and it resulted in a memory oversupply and price collapses. The 1985–1988, 1993–1994, 1998–2002 and the post pandemic oversupply. These were all cases where shortages followed by over corrections caused oversupply, financial losses due to low prices and fewer surviving companies. I think they're taking their time and are cautiously adding more capacity in such a way that prices won't end up collapsing again. Regardless, the result is still that we the consumers have to pay more.
At this point the remaining memory companies are… the ones that didn’t die during an over-supply collapse, right? I guess there’s been a strong evolutionary pressure against giving consumers what we want, haha.
If they gradually increase production capacity then prices stay high for 10+ years (or for as long as it takes for demand to crash) because a gradual increase in production takes that long for them to add enough capacity for current demand.
If they add enough capacity to meet current demand quickly then when demand crashes they still have billions of dollars in loans used to build capacity for demand that no longer exists and then they go bankrupt.
I would expect that OpenAI gets as much money as they ask for for the next 10 years.
There’s virtually infinite capital: if needed, more can be reallocated from the federal government (funded with debt), from public companies (funded with people’s retirement funds), from people’s pockets via wealth redistribution upwards, from offshore investment.
They will be allowed to strangle any part of the supply chain they want.
China already has a well developed DRAM industry, as DRAM is somewhat easier than logic, and can tolerate a much higher defect rate. The industry will figure this out.
Another point is I often see the money argument - like country X has more money, so they can afford to do more and better R&D, make more stuff.
This stuff comes out of factories, that need to be built, the machinery procured, engineers trained and hired.
I think the article has a giant blind spot as far as China is concerned , considering they have already a mature enough memory ecosystem via YMTC that Apple was considering sourcing from them. As well as continued expansion in the DRAM and HBM Fabs [1].
It feels like the memory cartel once again trying to incentivise their various govt to cough up some more tax breaks/funding to cushion the AI buildout bet that they made and the bubble seeming about to pop.
In any case if they leave the consumer market underserved it should be no surprise if before that 2030 prediction we are all on cheaper YMTC memory modules.
I think you're massively overestimating how much money is really accessible here. The parent comment's right that all of the easily available VC & private equity investment is basically used up. OpenAI was struggling to sell $600M of private equity, the big multi-billion dollar investment packages had lots of conditions and non-cash in it.
> more can be reallocated from the federal government (funded with debt)
While this is the most reliable funding, it's still not very accessible. OpenAI is a money pit, and their demands are growing quickly. The US government has started a bunch of very expensive spending. If OpenAI were to require yearly bundles of it's recent "$120B" deal, that's 6% of the US' discretionary budget. 12.5% of the non-military discretionary budget. (And the military is going to ask for a lot more money this year) Even the idea of just issuing more debt is dubious because they're going to want to do that to pay for the wars that are rapidly spiralling out of control.
None of this is saying that the US government can't or wouldn't pay for it, but it's non trivial and it's unclear how much Altman can threaten the US government "give me a trillion dollars or the economy explodes" without consequences.
Further deficit-spending isn't without it's risks for the US government either. Interests rates are already creeping up, and a careless explosion of deficit may well trigger a debt crisis.
> from public companies (funded with people’s retirement funds)
This would be at great cost. OpenAI would need to open up about it's financial performance to go public itself. With it's CFO being put on what is effectively Administrative Leave for pushing against going public, we can assume the financials are so catastrophic an IPO might bomb and take the company down with it. Nobody's going to be investing privately in a company that has no public takers.
Getting money through other companies is also running into limits. Big Tech has deep pockets but they've already started slowing down, switching to debt to finance AI investment, and similarly are increasingly pressured by their own shareholders to show results.
> from people’s pockets via wealth redistribution upwards
The practical mechanism of this is "AI companies raise their prices". That might also just crash the bubble if demand evaporates. For all the hype, the productivity benefit hasn't really shown up in economy-wide aggregates. The moment AI becomes "expensive", all the casual users will drop it. And the non-casual users are likely to follow. The idea of "AI tokens" as a job perk is cute, but exceedingly few are going to accept lower salary in order to use AI at their job.
There's simply not much money to take out of people's pockets these days, with how high cost of living has gotten.
> from offshore investment.
This is a pretty good source of money. The wealthy Arabian oil states have very deep slush funds, extensively investing in AI to get ties to US businesses and in the hope of diversifying their resource economies.
I'm a bit surprised the article makes no mention of Google's TurboQuant[0] introduced 26 days prior.
Given that TurboQuant results in a 6x reduction in memory usage for KV caches and up to 8x boost in speed, this optimization is already showing up in llama.cpp, enabling significantly bigger contexts without having to run a smaller model to fit it all in memory.
Some people thought it might significantly improve the RAM situation, though I remain a bit skeptical - the demand is probably still larger than the reduction turboquant brings.
TurboQuant is known across the industry to not be state of the art. There are superior schemes for KV quant at every bitrate. Eg, SpectralQuant: https://github.com/Dynamis-Labs/spectralquant among many, many papers.
