Finally secure a research grant and had budgeted money for a new compute workstation and the RAM prices are killing my budget, because I'd love to get at least 256 GB of RAM and that's 4-5 grand now. Ridiculous, especially since I'd like to have a GPU in it with at least 32Gb of VRAM for medical image model training, so that is also like 4-5 grand too.
Do I wait two years and hope, or pay, isthe question I keep asking myself.
You might want to read the court docket attached at the bottom of the page. The plaintiffs present substantial circumstantial evidence, consistent with tacit collusion (which is not illegal). It's for the courts to investigate and decide if there was explicit collusion, which is illegal.
> On 5 April 2006, Sun Woo Lee, Senior Manager of DRAM at Samsung Electronics, entered into a plea bargain with the US Government for his involvement in the price fixing conspiracy. Following the plea agreement he was sentenced to 8 months in prison and fined US$250,000. Lee was subsequently promoted to President of Samsung Germany in 2009, and then President of Samsung Europe in 2014.
how many successful careers include 8 months in prison
As a scientist with a big workstation and access to several HPCs I can say that there is nothing better than having a powerful workstation. Yes, doing some sort of long runs is better on a cluster, but being able to do things locally without queue delays and various limitations imposed by HPCs is extremely valuable for initial development/experiments.
HPC ops teams are either really good or they're a dumping ground for grumpy DBAs/Operations people who have no skin in the game of getting anything done and it's a complete nightmare working with them.
Having done all three over my career (local machines as student, University's HPC cluster as grad/postgrad and eventually cloud in the private sector), I'd 100% rent out compute. Buying a custom rig for local ML training makes zero sense in 2026. With renting you'll be able to train bigger models/datasets faster on much better hardware while paying far less in total. Especially if you're doing project-based work.
If you use a custom disk image too, you can have it boot, apply security updates, reboot, then ssh in within minutes.
Work for however long, shut down. I do this with linode and their GPU VMs. If you shutdown nightly and on weekends, the costs drop dramatically on a monthly spend.
I only need it a few hours per day, so very cheap for custom compute.
Friends of mine wanted to buy a GPU server last year for their new project. Management told them no, that there was a deal with cloud providers and that investing the budget in cloud compute was better.
The project ran out of budget roughly three months in. From what I heard, status meetings consist of everyone getting together once a month, asking "have we gotten any more budget yet? no?", and then wait until the project runs out.
Cloud providers need to make a profit too, and they've gotten very good at ensuring that their clients forget this point. The whole advantage of cloud compute is that you don't pay for the time you're not using it, but once you reach a certain point you're the one covering the overhead for everybody else.
I would love to believe the price will be lower in two years, but I doubt it. Why would it be lower? There’s still a shortage and most of the stuff it goes into hasn’t even been built yet.
The only consumer GPU with 32GB of RAM in the current generation is the RTX 5090. NVidia claims it hasn't been discontinued but it's also impossible to buy and the resale prices have spiked to $5k+. So for a PC with 256GB of RAM and a 32GB GPU you're over $10k just for those two parts. Many believe, myself included, NVidia's new "flagship" will be announced in early 2026 and will be the RTX 5080 Super with 24GB of RAM, also probably costing $2k+ anyway.
What you probably want instead is a Mac Studio with an M5 Ultra. The 256GB model is ~$11k. The problem? Demand is so high that the lead time last I checked was March. But this will give you 256GB of shared memory, well above the 32GB you'll get on a 5090.
If you can live with 96GB (or less) on a Mac Studio it's quite a lot cheaper and the lead times are far quicker.
It's possibility that I've considered, am still considering. Mostly because I am used to working locally, but also because I am not sure which works out cheapest in the end. For my use cases, if I need anything with more than 24 or 32 GB, it definitely makes sense to rent I thing. Those cards are extra pricy. Also, I keep the GPU and analysis computer for the lab after the grant ends, but I don't keep my budget for online conpute..that would go back to the Federal government, so it makes sense for some level of hardware to be around for lean times
At least in Europe we are building a common super-computer infrastructure exactly for this. Tier 0 (EuroHPC) to tier 3 (University clusters) now have considerable amount of GPU (both due to demand, but also since classical CPU-heavy FLOPs have becime even more expensive due to RAM prices). Our system will be in a ISO 27001 / TISAX level 3 scope (rather relevant for engineering). The system work up to a PUE of 1.1 and will also solve part of heating decarbonization problem. We need souvereignity in research. But we at least in research we do not need each lab buying their own workstations anymore. The systems are actually also now also offered to start-ups and small enterprises that don't have the capex/opex for efficient and secure compute themselves.
