Why Korean memory giants aren't rushing to expand DRAM supply
Posted by tecialist@reddit | hardware | View on Reddit | 174 comments
Posted by tecialist@reddit | hardware | View on Reddit | 174 comments
Working-Crab-2826@reddit
Why expand memory supply if you chan charge 200% more with artificial scarcity?
akuncoli@reddit
our only hope is china
just in case someone said "china will do same sell all to AI too " no, cxmt committed to keep make *)% capacity for DRAM and 20% for HBM ram,yes they still sell some to AI but not dicth casual consumer ike other greedy company like samsung micron sk hynix
Strazdas1@reddit
The dumbest statement in the thread full of dumb statement right here.
CXMT is retooling their DDR fabs into HBM fabs abandoning DDR production entirely because HBM is more profitable.
akuncoli@reddit
and you spread fake news
china already said will make 20% production for hbm ai 80% production for dram since beginning, they never said will make all for dram
https://www.techpowerup.com/346207/cxmt-reportedly-plans-to-dedicate-20-of-mass-production-capacity-to-hbm3-line-in-2026
at least china not greedy evil corporation like micron, literally ditch us all casual consumer when they got bigger money from AI company
Strazdas1@reddit
Look not at what they say, look at what they do. They are already regearing their old fabs to produce HBM.
China is literally doing all the same things but worse, yet you cant stop salivating over them.
NaXter24R@reddit
And why would they sell stuff for much less when they can charge the same ridiculous price? They're not doing charity either
TDYDave2@reddit
Sell for less to gain market share.
Classic move on how to penetrate an established market.
Strazdas1@reddit
And after they penetrate it what do they do? oh, thats right, they jack up the prices, just like all other markets they did this in.
TDYDave2@reddit
Never claimed otherwise.
NaXter24R@reddit
Yes, but right now other companies are making 6x the price. They'll sell for less for sure, but that would be 4x, not 1x...
TDYDave2@reddit
No one said they would sell at the old bottom price, just less than the current top price.
tooltalk01@reddit
Let me give it a try. The key point that you are missing is that 100% of all DRAM produced will be sold regardless of price point because the DRAM market is demand driven.
And in this situation, your market share is determined by the quantity of supply, not price. The EV market is not demand-driven -- eg, car shortage during COVID.
iDontSeedMyTorrents@reddit
If you sell 100% of what you are making, how is a lower price going to gain you market share?
akuncoli@reddit
huh? those greedy company (samsung micron etc) definitely profit at least 300% more than capital needed they need to making those ram and ssd. china will always sell cheaper product than competitor
beside, why salty ? samsung and micron literally ditch all of us casual people, dont defend greed multi billion company.
TDYDave2@reddit
True if you want to sell your 100% to the domestic market that is fine, but if you want to break into the international market, you have to give international customers a reason to switch.
Same tactic BYD is using to develop the Chinese car business internationally.
coffeesippingbastard@reddit
kinda. For better or worse, China does actual capitalism better than a lot of countries.
If there is room for SOMEONE to make profit, they will try. If China joins in and CXMT is making stupid amounts of money, there should be additional companies joining the fray to take their cut of the pie- and drive prices down.
The primary limitation in China is really lithography equipment. If they successfully build scalable 12nm process equipment then you may see additional competitors join. Unfortunately at best we're looking at 2028 for any changes.
hackenclaw@reddit
China is capacity constraint by Sanction.
If cxmt can expand without constraint, they would have done by now.
In a DRAM shortage market, they can fill up the market left out by Samsung, SK. even also sell the DRAM to AI companies that do not want to wait.
If the DRAM has oversupply, Cxmt still can rely on China to buy up all their excess production (China gov can force this). Which will kick out Samsung/SK.
If you look at situation like this, there is almost no downside for Cxmt to expand rapidly. The problem is the sanctions keeping them from acquiring enough equipments to expand.
Midiamp@reddit
Asian here. We have Kingbank memory that uses cxmt chips. Price is the same with Corsair, Team, Kingston, G.Skill.
TruthHistorical7515@reddit
Prices will be the same because there is severe supply shortage. Basic economics.
SireEvalish@reddit
“Artificial scarcity”
Jesus fucking Christ.
Jeep-Eep@reddit
What do you call it when (a) it's a set of operations that have been slapped multiple times for cartel bullshit, and (b) that slimy fuck Sam Altman buys some absurd number of memory wafers to deny them to the competition?
Strazdas1@reddit
False statement. There was only one case of provable cartel, 26 years ago.
Thats called real demand, not artificial scarcity.
elkond@reddit
bestie there's a difference between "we collude to artificially raise prices of existing inventory" and "we'd love to produce more, but laws of reality itself are not ours to control "
exilus92@reddit
None of what you describe is called "Artificial scarcity".
