RAM Is So Expensive, Samsung Won't Even Sell It To Samsung (pcworld.com) 87
A severe spike in global DRAM prices has pushed Samsung Semiconductor to refuse a long-term RAM order from its own sibling, Samsung Electronics. The move is forcing the smartphone division into short, expensive renegotiations, which will likely mean higher costs for consumer devices. PCWorld reports: Samsung subsidiaries are, naturally, going to look to Samsung Semiconductor first when they need parts. Such was reportedly the case for Samsung Electronics, in search of memory supplies for its newest smartphones as the company ramps up production for 2026 flagship designs. But with so much RAM hardware going into new "AI" data centers -- and those companies willing to pay top dollar for their hardware -- memory manufacturers like Samsung, SK Hynix, and Micron are prioritizing data center suppliers to maximize profits.
The end result, according to a report from SE Daily spotted by SamMobile, is that Samsung Semiconductor rejected the original order for smartphone DRAM chips from Samsung Electronics' Mobile Experience division. The smartphone manufacturing arm of the company had hoped to nail down pricing and supply for another year. But reports say that due to "chipflation," the phone-making division must renegotiate quarterly, with a long-term supply deal rejected by its corporate sibling. A short-term deal, with higher prices, was reportedly hammered out.
The end result, according to a report from SE Daily spotted by SamMobile, is that Samsung Semiconductor rejected the original order for smartphone DRAM chips from Samsung Electronics' Mobile Experience division. The smartphone manufacturing arm of the company had hoped to nail down pricing and supply for another year. But reports say that due to "chipflation," the phone-making division must renegotiate quarterly, with a long-term supply deal rejected by its corporate sibling. A short-term deal, with higher prices, was reportedly hammered out.
AI (Score:2, Insightful)
Macroeconomics 101 (Score:4, Insightful)
There will be a correction. There is ALWAYS a correction.
Re: (Score:1)
Right... And my 5090 didn't cost 2200 still. Any minute now they'll be giving them away in cereal boxes.
Re: (Score:2)
Re:Macroeconomics 101 (Score:5, Informative)
Be careful there. Lots of AI is being put to silly, useless, or unreasonable uses. OTOH, lots of it is being put to extremely productive uses. (OK, 20% improvement in output, but also an increase in expenses.)
ISTM, that PART of the AI hoopla is a bubble. Possibly much more than half. But the other half is not a bubble, and is growing rapidly. What the collapse will look like depends in part on how much the productive segment grows relative to the other part before it happens.
Re: (Score:2)
I do not think "productivity" is what is driving this. I think it is intelligence agencies using AI to go through half a century of collected data to determine who will be the winners and losers in society. "Society" itself is 'Fascist' and wants every social interaction to happen in an expected order.
I do not know how to explain this idea without writing an entire book... but, ignoring Reality will end in Reality reminding 'you' that it can not be ignored. This 'ignorance of Reality' frequently ends up in
Re: (Score:3)
The massive amount of capital put into AI is not going to yield results. There will be mostly losers and only a few winners. This is assuming AI sticks around and is the paradigm shift that many people believe it will be.
If it's entirely fake, just a bunch of nonsense that people wasted money on, then it will pop quickly. But if it's somewhat real, like the dot-com boom. We'll see society transformed, and a lot of failed businesses, some of them very stupid, scattered along the information superhighway.
If t
Re: (Score:2)
The massive amount of capital put into AI is not going to yield results.
To clarify my own statement. I mean not going to yield results proportional to the amount put into them. You can double the investment and not get double the return. I think this is reflected in the wacky P/E ratios we see in the market for AI related stocks.
Re: (Score:3)
This is not a bubble. This is going to be the way of things for a while. The economic activity AI is producing is going exponentially, and the inputs are trivial vs the future productivity. These are the facts. DRAM, GPUs and the infrastructure to power them just got slammed with a huge demand, which WILL drive up the cost. Thats not inflation, thats normal supply/demand curves coming into balance. Inflation is when all things rise equally. Here, these components are rising faster than the overall cost of things in the marketplace (though there is going to be spillover as many non-involved things have to pay for the increases to maintain their requisite supply (smartphones in this case).