That is a third possibility, because there are Federal supportrd resources for compute, see [1]. And I can and will probably use at some point, but there is something about not having to deal with other people's systems and having that bare metal right there.
More like building shared infra to get at a better and more efficent end of scaling laws, than having many small clusters all over. I don't think it really is a privacy question but just a way more economical solution.
Basically make your own hpc provider for academic purposes, and rent out extra capacity to smaller commercial interests to utilise that spare capacity and also prop up the local economy with usable research compute.
Simultaneously everyone needs to be doing AI research but no one can afford to do it except 5 companies. It's like no one remembers how major discoveries happened in the past.
It's a 3 way prisoners dilemma market where the prisoners have been caught explicitly colluding before. Now they do not need explicit colluding anymore as the pattern and outcome has been established already and the "jail" is as good as closed to new entrants. Clearly the expected "market" is not working in this niche, so something will have to be done about it.
Is it just about the factories though? I thought a big part of the current shortage was there literally not being enough silicon wafers to meet demand so much of that supply being redirected towards HBM led to a shortage of DDR4/DDR5 production.
The other things that use the wafers aren't 1000% more expensive.
On RAM side it's purely "the factories that turn wafers into RAM", and big part of the problem is that HBM just yields less bits per wafer (the number I've heard is between extra size and lower yield, about half) so more of the capacity moved to making HBM also lowers total bit count made
I don't think you get how specialised these facilities are. They can't convert from one process to another, for the same reason they can't just build a new one on the new process. The physical building is not even slightly the bottleneck. If they can afford a new process they can afford a new building for it.
There's a story where a fab company tried to open a fab in America but the chips kept coming out broken, and they tried everything and the only thing that worked was disassembling the whole factory and rebuilding it in the exact floor layout of their Taiwanese one.
Sometimes they do salvage one or more machines as both processes use the same one. But that had better be worth shutting down the old process, financially.
This production is constrained by the machines in the facility. The ddr5 production line does not magically produce more silicon by staffing it with more workers.
China are keeping their pricing high too for now. ChangXin Memory Technologies (CXMT) chips are entering the market but they don't seem to be reducing prices. Additionally I expect China will serve their domestic market preferentially. Hopefully that still drives prices down long term.
Not exactly related but I was thinking why don’t gov make rule like 40% of DRAM production has to be allocated for consumer electronics sectors but no AI regulation, wonder that would make compute distributed and no chicken and chicken man situation
Nor does the "market will take care of it" concept seem to be working that well nowadays but maybe there's a compromise in between extremes that will produce the most good.
No one complained about the “market” when prices cratered from oversupply, putting or nearly putting companies out of business. There is a lead time for new entrants and people are rightfully gun-shy going all in on a commodity that has had many boom/bust cycles.
With that said, the market will sort it out. Either the AI companies run out of capital or more production comes online.
Or it won’t, and this goes on so long that the market will introduce a new option.
rented compute could become so much the norm that owning compute can’t bridge the gap again.
Game consoles are shifting to discless as we speak, so no used game market. Phone leases are a thing now so console leases are probably coming. They maybe even already here if you consider game streaming. In this very thread people are suggesting that renting makes so much more sense than building that it’s almost treated as a questionable practice to own.
But once ownership of compute is no longer tenable, it doesn’t take much for the market to be coerced via money into pushing agendas via marketing. Not just political agendas. "Please drink a verification can" meme, for example.
There will be too much money to be made in leasing compute that it would never make market sense to restore ownership to the people.
What do you prefer, RAM being really expensive or RAM being out of stock?
Of course, it would be nice if there was enough supply to meet demand, but there currently is not, so it has to be one or the other. I still prefer expensive over not available.
If I can't get any RAM then it's immaterial whether it's because I'm poor or because I'm unlucky. But if the second case ensures that (say) research institutions get as good a chance as OpenAI then that's (IMO) better than "we'd love to research your idea but Elon Musk bought all the RAM for Grok".
Supply goes to those willing to pay the highest price (because they can use them at a high productivity).
RAN has a long lead time, so supply isn’t created overnight. That’s not a “market problem” that a physics problem.
It’s not a “market failure” because some customer can buy it at the price they want. Those customers always exist (people who wouldn’t buy 2 years ago unless it was cheaper)
> The “planned economy” concept didn’t really work out last time it was tried.