> Given that TurboQuant results in a 6x reduction in memory usage for KV caches
All depends on baseline. The "6x" is by stylistic comparison to a BF16 KV cache; not a state of the art 8 or 4 bit KV cache scheme.
Current "TurboQuant" implementations are about 3.8X-4.9X on compression (w/ the higher end taking some significant hits of GSM8K performance) and with about 80-100% baseline speed (no improvement, regression): https://github.com/vllm-project/vllm/pull/38479
For those not paying attention, it's probably worth sending this and ongoing discussion for vLLM https://github.com/vllm-project/vllm/issues/38171 and llama.cpp through your summarizer of choice - TurboQuant is fine, but not a magic bullet. Personally, I've been experimenting with DMS and I think it has a lot more promise and can be stacked with various quantization schemes.
The biggest savings in kvcache though is in improved model architecture. Gemma 4's SWA/global hybrid saves up to 10X kvcache, MLA/DSA (the latter that helps solve global attention compute) does as well, and using linear, SSM layers saves even more.
None of these reduce memory demand (Jevon's paradox, etc), though. Looking at my coding tools, I'm using about 10-15B cached tokens/mo currently (was 5-8B a couple months ago) and while I think I'm probably above average on the curve, I don't consider myself doing anything especially crazy and this year, between mainstream developers, and more and more agents, I don't think there's really any limit to the number of tokens that people will want to consume.
The work going into local models seems to be targeting lower RAM/VRAM which will definately help.
For example Gemma 4 32B, which you can run on an off-the-shelf laptop, is around the same or even higher intelligence level as the SOTA models from 2 years ago (e.g. gpt-4o). Probably by the time memory prices come down we will have something as smart as Opus 4.7 that can be run locally.
Bigger models of course have more embedded knowledge, but just knowing that they should make a tool call to do a web search can bypass a lot of that.
The net effect won’t be a memory use reduction to achieve the same thing. We’ll do more with the same amount of memory. Companies will increase the context windows of their offerings and people will use it.
I am not convinced that more context will be useful, practical use of current models at 1mil context window shows they get less effective as the window grows. Given model progress is slowing as well, perhaps we end up reaching a balance of context size and competency sooner than expected.
Stuff in more code. Stuff in more system prompt. Stuff in raw utf8 characters instead of tokens to fix strawberries. Stuff in WAY more reasoning steps.
Given the current tech, I also doubt there will be practical uses and I hope we’ll see the opposite of what I wrote. But given the current industry, I fully trust them so somehow fill their hardware.
Market history shows us than when the cost of something goes down, we do more with the same amount, not the same thing with less. But I deeply hope to be wrong here and the memory market will relax.
that will only increase the demand for RAM as models will now be usable in scenarios that weren't feasible prior, and the ceiling for model and context size is not even visible at this point
I hate to mention Jevons paradox as it has become cliche by now, but this is a textbook such scenario
I said it for many years that OS developers need to focus on over optimisations. If it wasnt a chip sgortage it would be the ever slowing progress on chip scaling.
But software optimisation helps all hardware and that doesnt drive sales.
Linux however, they dont have to worry about that. Maybe it is finally the era of Haiku OS as the ghost of BeOS rises!
I think RAM shortages would be the least of our problems…
Assuming China takes TSMC in one piece (unlikely without internal sabotage in the best case scenario), it would still probably take years before it produces another high end GPU or CPU.
We would probably be stuck with the existing inventory of equipment for a long time…
I am surprised we consider TSMC like a natural resource: isn't it really a combination of know-how and build-out according to that know-how? If smarts leave the country, perhaps this moves with them.
The risk with China taking over Taiwan is that they mostly expedite their own production research by a couple of years.
It kinda does resemble a natural resource though. The machines and technology in use at TSMC are so insanely complex, that there isn't a single person on earth who knows everything about how it works. TSMC functions only because of all of the pieces of the puzzle being together in the right place and arranged in just the right way. It's a very fragile balance that keeps it all running, and a major disruption could mean we get thrown back by a decade in chip-making technology.
What you say is absolutely true, and is a serious problem—but the way our system operates does not allow us to correct for it.
Anyone trying to spin up a competitor to TSMC would have to first overcome a significant financial hurdle: the capital investment to build all the industrial equipment needed for fabrication.
Then they'd have to convince institutions to choose them over TSMC when they're unproven, and likely objectively worse than TSMC, given that they would not have its decades of experience and process optimization.
This would be mitigated somewhat if our institutions had common-sense rules in place requiring multiple vendors for every part of their supply chain—note, not just "multiple bids, leading to picking a single vendor" but "multiple vendors actively supplying them at all times". But our system prioritizes efficiency over resiliency.
A wealthy nation-state with a sufficiently motivated voter base could certainly build up a meaningful competitor to TSMC over the course of, say, a decade or two (or three...). But it would require sustained investment at all levels—and not just investment in the simple financial sense; it requires people investing their time in education and research. Dedicating their lives to making the best chips in the world. And the only reason that would work is that it defies our system, and chooses to invest in plants that won't be finished for years, and then pay for chips that they know are inferior in quality, because they're our chips, and paying for them when they're lower quality is the only way to get them to be the best chips in the world.