AI investment is growing exponentially and has been for some time. Meanwhile, AI productivity is already slowing, and shows further signs that as we move forward it will slow still more. While tech companies are attempting to shovel more of it at the businesses and end-users that rely on them, there's only so much the current LLM driven systems are going to be able to accomplish without some massive change in basic operation that, thus far, doesn't appear to be happening. And while they continue to rake in
Re: (Score:2)
Re: Macroeconomics 101 (Score:2)
Are you saying internet/tech stocks weren't a bubble in 1999/2000 because they were the way of the future?
Re: Macroeconomics 101 (Score:3)
This is not a bubble. This is going to be the way of things for a while. The economic activity AI is producing is going exponentially
That's nice and all, but where are the returns? If they don't start coming soon, the whole thing is going to run out of gas.
Pets.com was doing really well, lots of advertising too, until it became clear that nobody was buying anything from it. Building it doesn't necessarily mean they'll come, but the advertising industry made bank. Lots of economic activity around the gold rush too, problem is there wasn't any gold.
Construction companies and hardware vendors are doing well, but after a while people stop bu
Re: (Score:2)
No, inflation is an expansion of the money supply. When the government prints more money than the population growth requires, banks lend more money at lower rates, people and businesses borrow more, and the surplus money chasing the same goods and services increases the demand and prices rise. That is economic inflation, my term, which is different from what laymen call inflation: some prices going up, as from tariffs or other taxes.
Re: (Score:2)
It's absolutely a bubble. All these buildouts and nobody finding anything AI is useful for.
The crash is going to result in some half-completed buildouts and the servers being sold for pennies on the dollar.
Re: (Score:2)
From 1998 through 2000, we had a bubble that was caused by generic business people insisting they be involved in the tech boom. It popped in 2000, bringing a lot of misery with it. The AI bubble is the same thing, you have people with ZERO understanding of AI trying to get into it JUST so they can have an AI company, get all that money from an IPO, then cash out when their company that doesn't have a real product collapses.
At the base of any bubble is real potential, but when clueless people with a lot
Re: (Score:2)
This most certainly *is* a bubble. Over 90% of US GDP growth is *nothing* but data center buildouts. There was the story here on /. a week or two ago, of two *completed* datacenters in Nvidia's home county... that are sitting there, useless, because there's no electricity to run them.
And circular support - Nvidia and Anthropic and is it M$? is not a good sign... esp. when Anthropic is still operating at a loss, and is going to need 10 times the current revenue is a few years...
Re: (Score:2)
I like to help.
Hey, remember when the x090's only cost about a grand? You know, back before bitcoin when an x080 was around $800. Someday, I'm going to find the guy who invented bitcoin and I'm going to punch him in the face several times. Then I'm go
Re: (Score:2)
5090 isn't a bubble.That's just a low volume, high margin niche product for people with a lot of money.
Re: (Score:2)
The 5090 is the only RTX 5000 series GPU that is better than the previous generation, just because it is "larger". The only real design improvements in RTX 5000 over 4000 series was in AI related areas(DLSS). That is one big reason the Radeon 9070XT has been doing well, because NVIDIA spent too much effort on AI and didn't really improve rendering performance for the RTX 5000 series.
NVIDIA has to realize that if AMD releases UDNA in 2026, it's going to be a BIG performance boost on the AMD side, so NVID
Re: (Score:3)
There will be a correction. There is ALWAYS a correction.
To the article? Yes there was... apparently it is not true at all [sammobile.com]
Re: (Score:2)
It's no different from you buying something for personal use but deciding to sell it to someone else who would offer you more money than what you paid for it instead of
Re: (Score:2)
There is nothing new about that at all.
I work for a company that would much rather sell product to end customers at full price than to keep a portion of our limited production output to ourselves for no revenue at all. It probably shouldn't surprise many people that we would rather keep customers happy by delivering on the promised schedule while taking in a lot of money from them for the privilege, and finding other ways to satisfy the internal need, such as sharing with other teams that might already hav
Free market at work (Score:2)
Sure, DIY PC builds will be a thing of the past, but that is the price of freedom right?