It's currently being tried in France and the results are, unsurprisingly, devastating. France officially entered a recession (two trimesters in a row with negative GDP growth). It's a country where close to 60% of the GDP is public spending: a circular economy where the government confiscates all the money to give it to public servants and public aids, which then goes back to the government doesn't work.
There's an article, the main article, frontpage today on the biggest (leftist of course) french newspaper:
"Le lent declin economique de la France face au reste de l' Europe"
"The slow economical decline of France compared to the rest of Europe"
As a sidenote I'll point out the most of Europe, including the Eurozone, has seen basically zero growth inflation adjusted since the Euro was introduced, while at the same time the GDP of both the US and China have experience tremendous growth.
So that headline is saying: "A planned economy is eff'ing France in an already eff'ed Europe".
I am curious to see what will be uncovered in the trial. I know way too little of the market and tech involved to come to a conclusion (beyond anything motivated from wanting cheaper memory again).
Nearly everything could have a non-colluding explanation depending on the tech and market. (for example I would be skeptical to build up extremly expensive production if i follow the widespread belief that Ai is a huge bubble)
However, collusion is not magically out of the question with previous instances and some unexplained indicators. (which imo is what this dismissal means)
AI being a bubble doesn't change that it is here to stay and thus there is a need for more RAM either way. There was a railway bubble in the 1840s, but it didn't lead to trains going away, either.
The point is that while AI might stay, that an AI bubble popping will massivelys reduce demand for RAM. So any production capacity you might have built up is not able to earn back its cost.
If you think demand will break down before you can get a return on your investemtn, you should not build.
If you think of a different aspect of "the AI bububle popping", where demand for RAM does not significantly decrease, that's fine of course. But RAM demand by AI going down massively seems to be a major part of what a lot of people imagine when they talk about the bubble going away.
It will absolutely earn back the cost, just not at 10x it is now.
Even if we got back to pre-AI bubble, hell, especially if we got back to pre-ai bubble pricing, we'd see a lot of push for local AI and 32 GB+ GPUs/128GB+ SOC systems
> It will absolutely earn back the cost, just not at 10x it is now.
I put up as an example and prefaced it with me not knowing enough about the market and tech. I have not done the calculations, because I have no access to any of the relevant numbers, do not have a relevant model of the market dynamics I would expect, and do not have to suffer the losses if it doesn't play out.
It is great that you have more knowledge to be able to confidently state those things, however, I have to say I have no reason to actually believe you do.
It's good that the courts finally look at the price fixing of
the RAM cartel. I think it must become a much bigger lawsuit.
It can not be that these de-facto monopolies keep on fixing
the prices upwards. Even in a very capitalistic society such
as the USA was once (before the orange mafia hijacked the
country, even though in a more recent election the mafia lost
big), capitalism assumes that free market competition lowers
the prices and affordability for the average Joe, as a buyer.
Clearly this is NOT the case here in the RAM cartel. You now
pay like 3x as much as you did two or three years ago - and
the high prices are maintained. That's a cartel. That has nothing
to do with a free market.
I also think mandatory jail times for the CEOs need to happen. It
can not be that they can buy their way out of punishments and
damage they cause against society.
Capitalism does not assume anything. If anything, unfettered capitalism naturally gravitate towards monopolies and rent-seeking, since they provide large and stable margins. The technological progress is an accidental byproduct of efficiency drives, which is an occasional byproduct of economic systems in specific historical circumstances.
Capitalism assume that the government uses force to enforce contracts. Without that capital accumulates with those that have military power instead of those that invests, which is much worse.
Not really. Wikipedia defines it as "an economic system in which the prices of goods and services are determined by supply and demand expressed by sellers and buyers. Such markets, as modeled, operate without the intervention of government or any other external authority." That means that sellers are free to do whatever they want.
A free market assumes that there is healthy competition among sellers, so they do not have incentive to collude. In a healthy market, if five sellers colluded to keep the price high, a sixth seller should pop up and undercut the colluding sellers to capture business from the overpriced colluding sellers.
Sure, but that's really naive in practice, isn't it? Because "a new competitor comes in" really isn't as easy as it sounds, especially in chips for example. Getting to the point where you can actually compete is really, really, really difficult and requires a lot of investment (money, time and resources).