Just google the definition or ask chatgpt.
FlyingBishop@reddit
It's important to note that Sam Altman has not yet bought an absurd number of memory wafers, he pinky promised that he would buy an absurd number of memory wafers. It's market manipulation, but also he might, but no one knows. It's hard to call it artificial scarcity when we're talking about 1-5 years from now. No one knows what the demand will be then. No one knows what the supply will be then, it's all guesswork.
Jeep-Eep@reddit
It was very much an attempt at it, and well, Iran's telling us very clearly what the future's looking like.
FlyingBishop@reddit
Iran is terrified and they are blowing shit up left and right. They cannot think about what the future is like, it is too bleak.
Jeep-Eep@reddit
Blowing up shit left and right, but they're not unreasonably scared, nor are they unsystematic. They're executing quite cleanly on a battle plan here and escalating in a controlled, deliberate fashion.
FlyingBishop@reddit
There is nothing clean about what they are doing. They are exercising some restraint, and they know what they can do but they don't know what they're going to do tomorrow.
oioioi9537@reddit
For these people words mean nothing at this point, throw all definition out of the windows and just vibe vomit words out
Working-Crab-2826@reddit
How do the boots taste?
oioioi9537@reddit
calling this artificial scarcity is a roblox gamer take. pay attention in school
Working-Crab-2826@reddit
Avoiding the question is a Roblox gamer take. Pay attention in school.
oioioi9537@reddit
Look nephew at least look up the definition of a word before you use them next time. Or at least try to have an actual logical argument instead of ad hominems lol
Working-Crab-2826@reddit
I mean, you’re the only calling me things here. Twice already.
oioioi9537@reddit
You haven't responded to anyone with any substance not just me lol. Typical gamer talk
Working-Crab-2826@reddit
Third time now. Talk about ad hominems.
Try looking at yourself in the mirror for once, gamer boy.
oioioi9537@reddit
Then explain to me how its artificial scarcity nephew
Haseki-Hurrem-Sultan@reddit
Honestly the mods should just ban people with these low effort takes. It just drags the quality of the subreddit down.
Raikaru@reddit
You just don't understand anything about how things work. In order to build more factories, it would take a year+, and they can't know if demand will even be the same then. If you guys truly think the "AI Bubble" will pop, then that would solve things before they even need to build.
Working-Crab-2826@reddit
I think it’s clear that it is you who does not understand how things work.
theekumquat@reddit
You gonna say anything of substance or just keep repeating nonsense lol
iDontSeedMyTorrents@reddit
250 upvotes as of this post.
This sub has deteriorated so far in terms of quality, overrun by idiot gamers.
StarbeamII@reddit
Fanned on by the likes of Gamersnexus spouting misinfo at them
gokogt386@reddit
All you geniuses keep praying to every god under the sun that the AI bubble crashes hard and then turn right around and expect hardware companies to throw up a million factories to deal with the demand crunch from said bubble
6198573@reddit
They are separate things
An AI bubble might exist, meaning companies are massively overspending on it based on its perceived future value
But AI is a real product with real uses and applications
Even if an AI bubble exists and it bursts, we will still need datacenters and the demand for components won't disappear.
In the short term demand would be lower than what it is right now, but it would undoubtedly still be higher than it was before AI
The Internet didn't disappear when the dotcom bubble burst, and its now much more massive than it was back then
DueAnalysis2@reddit
It's not that the demand would disappear, but that it would dramatically lower. Companies are reluctant to repeat the COVID mistake of overinvesting in capacity when there isn't a clear signal that the demand will last at this level going forward.
If the AI bubble bursts, we'll probably be left with 2, maybe 4 major providers. All the other competition, and their data centre providers would be out and demand for components would fall correspondingly.
FlyingBishop@reddit
I don't think there's any chance of demand for RAM falling, not in a numeric sense. There's probably profitable demand for 5x as much RAM as is being produced on an annual basis right now. Might be razor-thin margins at that volume, but I think there would be margins. So demand might get cut in half but there's still demand for 2-3x as much RAM.
Jeep-Eep@reddit
Eh, when the bubble goes one major sink for it will be gone, most data center ops don't trash blades and cache like genai 'work'; when that crap ends that will be serious demand destruction and it looks like Iran has decreed it.
FlyingBishop@reddit
Iran knocked out the helium supply which is going to kill RAM supply momentarily. Iran isn't doing anything to diminish RAM supply. Your mistake is thinking that AI bubble popping means the AI market goes away, as others have said, the Internet itself only grew larger when the dotcom bubble popped in 2000.