The same thing everyone who's paying attention to the real world (and not the financial fantasy world) does: that OpenAI's purchase commitments are wildly unrealistic and unsustainable.
What’s the lose scenario for them? They’re basically a cartel, and you need ram irregardless. If they make less it’s still a cost:demand, just not the most optimal for them. They’ve done that math, and figure this is the best risk and reward for them. Your goodwill or opinion doesn’t matter to them, because you need them more than they need you.
I fear the author and most commenters are not aware of the law of demand and supply. If there is demand for consumer RAM, there will be supply for consumer RAM. It just takes time and risk-assessment to scale up operations.
We have RAM shortage now, we will have very cheap RAM tomorrow. It’s not like production is bottlenecked by raw materials. Chip companies just need to assess if the demand by AI companies will last so it’s better to scale up, or perhaps they should wait it out instead of oversupplying and cutting into their profits.
We're talking about advanced semiconductor manufacture. It takes years and 100s millions to billions of dollars to scale up operations. That's something you don't do unless you know there's demand to sustain it in future.
I'm personally hoping that one of the AI or data center companies is suddenly unable to pay for their bills and deflate the entire industry. Probably the only hope of things getting better before the 2030s.
I just checked my gaming PC I built a few years ago with 64GB of DDR5 RAM, its actually gone up in value, that is unheard of generally.
Think I will scrap my PC and sell its parts.
I wonder if there are any niche companies building decent rigs with DDR3 and 5/6th generation Intel CPUs out there, it is cheap and might be a business opportunity?
I work at an e-waste recycling company. I have several dozen trays of RAM in my inventory, ~90% of it DDR3. DDR3 was selling as of a month ago, but I haven't tried to sell any RAM since. I'm looking forward to doing a huge one this week.
This is simple extrapolation from current demand, nothing more. And that's a borderline silly analysis because it assumes the AI bubble won't burst. The great misadventure in the Persian Gulf probably accelerates that because we're almost certainly going to be facing a recession.
Another thing I've been thinking about is what happens when the next generation of NVidia chips comes out? I suspect NVidia is going to delay this to milk the current demand but at some point you'll be able to buy something that's better than the H100 or B200 or whatever the current state-of-the-art for half the price. And what's that going to do to the trillions in AI DC investment?
I'm interested when the next bump in DRAM chip density is coming. That's going to change things although it seems like much of production has moved from consumer DRAM chips to HBM chips. So maybe that won't help at all.
I do think that companies will start seeing little ot no return from billions spent on AI and that's going to be aproblem. I also think that the hudnreds of billions of capital expenditure of OpenAI is going to come crashing down as there just isn't any even theoretical future revenue that can pay for all that.
I fear that the real reason we do have a shortage, I mean, the real reason for the demand, is AI companies scooping what they can so that their competitors, whether existing or incumbent, can’t get to it.
I’m a bit of an optimist. I think this will smack the hands of developers who don’t manage RAM well and future apps will necessarily be more memory-efficient.
Using a lot less RAM often implies using more CPU, so even with inflated RAM prices, it's not a good tradeoff (at least not in general).
Only if the software is optimised for either in the first place.
Ton of software out there where optimisation of both memory and cpu has been pushed to the side because development hours is more costly than a bit of extra resource usage.
Or just using less electron and writing less shit code.
Ok so Samsung, SK Hynix and Micron do not have the capacity to meet demand. Also, what little capacity they do have they are allocating to HBM over DRAM. Based on my limited knowledge HBM can not be easily repurposed for consumer electronics. Translation: main street is cooked for the next 3-4 years.
It doesn't stop there though. OpenAI is currently mired in a capital crunch. Their last round just about sucked all the dry powder out of the private markets. Folks are now starting to ask difficult questions about their burn rate and revenue. It is increasingly looking like they might not commit to the purchase order they made which kick-started this whole panic over RAM.
Soo ... how sure are we that the memory makers themselves are not going to be the ones holding the bag?
Don’t the memory makers always get left holding the bag? I feel this has happened at least three times before.
DRAM and to a lesser degree storage are notorious for their feast and famine cycles
(Well that and collusion)
To add a more local hurdle as well, the Dutch power grid is at capacity and its managing company is now telling companies that planned to build a datacenter that they can't be connected to the grid until 2030, even though said companies already paid for and got guarantees about that connection.
That is, memory capacity is reserved for datacenters yet to be built, but this will do weird things if said datacenter construction is postponed or cancelled altogether.
> the Dutch power grid is at capacity and its managing company is now telling companies that planned to build a datacenter that they can't be connected to the grid until 2030, even though said companies already paid for and got guarantees about that connection.
Are the Netherlands a large proportion of global datacenters?