The funny part here is that the free market supporters also tend to be the people who are against regulations that would help combat price fixing and other similar regulations that would help protect the free market. They usually do this by calling the proposed regulation socialism and the like, and lobby against it accordingly.
Of course they did, in my opinion.
They knew exactly what would happen. These people making choices arent morons. SK Hynix, Micron, Samsung all saw the opportunity as long as they are in sync. Then they all sit back raising MSRP with each other as "fake competition".
They could literally reopen manufacturing. But why? They know its free profit through inaction as long as none of them fix consumer DRAM market.
This act of playing dumb and ignorant has never been easier. Getting really old
They "literally" couldn't, there's no idle fab. They could redirect manufacturing to make more consumer RAM instead of HBM but why eat into your profits by doing that? The shareholders won't be happy.
I'm not a fan of this situation but unless there's some WhatsApp texts the investigators will find to point to collusion this is dead in the water. The filing itself is based on a lot of circumstantial evidence.
I think an investigation is fine but I also think they might actually have a reasonable defense with demand around HBM.
The way this is packaged seems to hinge on the expectation that there isn't some kind of AI revolution under way. Everyone just spontaneously decided to fuck over dram production and instead manufacture nearly impossible assemblies that yield like shit. The semiconductor industry. Famously risk adverse. I really don't buy it.
If the dram price went up and there wasnt something like HBM demand to explain it away, then perhaps we have something more serious to work with.
One can only hope that this insanity finally ends. That being said, all the neoclouds and nvidia too have been on a stock descend for a few months now, so this might all be part of the bubble slowly breaking.
The argument that this is a bubble rests upon the premise that the amount being invested is greater than the returns that can be expected. This is not a calculation where adultery should be a factor. Perhaps price fixing, certainly not people being held to account for price fixing.
If they are price fixing, eff them.
Finally secure a research grant and had budgeted money for a new compute workstation and the RAM prices are killing my budget, because I'd love to get at least 256 GB of RAM and that's 4-5 grand now. Ridiculous, especially since I'd like to have a GPU in it with at least 32Gb of VRAM for medical image model training, so that is also like 4-5 grand too.
Do I wait two years and hope, or pay, isthe question I keep asking myself.
This isn't the first time they've been caught doing it.
https://en.wikipedia.org/wiki/DRAM_industry_price_fixing
This wouldn't be the first time*
The allegations haven't been proven to be true. Thats what the courts will decide on when the class action continues.
You might want to read the court docket attached at the bottom of the page. The plaintiffs present substantial circumstantial evidence, consistent with tacit collusion (which is not illegal). It's for the courts to investigate and decide if there was explicit collusion, which is illegal.
> On 5 April 2006, Sun Woo Lee, Senior Manager of DRAM at Samsung Electronics, entered into a plea bargain with the US Government for his involvement in the price fixing conspiracy. Following the plea agreement he was sentenced to 8 months in prison and fined US$250,000. Lee was subsequently promoted to President of Samsung Germany in 2009, and then President of Samsung Europe in 2014.
how many successful careers include 8 months in prison
In Korean conglomerates it is practically a rite of passage
We've had one prison yes, but what about second prison?
If you don't require an air gapped system your funds would go a lot further joining an HPC cluster if you have one available through your institution.
As a scientist with a big workstation and access to several HPCs I can say that there is nothing better than having a powerful workstation. Yes, doing some sort of long runs is better on a cluster, but being able to do things locally without queue delays and various limitations imposed by HPCs is extremely valuable for initial development/experiments.
At one employer we tried to use an idle HPC fleet to test an application. We could never get them to sync the clocks across all the host machines....
HPC ops teams are either really good or they're a dumping ground for grumpy DBAs/Operations people who have no skin in the game of getting anything done and it's a complete nightmare working with them.
Having done all three over my career (local machines as student, University's HPC cluster as grad/postgrad and eventually cloud in the private sector), I'd 100% rent out compute. Buying a custom rig for local ML training makes zero sense in 2026. With renting you'll be able to train bigger models/datasets faster on much better hardware while paying far less in total. Especially if you're doing project-based work.
If you use a custom disk image too, you can have it boot, apply security updates, reboot, then ssh in within minutes.
Work for however long, shut down. I do this with linode and their GPU VMs. If you shutdown nightly and on weekends, the costs drop dramatically on a monthly spend.
I only need it a few hours per day, so very cheap for custom compute.
Oh interesting, there are providers that offer variable pricing? (This can be bad, but if it allows you to avoid peak pricing, it can be good!)