Re: (Score:2)
Sure, DIY PC builds will be a thing of the past, but that is the price of freedom right?
Wish we could say a couple of decades of leftover RAM is gonna sell like fucking Charizard one day, but I kinda doubt it.
Re:Free market at work (Score:5, Funny)
Sure, DIY PC builds will be a thing of the past, but that is the price of freedom right?
32 bit OS's coming back because no one will be able to afford 4GB :-)
Re: Free market at work (Score:1)
what percent of people make diy pc out of all desktop, laptop, corporate pc? not a "freedom" if useless to almost all humans. A luxury ...
Maybe (Score:5, Interesting)
Maybe this will me be good. If the software people had to start caring about resource consumption again, instead of just thinking "hey lets use Electron" no reason my stupid IM can't have 4gigs of resident memory, after users will just buy more RAM!
Re: (Score:2)
Re: (Score:2)
Fair enough - it is easy to forget just how much real functionality there actually is in these stacks. It is nice to live in a world where a handful of lines of glue code yield a rich application.
However, there is a lot of stuff that does not *need* all that and generally isnt worth the trade off for many/most users. There is also the reality that all that to frequently gets delivered in the laziest way possible. Rather than a few shared libs that the OS could map into multiple virtual address spaces, we g
Re: Maybe (Score:2)
That's what indexing and string interning is for. You know, hashmaps and stuff.
Re: (Score:2)
I wouldn't hold my breath ...
My view is: that battle has been lost, for a long time now ... ...
Bloat is here to stay
Don't get me wrong, I wish you were right though.
time for the AI bubble to burst (Score:3)
On a more serious note, the DDR5 modules I got for 45 last year June are now at 290, what I got for 55 is now 480. And it's all out of stock...
But, as someone here mentioned in the comments of the last memory story, the fact there's no capacity buildout does mean that the common view is that the bubble will pop within the next few years.
Re:time for the AI bubble to burst (Score:5, Interesting)
Re: (Score:3)
Re:Hollywood Accounting Mk. 2 (Score:5, Interesting)
Samsung has famously not allowed their subsidiaries to subsidize each other. If the price of their own memory is not competitive, the memory that goes into their other products shall be bought from outside. The memory subsidiary, like any other, has to be able to stand on it's own, otherwise you end up with dead weight everywhere.
This is now the flip side of that. The market conditions are unpredictable, so they won't make long term commitments, and if our their electronics cannot pay the going rate of their memory, they shall not piss their potential earnings into the wind either.
I would call maybe some internal protectionism is in order. Their consumer electronics won't be able to source memory from anywhere else either, might as well close shop. But they are pretty much the only consumer electronics company that could deliver based on their own supply. Could use it to wipe the floor with competition...
Who knows, maybe the higher ups have not weighed in yet. And let's not forget Samsung like any other chaebol is heavily intertwined with SK government, too. Whether it's owned by the Lee family or SK govt, and/or is the SK govt owned by Samsung instead... these are questions which depend a lot on who is asking, when, and in what context... The current memory bubble is unprecedented in semiconductor history, and there is no playbook to follow. But consumer electronics are pretty much completely priced out of the memory market, and we can expect all sorts of weird stuff to follow.
And it's all because of LLMs (Score:2)
The AI craze is causing so much damage on so many fronts. It could end society as we know it, in ways that have nothing to do with Science Fiction scenarios such as taking over Earth and supplanting Man.
AI is causing disruption of markets, loss of jobs, negative mental health effects, increases in AGW, and on and on. And increasingly, it's being used to make propaganda more pervasive and more effective. By not pushing back harder against these trends, the 99% of us are voluntarily putting the yokes designed
Re: (Score:2)
It seems like less than a week ago that the /. experts were telling us the only thing gating AI was gigawatts.
Re: (Score:2)
You left out the robots. Enslaved to do what when the robots are cheaper?
Re: (Score:2)
Re: (Score:2)
Short sighted pricing vs. long term relationships (Score:2)
When COVID subsided, their stock crashed 70%, as not only did they lose the
It's the Shoe Event Horizon (Score:2)
... except instead of shoes [fandom.com] becoming the only profitable product to manufacture, it's chatbots. Nobody knows why, but when it's all over, the only survivors will be those who evolved into computer-illiterate deaf-mutes.