Amsterdam hosts a major internet exchange. It's not a bad place to build a datacenter and there are many. Northern latitude brings free air cooling, but also additional distance to clients. Lots of peers in AMS-IX, but not a lot of oceananic cable landings (one with two paths to the US, but most of the submarine cables land nearby in Europe)
Yes. Amsterdam has one of the largest IXPs (AMS-IX) in Europe and is also one of the largest European markets for Internet Infrastructure services (i.e. hosting, DNS provision, domain name registration, etc.)
Is that relevant? The grid in every country is getting ridiculously stressed by datacenters.
What the grid looks like in different countries is very different. The Dutch power grid is already almost 50% renewables, which is an inconvenience for adding capacity because that's around where you have to start really dealing with storage in order to add more.
In most other places the percentage is significantly less than that and then you can easily add more of the cheap-but-intermittent stuff because a cloudy day only requires you to make up a 10% shortfall instead of a 50% one, which existing hydro or natural gas plants can handle without new storage when are more of them to begin with.
That guarantee is not as much of a guarantee as stated in the media. You get a guarantee it will be planned at a certain time (as in looked at), not that it will be build. The cost of doing business is taking risks and mitigating them. There is a reason the nuclear plant in Borsele was build: an aluminium smelter. Maybe you should arrange for something similar as a datacenter (no politician will fall on a sword for that but you can try). The (original) power draw is about the same 80-100MW.
The Radeon VII came out in 2019 as a $700 consumer GPU with an 1TB/s HBM2 memory subsystem which is more than any consumer GPU you can get today, including the high-end ones afaik. At that point in time, there was a whole lineup of AMD GPUs with HBM going down into the midrange.
If they could make this stuff and sell it to regular people a decade ago for very palatable prices, why do they come up with the idea that this is the technology of the gods, unaffordable by mere mortals?
I have been wondering this recently. It was the convention that if you wanted to keep costs down, try to keep the memory bus size down as low as possible. Still remember the awful Radeon 9200 SE - 64bit data bus that strangled an already slow GPU.
Heck, I have a phone with a 16bit memory bus for instance. The high(ish) clock rate only makes up the difference slightly.
But with general prices on all components going up, it might not be such a big factor any more.
HBM migght make sense for higher end products which can free up space for the lower end that will never use the tech.
I was gonna say, I still use an AMD Vega that uses HBM2.
> 1TB/s HBM2 memory subsystem which is more than any consumer GPU you can get today
5090 has 1.8 TB/s?
That card only had 16GB of memory; its memory bandwidth was 1TB/s.
The Pro variant had 32GB, I had one in a 2019 Mac Pro
You're saying this in a world where AMD's highest end consumer GPU in 2026 is also limited to 16 GB.
RX7900 XTX has 24GB
The market is already stagnated. Even if OpenAI doesn’t buy what they reserved other players will do so. SK Hynix CEO said there is a 20% gap between supply and demand per year. And that doesn’t account the shock effect that will take place the moment prices normalize and everyone and their dog will go out and start buying inventory to avoid the next crisis. I for one would certainly buy more than I currently need just in case.
What kind of consumer electronics can you build with HBM? That's the startup you should be founding...
AMD has built some consumer GPUs in the recent past with HBM - RX Vega and Radeon VII (although I assume not all "HBM" is created equal).
My vega 56 still has 400gb/s of memory which is still insane for how old the card is.
AMD's Hawaii architecture had 320GB/s on a 512b GDDR5 bus in 2013.
The Fiji XT architecture after it had 512GB/S on a 4096b HBM bus in 2015.
The Vega architecture did have 400GB/s or so in 2017, which was a bit of a downgrade.
But wouldn’t you rather hbm prices come down first ? Memory makers will be fine. There is practically infinite demand. Unless you get china style rationing of compute per person world wide.
The real issue is everyone wanting to upgrade to hbm, ddr5, and nvme5 at the same time.
I am betting the pendulum swings faster to the other side to excess capacity as all the construction lies of Altman fall through with financiers waking up the the fact they can't build the infrasctructure as fast nor make any profits on that infrastructure that will get built.
> Soo ... how sure are we that the memory makers themselves are not going to be the ones holding the bag?
We aren't. The remaining memory manufacturers fear getting caught in a "pork cycle" yet again - that is why there's only the three large ones left anyway.
If they don't expand capacity much, the only negative consequences I foresee happening for them is that they might lose spending discipline, and that systems will be set up to make do with a little less memory. Apart from that, it's just very high profits followed by more or less regular profits.
They could wind up losing all their business to China though.
China has memory makers who are creeping up through the stages of production maturity, and once they hit then there's no going back.
If the existing makers can't meet supply such that Chinese exports get their foot in the door, they may find they never get ahead again due to volume - that domestic market is huge so they have scale, and the gaming market isn't going to care because they get anything at the moment, which is all you'll need for enterprise to say "are we really afraid of memory in this business?"
Memory makers did get themselves into this situation by selling all wafers for empty promises and alienating everyone but OpenAI tbh. I do hope they end up holding the bag once again, cause after covid and the cartel thing they don't seem to ever learn their lesson on how to have the tiniest amount of integrity.