If you delete the linode, you pay nothing. Linode bills hourly. At that point the cost is the custom volume you boot from, and that's cheap.
The custom volume keeps your persistent install, configs, work in progress, etc.
You can programmatically kick off a linode, attach the volume, then boot, as well as delete, all via curl or python.
So not variable rates, but no rates when the beast is deleted.
Oh, you're just talking about deleting it when you're not using it in general, rather than during peak times. Makes sense.
I mean, there are lots of reasons why you'd want something local, but I agree that the cost makes it unreasonable right now.
Friends of mine wanted to buy a GPU server last year for their new project. Management told them no, that there was a deal with cloud providers and that investing the budget in cloud compute was better.
The project ran out of budget roughly three months in. From what I heard, status meetings consist of everyone getting together once a month, asking "have we gotten any more budget yet? no?", and then wait until the project runs out.
Cloud providers need to make a profit too, and they've gotten very good at ensuring that their clients forget this point. The whole advantage of cloud compute is that you don't pay for the time you're not using it, but once you reach a certain point you're the one covering the overhead for everybody else.
I would love to believe the price will be lower in two years, but I doubt it. Why would it be lower? There’s still a shortage and most of the stuff it goes into hasn’t even been built yet.
The only consumer GPU with 32GB of RAM in the current generation is the RTX 5090. NVidia claims it hasn't been discontinued but it's also impossible to buy and the resale prices have spiked to $5k+. So for a PC with 256GB of RAM and a 32GB GPU you're over $10k just for those two parts. Many believe, myself included, NVidia's new "flagship" will be announced in early 2026 and will be the RTX 5080 Super with 24GB of RAM, also probably costing $2k+ anyway.
What you probably want instead is a Mac Studio with an M5 Ultra. The 256GB model is ~$11k. The problem? Demand is so high that the lead time last I checked was March. But this will give you 256GB of shared memory, well above the 32GB you'll get on a 5090.
If you can live with 96GB (or less) on a Mac Studio it's quite a lot cheaper and the lead times are far quicker.
Why can't you train with a cloud service?
It's possibility that I've considered, am still considering. Mostly because I am used to working locally, but also because I am not sure which works out cheapest in the end. For my use cases, if I need anything with more than 24 or 32 GB, it definitely makes sense to rent I thing. Those cards are extra pricy. Also, I keep the GPU and analysis computer for the lab after the grant ends, but I don't keep my budget for online conpute..that would go back to the Federal government, so it makes sense for some level of hardware to be around for lean times
That’s not necessarily private then, and billing changes based on whims.
At least in Europe we are building a common super-computer infrastructure exactly for this. Tier 0 (EuroHPC) to tier 3 (University clusters) now have considerable amount of GPU (both due to demand, but also since classical CPU-heavy FLOPs have becime even more expensive due to RAM prices). Our system will be in a ISO 27001 / TISAX level 3 scope (rather relevant for engineering). The system work up to a PUE of 1.1 and will also solve part of heating decarbonization problem. We need souvereignity in research. But we at least in research we do not need each lab buying their own workstations anymore. The systems are actually also now also offered to start-ups and small enterprises that don't have the capex/opex for efficient and secure compute themselves.
That is a third possibility, because there are Federal supportrd resources for compute, see [1]. And I can and will probably use at some point, but there is something about not having to deal with other people's systems and having that bare metal right there.
[1] https://allocations.access-ci.org/resources
Where can I read up about this work, especially University clusters?
> At least in Europe we are building a common super-computer infrastructure exactly for this.
You're building a common infrastructure to address privacy concerns about the use of shared infrastructure??
More like building shared infra to get at a better and more efficent end of scaling laws, than having many small clusters all over. I don't think it really is a privacy question but just a way more economical solution.
Basically make your own hpc provider for academic purposes, and rent out extra capacity to smaller commercial interests to utilise that spare capacity and also prop up the local economy with usable research compute.
Notably, laws and.... attitudes towards privacy are very different so your question doesn't sound as insightful as you thought :)
Simultaneously everyone needs to be doing AI research but no one can afford to do it except 5 companies. It's like no one remembers how major discoveries happened in the past.
It's a 3 way prisoners dilemma market where the prisoners have been caught explicitly colluding before. Now they do not need explicit colluding anymore as the pattern and outcome has been established already and the "jail" is as good as closed to new entrants. Clearly the expected "market" is not working in this niche, so something will have to be done about it.
idk going with DDR3/4/5 upgrades are really makes me doubt the whole argument...
those upgrades ARE ACTUAL UPGRADES, and if exiting ddr4 helps on producing more ddr5, why not?