This can't be the right way to run Samsung (Score:2)
Re: (Score:1)
I don't know how their structured sounds more like a parent holding company with subsidiaries that are their own legal entities not just divisions/departments.
So they probably independently have their own CEOs. The folks running the holding company though might very well be asking, well why would we not want each sub to make itself as profitable as it can be.
They only reason to step in is if/when Samsung Electronics is actually endangered in terms of market share. If they have to design around cheaper slow
Ram is wasted by the vibe coder generation (Score:2, Offtopic)
Re: (Score:2)
Given that a large percentage of the current AI investment is "me too" investment that will fail, then hop
Re: (Score:2)
LLM's eat RAM... (Score:5, Informative)
Most people don't realize how much of a pig LLM's actually are. All the focus is on the big AI machines PSU's, multi kW fan boards, GPU's, etc... Most of them have maybe 8 2.5" disk slots. They're not storage beasts. But what they do have is HBM processors with multiple Tb's of registered ECC DDR5. The OS boots locally, and then the RAM receives the training data from a remote data lake. It gets loaded into RAM and stays there. Multiple Tb's per node.
T
As an alternative (Score:2)
Re: (Score:2)
Market has gone crazy (Score:2)
So, AI firms are gobbling up ram to power AI-services that most consumers do not want (apart from making the occasional meme or whatever), which is at the same time hurting consumers by increasing ram pricing for everyone.
So it's a double whammy. This would be like ice cream companies monopolising drinking water to make, say, pee-flavoured slushies.
That explains it ? (Score:2)
Workstations (Score:2)
vast demand for AI (Score:5, Interesting)
I see comments here claiming it's a bubble and most people don't want it, but the numbers say otherwise. OpenAI claims to have about 700 million active weekly customers. Anthropic is forecasting "$70B in revenue by 2028" and I believe it.
https://techcrunch.com/2025/11... [techcrunch.com]
Google's Gemini 3 is incredibly good in my experience, as is Anthropic's Claude Opus 4.5. Google has been forced to rate-limit access to Gemini 3 so as not to be overwhelmed.
I use these products every day for coding assistance and they are amazing. The difference in quality of the models compared to this time last year is stunning. But I suspect there is a limit to what these 'conventional' LLM's will be able to do with the current architectures. Throwing exponentially more hardware at the problem will have diminishing returns. Considerable research is happening to try to get good results while using fewer resources, so it could be that this will eventually put a ceiling on RAM consumption.
Re: (Score:2)
By your uid number I'm guessing you missed the Netscape bubble. I recall asking engineers why would netscape be so valuable if they lose money? They'd just shrug it off and say because. I also recall a finance gu
Re: (Score:2)
>> Are they 700M unique users? How many are paying?
Good question. The numbers may be fudged but it looks like unique users;
https://www.cnbc.com/2025/08/0... [cnbc.com]
"700 million weekly active users, with usage growing 4X year over year."
"OpenAI now counts five million paying business users, up from three million in June"
As for being extremely popular while losing money, that's the commonplace tech trajectory now. You spend whatever investment money it takes to achieve market dominance, and then it's gravy trai
Re: (Score:2)
You spend whatever investment money it takes to achieve market dominance, and then it's gravy train.
Historically speaking, that last part is frequently elusive. Lots of crashing and burning of once "sure thing" tech companies. Some have survived, but it's hardly an assumption that can be made.
Re: (Score:2)
>> Lots of crashing and burning
Very true, but it's a 'big bet' scenario where the winners get most of the jackpot. This has happened over and over in tech. And I think most people who don't use the AI products on a daily basis are not aware of how capable they already are. I pay double to use the best, most recent models in my work and they are far and away superior to the offerings from just 6 months ago. In a few months these current best models will be merely commonplace.
Re: (Score:2)
Every AI query should be billed at the actual build-out and electricity consumption price instead of hiding the enormous waste of resources within everyone's electric
Re: (Score:2)
>> 70 billion is nothing compared to the investment costs
True but the trend line is way up. They need to spend that money to capture market share or Google will get it.