> Memory makers did get themselves into this situation by selling all wafers for empty promises and alienating everyone but OpenAI tbh.
Wasn't the problem here that OpenAI was negotiating with Samsung and SK Hynix at the same time without the other one knowing about it? People only realized the implications when they announced both deals at once.
While we're giving away bags, I'd like HDD manufacturers to get some too.
That wouldn’t help if another one goes bankrupt that’ll only make things worse.
Sounds like they're too big to fail, maybe we should bail them out to reinforce that they will get rewarded for making bad decisions.
Permanent public ownership of (very large stakes in) these companies doesn't seem like such a bad idea anymore, does it? It's what we used to have for most of the 20th century at least in Europe.
Good point. I think both AI companies and hardware makers should pay for the damage they caused to us here.
They act as a de-facto monopoly and milk us. Why is this allowed?
It's a business with huge up-front capital expenses and typically very low margins. Supply is scaling up slowly because it's hard, and if you overshoot, you go out of business.
Nobody is "allowing" this. It's a natural property of being both advanced technology and a commodity at the same time.
The strange deals on the entire future output are what was allowed. Try to do the same thing with onions and the government understands you are a criminal.
https://en.wikipedia.org/wiki/Onion_Futures_Act
Because for the last 60 years we've allowed big business to buy and hollow out our legal and education systems.
Allowed? We live in a neoliberal world where corporate monopolies / oligopolies aren’t even remotely regulated. If you try to do even the gentlest regulation of companies people scream about communism and totalitarianism. Unless the regulation serves the monopolies by making it harder to enter the market.
It started with raegan, and even parties on the “left” in the west believe in it with very few exceptions.
> We live in a neoliberal world where corporate monopolies / oligopolies aren’t even remotely regulated. If you try to do even the gentlest regulation of companies people scream about communism and totalitarianism. Unless the regulation serves the monopolies by making it harder to enter the market.
The thing that enables this is pretty obvious. The population is divided into two camps, the first of which holds the heuristic that regulations are "communism and totalitarianism" and this camp is used to prevent e.g. antitrust rules/enforcement. The second camp holds the heuristic that companies need to be aggressively "regulated" and this camp is used to create/sustain rules making it harder to enter the market.
The problem is that ordinary people don't have the resources to dive into the details of any given proposal but the companies do. So what we need is a simple heuristic for ordinary people to distinguish them: Make the majority of "regulations" apply only to companies with more than 20% market share. No one is allowed to dump industrial waste in the river but only dominant companies have bureaucratic reporting requirements etc. Allow private lawsuits against dominant companies for certain offenses but only government-initiated prosecutions against smaller ones, the latter preventing incumbents from miring new challengers in litigation and requiring proof beyond a reasonable doubt.
This even makes logical sense, because most of the rules are attempts to mitigate an uncompetitive market, so applying them to new entrants or markets with >5 competitors is more likely to be deleterious, i.e. drive further consolidation. Whereas if the market is already consolidated then the thicket of rules constrains the incumbents from abusing their dominance in the uncompetitive market while encouraging new entrants who are below the threshold.
> Soo ... how sure are we that the memory makers themselves are not going to be the ones holding the bag?
I hope they do, they did not have to agree to sell so much RAM to one customer. They’ve been caught colluding and price fixing more than once, I hope they take it in the shorts and new competitors arise or they go bankrupt and new management takes over the existing plants.
Don’t put all your eggs in the one basket is how the old saying goes.
This will result in demand destruction which will starve the enterprise which will starve the hyperscaler. theres no situation where people not being able to afford hardware for 4 years results in the bubble not popping
The people who fucked over consumers are left holding the back that they sold us out over?
Oh no!
They won't be, prices are high because they are refusing to build capacity for demand that may evaporated by the time they are done. They are holding back and building only enough so when the bubble pops they will be fine.
So the ML hate is weaponized in the form of memory demand collapse FUD, and the public at large has to pay through their nose for it... thanks party poopers!
I don't think its from the ML collapse FUD, its most likely from the multiple time's in the past when they overbuilt and it resulted in a memory oversupply and price collapses. The 1985–1988, 1993–1994, 1998–2002 and the post pandemic oversupply. These were all cases where shortages followed by over corrections caused oversupply, financial losses due to low prices and fewer surviving companies. I think they're taking their time and are cautiously adding more capacity in such a way that prices won't end up collapsing again. Regardless, the result is still that we the consumers have to pay more.
At this point the remaining memory companies are… the ones that didn’t die during an over-supply collapse, right? I guess there’s been a strong evolutionary pressure against giving consumers what we want, haha.
its not like all the RAM is passing the same machine, they can gradually increase machines and observe the change in demand, and smoothly match it.
If they gradually increase production capacity then prices stay high for 10+ years (or for as long as it takes for demand to crash) because a gradual increase in production takes that long for them to add enough capacity for current demand.