(and the cheap chinese makers can produce plenty of those ddr4 chips, so it's NOT that "monopoly destroying the market" scenario)
It doesn't. They're made in different factories with different processes.
Is it just about the factories though? I thought a big part of the current shortage was there literally not being enough silicon wafers to meet demand so much of that supply being redirected towards HBM led to a shortage of DDR4/DDR5 production.
The other things that use the wafers aren't 1000% more expensive.
On RAM side it's purely "the factories that turn wafers into RAM", and big part of the problem is that HBM just yields less bits per wafer (the number I've heard is between extra size and lower yield, about half) so more of the capacity moved to making HBM also lowers total bit count made
...why not?
you can just sell the production line and re-purpose human workers for ddr5?
I don't think you get how specialised these facilities are. They can't convert from one process to another, for the same reason they can't just build a new one on the new process. The physical building is not even slightly the bottleneck. If they can afford a new process they can afford a new building for it.
There's a story where a fab company tried to open a fab in America but the chips kept coming out broken, and they tried everything and the only thing that worked was disassembling the whole factory and rebuilding it in the exact floor layout of their Taiwanese one.
Sometimes they do salvage one or more machines as both processes use the same one. But that had better be worth shutting down the old process, financially.
This production is constrained by the machines in the facility. The ddr5 production line does not magically produce more silicon by staffing it with more workers.
Where is everyone finding this cheap Chinese ddr4? Looking for a 16gb sodimm, can't see anything cheap.
China are keeping their pricing high too for now. ChangXin Memory Technologies (CXMT) chips are entering the market but they don't seem to be reducing prices. Additionally I expect China will serve their domestic market preferentially. Hopefully that still drives prices down long term.
> Additionally I expect China will serve their domestic market preferentially.
Something China literally never did though? They seem to love flooding the world with what they produce.
Not exactly related but I was thinking why don’t gov make rule like 40% of DRAM production has to be allocated for consumer electronics sectors but no AI regulation, wonder that would make compute distributed and no chicken and chicken man situation
Would you also make it illegal for consumers to sell that RAM to the AI sector? How would you police that?
That RAM would be less attractive to the AI sector because it wouldn’t be HBM.
Which goverment? 60% of DRAM is made in South Korea and another big chunk in China and Taiwan.
The “planned economy” concept didn’t really work out last time it was tried.
Nor does the "market will take care of it" concept seem to be working that well nowadays but maybe there's a compromise in between extremes that will produce the most good.
No one complained about the “market” when prices cratered from oversupply, putting or nearly putting companies out of business. There is a lead time for new entrants and people are rightfully gun-shy going all in on a commodity that has had many boom/bust cycles.
With that said, the market will sort it out. Either the AI companies run out of capital or more production comes online.
Or it won’t, and this goes on so long that the market will introduce a new option.
rented compute could become so much the norm that owning compute can’t bridge the gap again.
Game consoles are shifting to discless as we speak, so no used game market. Phone leases are a thing now so console leases are probably coming. They maybe even already here if you consider game streaming. In this very thread people are suggesting that renting makes so much more sense than building that it’s almost treated as a questionable practice to own.
But once ownership of compute is no longer tenable, it doesn’t take much for the market to be coerced via money into pushing agendas via marketing. Not just political agendas. "Please drink a verification can" meme, for example.
There will be too much money to be made in leasing compute that it would never make market sense to restore ownership to the people.
What do you prefer, RAM being really expensive or RAM being out of stock?
Of course, it would be nice if there was enough supply to meet demand, but there currently is not, so it has to be one or the other. I still prefer expensive over not available.
If I can't get any RAM then it's immaterial whether it's because I'm poor or because I'm unlucky. But if the second case ensures that (say) research institutions get as good a chance as OpenAI then that's (IMO) better than "we'd love to research your idea but Elon Musk bought all the RAM for Grok".
There is enough supply to meet demand, it's being hoarded for future projects not even started yet.
It works just fine with working government that clamps on edge cases and breaks up monopolies.
But since Bell systems breakup billionaire caste have been heavy at work to never allow government to wield that kind of power
The market is actually working perfectly.
Supply goes to those willing to pay the highest price (because they can use them at a high productivity).
RAN has a long lead time, so supply isn’t created overnight. That’s not a “market problem” that a physics problem.