As of August "OpenAI now counts five million paying business users, up from three million in June, as enterprises and educators embrace artificial intelligence tools."
I agree with your complaints about the price inflation and other detriments to society. The question is, are the impending productivity improvements going to more than off
Re: (Score:2)
The trend even in theory doesn't seem to keep pace with the depreciation on the assets.
My subjective experience is opposite, after an initial rapid improvement in LLM behavior, the subjective experience has plateaued. Doing a better job with getting the right stuff into context without having to manually stuff it with certain tools, and that counts for a lot, but given the same context the outputs are about as unreliable as they have been, including gemini 3. If it generates too much code, then it's more
Re: (Score:2)
>> the marketing has been obnoxious
They hype it to high heaven for sure.
>> If it generates too much code, then it's more trouble to fix that code than it's worth
I totally agree, but this is where us human beans still come into play. I am glad there is still a role for us. You have to know how much you can expect the AI to do for you reliably. Push it beyond that and it will struggle and possibly fail, much like a human employee. You have to break the work out into digestible pieces.
>> the
Re: (Score:2)
I suppose I wasn't clear, when I said they are dutifully generating code, I mean they *are* using the AI tools. So the leadership is left with the possibilities that either AI isn't fit for the task of suddenly halving their headcount without any transition plan or the employees are to blame, and so they are deciding the employees are to blame.
The leadership cut entire teams and then just assigned their projects to the other half who had never seen the codebase, never used or talked to users of those projec
Re: (Score:2)
Yeah, and how many of those users would pay for an LLM if forced to? They haven't monetized it yet, I do see people using AI. I don't see companies buying AI employees to do their work in the next 5 years. The AI industry does. If we don't have AI that is AGI (whatever that means) in a year an a half, the bubble pops.
Re: (Score:3)
Re: (Score:1)
Re: vast demand for AI (Score:2)
I also believe numbers supplied by companies!! Go Team AI !!
Re: (Score:2)
Oh yeah, Gemini is great! [buzzfeed.com]
I finally just had to block the AI search results on Google because they are so much worse than worthless every single time. The "citations" linked never say what Google claims they say. Every time I search for information on something I know about already I can see that Gemini doesn't know shit.
Apple overpriced RAM upgrades (Score:5, Funny)
Who would have thought that Apple's ludicrously expensive RAM upgrades would some day align perfectly with current RAM prices...
Re: (Score:2)
AI sucks. (Score:2)
I am not a fan of big government, but fucking do something for the retail market.
Re: (Score:2)
It's really damn annoying to get screwed by market conditions. If it weren't for the other disastrous consequences, I'd be praying for an AI bubble to burst so these things can go back on the open market.
Does anyone accept billionaires want this? (Score:3)
We could just tell the billionaires no. We would have to take their money away because money is power but we could do that. There's about 8,000 of them. There's 8 billion of us.
We could just tell them no.
Normal Behavior (Score:2)
I used to work for a very large Japanese electric company conglomerate. When some product was in short supply the internal customers were the first to have their supply cut. The company believed it was most important to keep the outside paying customers happy. It makes sense.
This will turn out great (Score:2)
It doesn't matter (Score:2)
The Tech Bro push for AGI is to Overlord everyone (Score:2)
The objective is the control of an AGI to use against everyone else. Basically, be the one with most powerful ring to rule them all. While everyone else is paying to forge it's creation, investment does not mean control...
Many read The Lord of Rings series [goodreads.com] as a fable, but others read it as a tweak-able (if only) business plan.
These types write stuff like the Techno-Optimist Manifesto [wikipedia.org]. Which is a really a pitch for you to invest in your own demise or be forced to submit to more precise permanent over lording
Samsung always pisses on Samsung (Score:2)
Samsung is collection of several companies and if you've ever spent any time working with them you quickly realize that they all prioritize other Samsung companies below other customers. I don't know whether it's because of anti-trust concerns, or market strategy, or just rivalry, but I've never seen any Samsung company that operated any differently. I worked quite a bit with Samsung Mobile and S.LSI, who are even quite interdependent (though S.LSI depends more on Samsung Mobile than the reverse), and they