If they add enough capacity to meet current demand quickly then when demand crashes they still have billions of dollars in loans used to build capacity for demand that no longer exists and then they go bankrupt.
I would expect that OpenAI gets as much money as they ask for for the next 10 years.
There’s virtually infinite capital: if needed, more can be reallocated from the federal government (funded with debt), from public companies (funded with people’s retirement funds), from people’s pockets via wealth redistribution upwards, from offshore investment.
They will be allowed to strangle any part of the supply chain they want.
I'm guessing they become pets.com within the year. At least I hope.
China already has a well developed DRAM industry, as DRAM is somewhat easier than logic, and can tolerate a much higher defect rate. The industry will figure this out.
Another point is I often see the money argument - like country X has more money, so they can afford to do more and better R&D, make more stuff.
This stuff comes out of factories, that need to be built, the machinery procured, engineers trained and hired.
If China capitalises on the big three focusing on data centre team, the big three might have a very hard time post bubble
I think the article has a giant blind spot as far as China is concerned , considering they have already a mature enough memory ecosystem via YMTC that Apple was considering sourcing from them. As well as continued expansion in the DRAM and HBM Fabs [1]. It feels like the memory cartel once again trying to incentivise their various govt to cough up some more tax breaks/funding to cushion the AI buildout bet that they made and the bubble seeming about to pop. In any case if they leave the consumer market underserved it should be no surprise if before that 2030 prediction we are all on cheaper YMTC memory modules.
[1]https://www.tomshardware.com/tech-industry/semiconductors/ym...
Maybe if they had no competitors...
I think you're massively overestimating how much money is really accessible here. The parent comment's right that all of the easily available VC & private equity investment is basically used up. OpenAI was struggling to sell $600M of private equity, the big multi-billion dollar investment packages had lots of conditions and non-cash in it.
> more can be reallocated from the federal government (funded with debt)
While this is the most reliable funding, it's still not very accessible. OpenAI is a money pit, and their demands are growing quickly. The US government has started a bunch of very expensive spending. If OpenAI were to require yearly bundles of it's recent "$120B" deal, that's 6% of the US' discretionary budget. 12.5% of the non-military discretionary budget. (And the military is going to ask for a lot more money this year) Even the idea of just issuing more debt is dubious because they're going to want to do that to pay for the wars that are rapidly spiralling out of control.
None of this is saying that the US government can't or wouldn't pay for it, but it's non trivial and it's unclear how much Altman can threaten the US government "give me a trillion dollars or the economy explodes" without consequences.
Further deficit-spending isn't without it's risks for the US government either. Interests rates are already creeping up, and a careless explosion of deficit may well trigger a debt crisis.
> from public companies (funded with people’s retirement funds)
This would be at great cost. OpenAI would need to open up about it's financial performance to go public itself. With it's CFO being put on what is effectively Administrative Leave for pushing against going public, we can assume the financials are so catastrophic an IPO might bomb and take the company down with it. Nobody's going to be investing privately in a company that has no public takers.
Getting money through other companies is also running into limits. Big Tech has deep pockets but they've already started slowing down, switching to debt to finance AI investment, and similarly are increasingly pressured by their own shareholders to show results.
> from people’s pockets via wealth redistribution upwards
The practical mechanism of this is "AI companies raise their prices". That might also just crash the bubble if demand evaporates. For all the hype, the productivity benefit hasn't really shown up in economy-wide aggregates. The moment AI becomes "expensive", all the casual users will drop it. And the non-casual users are likely to follow. The idea of "AI tokens" as a job perk is cute, but exceedingly few are going to accept lower salary in order to use AI at their job.
There's simply not much money to take out of people's pockets these days, with how high cost of living has gotten.
> from offshore investment.
This is a pretty good source of money. The wealthy Arabian oil states have very deep slush funds, extensively investing in AI to get ties to US businesses and in the hope of diversifying their resource economies.
...
...
"Was". Was a good source of money.
I'm genuinely curious to find out how many billions they get every year from now.
love that theres virtually infinite capital there. meanwhile in the rest of the world there is virtually no food.
are you kidding? if spent all that money on food you guys would just use it to bullshit all day and make funny pictures, while if we spend it on AI..
I'm a bit surprised the article makes no mention of Google's TurboQuant[0] introduced 26 days prior.
Given that TurboQuant results in a 6x reduction in memory usage for KV caches and up to 8x boost in speed, this optimization is already showing up in llama.cpp, enabling significantly bigger contexts without having to run a smaller model to fit it all in memory.
Some people thought it might significantly improve the RAM situation, though I remain a bit skeptical - the demand is probably still larger than the reduction turboquant brings.
[0] https://news.ycombinator.com/item?id=47513475
TurboQuant is known across the industry to not be state of the art. There are superior schemes for KV quant at every bitrate. Eg, SpectralQuant: https://github.com/Dynamis-Labs/spectralquant among many, many papers.
> Given that TurboQuant results in a 6x reduction in memory usage for KV caches
All depends on baseline. The "6x" is by stylistic comparison to a BF16 KV cache; not a state of the art 8 or 4 bit KV cache scheme.