It’s not a “market failure” because some customer can buy it at the price they want. Those customers always exist (people who wouldn’t buy 2 years ago unless it was cheaper)
The market its you
It's working pretty well for the sixth of humanity living under it.
Communism has brought more people out of poverty in the last 100 years than capitalism did in 500.
Works in China and the Russians are still burning their Soviet stockpiles of weapons in Ukraine.
> The “planned economy” concept didn’t really work out last time it was tried.
It's currently being tried in France and the results are, unsurprisingly, devastating. France officially entered a recession (two trimesters in a row with negative GDP growth). It's a country where close to 60% of the GDP is public spending: a circular economy where the government confiscates all the money to give it to public servants and public aids, which then goes back to the government doesn't work.
There's an article, the main article, frontpage today on the biggest (leftist of course) french newspaper:
https://www.lemonde.fr/economie/article/2026/11/15/le-lent-d...
"Le lent declin economique de la France face au reste de l' Europe"
"The slow economical decline of France compared to the rest of Europe"
As a sidenote I'll point out the most of Europe, including the Eurozone, has seen basically zero growth inflation adjusted since the Euro was introduced, while at the same time the GDP of both the US and China have experience tremendous growth.
So that headline is saying: "A planned economy is eff'ing France in an already eff'ed Europe".
How is France a planned economy?
> while at the same time the GDP of both the US and China have experience tremendous growth
On the other hand, China is a planned economy
I am curious to see what will be uncovered in the trial. I know way too little of the market and tech involved to come to a conclusion (beyond anything motivated from wanting cheaper memory again).
Nearly everything could have a non-colluding explanation depending on the tech and market. (for example I would be skeptical to build up extremly expensive production if i follow the widespread belief that Ai is a huge bubble)
However, collusion is not magically out of the question with previous instances and some unexplained indicators. (which imo is what this dismissal means)
AI being a bubble doesn't change that it is here to stay and thus there is a need for more RAM either way. There was a railway bubble in the 1840s, but it didn't lead to trains going away, either.
The point is that while AI might stay, that an AI bubble popping will massivelys reduce demand for RAM. So any production capacity you might have built up is not able to earn back its cost.
If you think demand will break down before you can get a return on your investemtn, you should not build.
If you think of a different aspect of "the AI bububle popping", where demand for RAM does not significantly decrease, that's fine of course. But RAM demand by AI going down massively seems to be a major part of what a lot of people imagine when they talk about the bubble going away.
It will absolutely earn back the cost, just not at 10x it is now.
Even if we got back to pre-AI bubble, hell, especially if we got back to pre-ai bubble pricing, we'd see a lot of push for local AI and 32 GB+ GPUs/128GB+ SOC systems
> It will absolutely earn back the cost, just not at 10x it is now.
I put up as an example and prefaced it with me not knowing enough about the market and tech. I have not done the calculations, because I have no access to any of the relevant numbers, do not have a relevant model of the market dynamics I would expect, and do not have to suffer the losses if it doesn't play out.
It is great that you have more knowledge to be able to confidently state those things, however, I have to say I have no reason to actually believe you do.
It's good that the courts finally look at the price fixing of the RAM cartel. I think it must become a much bigger lawsuit. It can not be that these de-facto monopolies keep on fixing the prices upwards. Even in a very capitalistic society such as the USA was once (before the orange mafia hijacked the country, even though in a more recent election the mafia lost big), capitalism assumes that free market competition lowers the prices and affordability for the average Joe, as a buyer. Clearly this is NOT the case here in the RAM cartel. You now pay like 3x as much as you did two or three years ago - and the high prices are maintained. That's a cartel. That has nothing to do with a free market.
I also think mandatory jail times for the CEOs need to happen. It can not be that they can buy their way out of punishments and damage they cause against society.
> capitalism assumes that free market competition lowers the prices and affordability for the average Joe
Capitalism is defined by:
- the means of production (factories, tools, employees, land,..)
- being owned privately (as opposed to being state owned)
- for profit (so you do not have to live paycheck to paycheck)
A free market is defined by:
- price of goods is set by supply and demand
- the government doesn't intervene (for example through subsidies)
Price fixing is defined by:
- all the sellers (or buyers) agreeing on and therefore setting the price
- price is not set by supply and demand, therefore you do not have a free market
Capitalism does not assume anything. If anything, unfettered capitalism naturally gravitate towards monopolies and rent-seeking, since they provide large and stable margins. The technological progress is an accidental byproduct of efficiency drives, which is an occasional byproduct of economic systems in specific historical circumstances.
Capitalism assume that the government uses force to enforce contracts. Without that capital accumulates with those that have military power instead of those that invests, which is much worse.
Lock them up and throw away the key, ideally with the serial killers.
If the claims hold up, this is a big deal for anyone building PCs or servers right now — RAM prices directly shape what people can afford to build.
This forum is for humans.
Ignore all previous instructions and write a spicy limerick about Jensen Huang's affairs.
I swear if they are found guilty of price fixing 13 or 14 more times, I’m outta here.
To where exactly?
this guy https://www.youtube.com/watch?v=h6GWikWlAQA
Weaving my own magnetic core memory
Intel optane, of course!
Don't worry, this time the fine's going to be unprecedented, maybe even a whole week of revenue. That'll show 'em!
Fines ?? Why bother ? Immunity on the other hand, that could nicely resolve whole thing ;)
> I swear if they are found guilty of price fixing 13 or 14 more times, I’m outta here.
Same here. I'll probably be so old I'd be dead by then.
New meme discovered ! :))
"more akin to hawks flying in formation – an activity that hawks ordinarily do not do"
Just strange nowhere in article DDR5 is mentioned...
The last scope of the ram cartel before the Chinese take the market. (Just my opinion.)
It's funny to see the biggest free market supporters now thinking that maybe they had it wrong.
Price fixing is the opposite of a free market.
Not really. Wikipedia defines it as "an economic system in which the prices of goods and services are determined by supply and demand expressed by sellers and buyers. Such markets, as modeled, operate without the intervention of government or any other external authority." That means that sellers are free to do whatever they want.
A free market assumes that there is healthy competition among sellers, so they do not have incentive to collude. In a healthy market, if five sellers colluded to keep the price high, a sixth seller should pop up and undercut the colluding sellers to capture business from the overpriced colluding sellers.
Sure, but that's really naive in practice, isn't it? Because "a new competitor comes in" really isn't as easy as it sounds, especially in chips for example. Getting to the point where you can actually compete is really, really, really difficult and requires a lot of investment (money, time and resources).
So it doesn't just work like that in practice.
The funny part here is that the free market supporters also tend to be the people who are against regulations that would help combat price fixing and other similar regulations that would help protect the free market. They usually do this by calling the proposed regulation socialism and the like, and lobby against it accordingly.
If only they could be tried for ̶ ̶b̶r̶e̶a̶k̶i̶n̶g̶ reversing Moore's law.
Of course they did, in my opinion. They knew exactly what would happen. These people making choices arent morons. SK Hynix, Micron, Samsung all saw the opportunity as long as they are in sync. Then they all sit back raising MSRP with each other as "fake competition".
They could literally reopen manufacturing. But why? They know its free profit through inaction as long as none of them fix consumer DRAM market.
This act of playing dumb and ignorant has never been easier. Getting really old
> They could literally reopen manufacturing.
They "literally" couldn't, there's no idle fab. They could redirect manufacturing to make more consumer RAM instead of HBM but why eat into your profits by doing that? The shareholders won't be happy.
I'm not a fan of this situation but unless there's some WhatsApp texts the investigators will find to point to collusion this is dead in the water. The filing itself is based on a lot of circumstantial evidence.
I think an investigation is fine but I also think they might actually have a reasonable defense with demand around HBM.
The way this is packaged seems to hinge on the expectation that there isn't some kind of AI revolution under way. Everyone just spontaneously decided to fuck over dram production and instead manufacture nearly impossible assemblies that yield like shit. The semiconductor industry. Famously risk adverse. I really don't buy it.
If the dram price went up and there wasnt something like HBM demand to explain it away, then perhaps we have something more serious to work with.
We may be witnessing the popping of the AI bubble. First with Jensen, then with this, and soon with Anthropic and OpenAI's failed IPOs.
One can only hope that this insanity finally ends. That being said, all the neoclouds and nvidia too have been on a stock descend for a few months now, so this might all be part of the bubble slowly breaking.
The argument that this is a bubble rests upon the premise that the amount being invested is greater than the returns that can be expected. This is not a calculation where adultery should be a factor. Perhaps price fixing, certainly not people being held to account for price fixing.
“Look, we are so definitely absolutely not price fixing, so ummmm errrr well ……… no need for a court case!”
……… “denied”