BTW, a number of corrections. The TurboQuant paper was submitted to Arxiv back in April 2025: https://arxiv.org/abs/2504.19874
Current "TurboQuant" implementations are about 3.8X-4.9X on compression (w/ the higher end taking some significant hits of GSM8K performance) and with about 80-100% baseline speed (no improvement, regression): https://github.com/vllm-project/vllm/pull/38479
For those not paying attention, it's probably worth sending this and ongoing discussion for vLLM https://github.com/vllm-project/vllm/issues/38171 and llama.cpp through your summarizer of choice - TurboQuant is fine, but not a magic bullet. Personally, I've been experimenting with DMS and I think it has a lot more promise and can be stacked with various quantization schemes.
The biggest savings in kvcache though is in improved model architecture. Gemma 4's SWA/global hybrid saves up to 10X kvcache, MLA/DSA (the latter that helps solve global attention compute) does as well, and using linear, SSM layers saves even more.
None of these reduce memory demand (Jevon's paradox, etc), though. Looking at my coding tools, I'm using about 10-15B cached tokens/mo currently (was 5-8B a couple months ago) and while I think I'm probably above average on the curve, I don't consider myself doing anything especially crazy and this year, between mainstream developers, and more and more agents, I don't think there's really any limit to the number of tokens that people will want to consume.
The work going into local models seems to be targeting lower RAM/VRAM which will definately help.
For example Gemma 4 32B, which you can run on an off-the-shelf laptop, is around the same or even higher intelligence level as the SOTA models from 2 years ago (e.g. gpt-4o). Probably by the time memory prices come down we will have something as smart as Opus 4.7 that can be run locally.
Bigger models of course have more embedded knowledge, but just knowing that they should make a tool call to do a web search can bypass a lot of that.
The net effect won’t be a memory use reduction to achieve the same thing. We’ll do more with the same amount of memory. Companies will increase the context windows of their offerings and people will use it.
That is the sad reality of the future of memory.
I am not convinced that more context will be useful, practical use of current models at 1mil context window shows they get less effective as the window grows. Given model progress is slowing as well, perhaps we end up reaching a balance of context size and competency sooner than expected.
Stuff in more code. Stuff in more system prompt. Stuff in raw utf8 characters instead of tokens to fix strawberries. Stuff in WAY more reasoning steps.
Given the current tech, I also doubt there will be practical uses and I hope we’ll see the opposite of what I wrote. But given the current industry, I fully trust them so somehow fill their hardware.
Market history shows us than when the cost of something goes down, we do more with the same amount, not the same thing with less. But I deeply hope to be wrong here and the memory market will relax.
You still need to hold the model in memory. If you have for example 16 GB ram, the gains aren't that much
That's not what consumes the most memory at scale. The KV caches are per-user.
You can still use as much memory, but fit more things into it, so I don’t think the current market hogs will let go easily.
that will only increase the demand for RAM as models will now be usable in scenarios that weren't feasible prior, and the ceiling for model and context size is not even visible at this point
I hate to mention Jevons paradox as it has become cliche by now, but this is a textbook such scenario
I'm a bit surprised the article makes no mention of China's new memory companies.
[0] https://techwireasia.com/2026/04/chinese-memory-chips-ymtc-c...
The era of optimisation is finally here. I'm excited.
I said it for many years that OS developers need to focus on over optimisations. If it wasnt a chip sgortage it would be the ever slowing progress on chip scaling.
But software optimisation helps all hardware and that doesnt drive sales.
Linux however, they dont have to worry about that. Maybe it is finally the era of Haiku OS as the ghost of BeOS rises!
Wait until China invades Taiwan.. (ok, it's not too likely, but what if?)
I think RAM shortages would be the least of our problems…
Assuming China takes TSMC in one piece (unlikely without internal sabotage in the best case scenario), it would still probably take years before it produces another high end GPU or CPU.
We would probably be stuck with the existing inventory of equipment for a long time…
I am surprised we consider TSMC like a natural resource: isn't it really a combination of know-how and build-out according to that know-how? If smarts leave the country, perhaps this moves with them.
The risk with China taking over Taiwan is that they mostly expedite their own production research by a couple of years.
It kinda does resemble a natural resource though. The machines and technology in use at TSMC are so insanely complex, that there isn't a single person on earth who knows everything about how it works. TSMC functions only because of all of the pieces of the puzzle being together in the right place and arranged in just the right way. It's a very fragile balance that keeps it all running, and a major disruption could mean we get thrown back by a decade in chip-making technology.
What you say is absolutely true, and is a serious problem—but the way our system operates does not allow us to correct for it.
Anyone trying to spin up a competitor to TSMC would have to first overcome a significant financial hurdle: the capital investment to build all the industrial equipment needed for fabrication.
Then they'd have to convince institutions to choose them over TSMC when they're unproven, and likely objectively worse than TSMC, given that they would not have its decades of experience and process optimization.
This would be mitigated somewhat if our institutions had common-sense rules in place requiring multiple vendors for every part of their supply chain—note, not just "multiple bids, leading to picking a single vendor" but "multiple vendors actively supplying them at all times". But our system prioritizes efficiency over resiliency.
A wealthy nation-state with a sufficiently motivated voter base could certainly build up a meaningful competitor to TSMC over the course of, say, a decade or two (or three...). But it would require sustained investment at all levels—and not just investment in the simple financial sense; it requires people investing their time in education and research. Dedicating their lives to making the best chips in the world. And the only reason that would work is that it defies our system, and chooses to invest in plants that won't be finished for years, and then pay for chips that they know are inferior in quality, because they're our chips, and paying for them when they're lower quality is the only way to get them to be the best chips in the world.
China is 10 years into what you describe, no?
> the way our system operates
They have the other system.
the scientists will switch sides with minimal issues, like they did after WWII
It seems that RAM manufacturers are still reluctant to increase production. They know something that investors don't about long term RAM demands?
The same thing everyone who's paying attention to the real world (and not the financial fantasy world) does: that OpenAI's purchase commitments are wildly unrealistic and unsustainable.
What’s the lose scenario for them? They’re basically a cartel, and you need ram irregardless. If they make less it’s still a cost:demand, just not the most optimal for them. They’ve done that math, and figure this is the best risk and reward for them. Your goodwill or opinion doesn’t matter to them, because you need them more than they need you.
> They’re basically a cartel,
The lawsuits in the past prove that statement to not be basically but actually.
I want those AI companies that drove the prices up, to pay an immediate back-tax to all of us.
I don't want to pay more because of AI companies driving the price up. That is milking.
I fear the author and most commenters are not aware of the law of demand and supply. If there is demand for consumer RAM, there will be supply for consumer RAM. It just takes time and risk-assessment to scale up operations.
We have RAM shortage now, we will have very cheap RAM tomorrow. It’s not like production is bottlenecked by raw materials. Chip companies just need to assess if the demand by AI companies will last so it’s better to scale up, or perhaps they should wait it out instead of oversupplying and cutting into their profits.
We're talking about advanced semiconductor manufacture. It takes years and 100s millions to billions of dollars to scale up operations. That's something you don't do unless you know there's demand to sustain it in future.
The law of supply and demand works in a perfect competition market.
There are two RAM suppliers...
If only we have not allowed oligopolies to exist. Meanwhile, EU is not in the race at all and US has very few fabs.
I'm personally hoping that one of the AI or data center companies is suddenly unable to pay for their bills and deflate the entire industry. Probably the only hope of things getting better before the 2030s.
That’s likely to happen if all the talks about OpenAI pulling out of their wafer deals are true.
I just checked my gaming PC I built a few years ago with 64GB of DDR5 RAM, its actually gone up in value, that is unheard of generally.
Think I will scrap my PC and sell its parts.
I wonder if there are any niche companies building decent rigs with DDR3 and 5/6th generation Intel CPUs out there, it is cheap and might be a business opportunity?
I am still running a DDR3 2nd Gen i7. 32GB RAM, it is surprisingly comfortable but I also dont push it too hard.
I work at an e-waste recycling company. I have several dozen trays of RAM in my inventory, ~90% of it DDR3. DDR3 was selling as of a month ago, but I haven't tried to sell any RAM since. I'm looking forward to doing a huge one this week.
Thank god they shut down 3D XPoint.
This is simple extrapolation from current demand, nothing more. And that's a borderline silly analysis because it assumes the AI bubble won't burst. The great misadventure in the Persian Gulf probably accelerates that because we're almost certainly going to be facing a recession.
Another thing I've been thinking about is what happens when the next generation of NVidia chips comes out? I suspect NVidia is going to delay this to milk the current demand but at some point you'll be able to buy something that's better than the H100 or B200 or whatever the current state-of-the-art for half the price. And what's that going to do to the trillions in AI DC investment?
I'm interested when the next bump in DRAM chip density is coming. That's going to change things although it seems like much of production has moved from consumer DRAM chips to HBM chips. So maybe that won't help at all.
I do think that companies will start seeing little ot no return from billions spent on AI and that's going to be aproblem. I also think that the hudnreds of billions of capital expenditure of OpenAI is going to come crashing down as there just isn't any even theoretical future revenue that can pay for all that.
can't read the article due to a paywall.
https://archive.is/QGU89
here is the source: https://www.reuters.com/world/asia-pacific/south-koreas-sk-g...
I fear that the real reason we do have a shortage, I mean, the real reason for the demand, is AI companies scooping what they can so that their competitors, whether existing or incumbent, can’t get to it.
This was one of the theories behind the wafer buyout by OpenAI indeed. Pretty efficient way to make everyone panic and cut off of new hardware.
Was it debubked in any way (e.g. by OpenAI actually showing what they do with the wafers?)
Fabricated shortage to fasten US Chip Act and US Chip Security Act
Expect shortages across the board. RAM? That's the tip of the iceberg, think food and gas.