Strazdas1@reddit
Gulf helium isnt even the largest supply of helium.
no6969el@reddit
It's because once it popped only the practical most realistic uses existed. Which is exactly why we are losing Sora and it seems like it's going to repurpose into something more useful for them.
FlyingBishop@reddit
I never used Sora, it seemed like a silly tech demo to me. But OpenAI, contrary to the popular perception, isn't the largest user of RAM in the world. They're not even second-largest. If more RAM becomes available there are literally millions of users waiting to jump.
Jeep-Eep@reddit
AI takes unholy amounts of juice; while they cut the helium, they also cut the natgas and that's a demand destruction locked in because even if they got a peace today that blasted infrastructure won't be running at full speed again before Q3.
Internet was also useful unlike AI which has been enraging everyone except a small but vocal array of idiots and without the vast cash supply behind it, copyright will obliterate it because copyright is a loadbearing function.
FlyingBishop@reddit
You're using AI every day and you don't even know it. LLMs are not even half of AI, they're not even 1/10th of RAM demand.
Jeep-Eep@reddit
They're the biggest driver of the demand spike before the War, and I wasn't talking about the background, useful ML applications here, I was talking about what's driving the current situation.
FlyingBishop@reddit
LLMs are a useful application, just because the hype is wild doesn't mean they're useful. LLMs are basically the state of the art for machine translation, nobody is going to be using MT models that aren't essentially LLMs in 5 years.
fastheadcrab@reddit
Perhaps, but only if the AI craze continues unabated. The high energy prices will slow data center construction. It's also likely that cloud AI providers will probably have to raise prices as they are already subsidizing usage quite a bit.
Jeep-Eep@reddit
I mean, with the chinese operations coming I'd at least try to up volume because otherwise that's asked them to undercut you and eat your lunch.
Reasonable-Sea3407@reddit
Internet is indeed bigger but hardware during which last bubble pop is worthless. Same is the fate of current hardware. All will be flood to market after 3 to 4 years when next better hardware will come. Question is will ai companies survive these 4 years to buy the next batch. If yes than we are screwed if not than manufacturer are screwed so they are making as much bank as they could during boom to survive the bust.
6198573@reddit
I think the broad the discussion is about memory makers increasing their production capabilities, for current and future hardware
And hardware technology isn't moving that fast at the moment, DDR4 is still relevant for a lot of servers so i don't see something like DDR5 becoming useless any time soon
I just don't see demand for components going down. Some AI companies that over-invested and over-promissed might fold, but the technology itself will find its place
And new datacenters will continue to be built and even consumers/companies will look for hardware to run their own local AI models
Reasonable-Sea3407@reddit
Ddr5 will not become useless but higher capacity of ddr5 will be possible so most data centre replace the lower capacity ram of today for the higher one of future unless companies doesn't make one. So ram of today will be replaced by data center in few years one way or another.
Companies simply bought too much to not get behind but only few will survive and those will rule the world for next few decade but rest will pop with the bubble. Happened with the process than graphic card hard drive now with ai models. Only 2 to 3 will survive at the end and all other will go bankrupt or give up on ai race. So one way or another next batch will only be bought by few companies unlike now. So manufacturers are making bank now because they know bust is coming. I just hope my hardware does not fail till than because i simply can afford to fix my pc at these prices.
zoson@reddit
The problem with the argument you are making is we know that none of the manufacturers are running their ddr5 production at maximum capacity. this is fab space completely separate from HBM. it's been reported on several times. they are creating an artificial bottleneck.
StarbeamII@reddit
Source on memory makers not currently running their production at maximum capacity?
zoson@reddit
There are TONS of articles about it. Just google something like:
They are all stating that they are being cautious because of the last memory bust. Some are making up nonsense about not being able to bring the fabs up to capacity due to a lack of cleanroom space, but that is complete nonsense.
StarbeamII@reddit
The very first hit says:
zoson@reddit
You clearly don't know how a search algorithm works because that is this article. Imagine citing something as proof of itself.
StarbeamII@reddit
I am quoting a newspaper article, which was the very first result for what you suggested I search, as proof directly contradicting your claim that "none of the manufacturers are running their ddr5 production at maximum capacity". Meanwhile, you cite an AI search result rather than any actual sources or articles, despite claiming that there are "TONS of articles" claiming as such.
zoson@reddit
You clearly cherry picked. There is a more recent Tweaktown(written on the 18th) article AND a Bloomberg(written on the 8th) article that discuss it and are the other top hits. Get blocked for being an awful person who operates 'prove my bias' rather than 'review the proof for the answer.'
StarbeamII@reddit
Source on memory makers not running their production at maximum capacity?
Y0tsuya@reddit
Those people would sell their own mothers for a chance to buy cheap parts for their next build. Their wet dream is Chinese companies taking over everything to "smash the cartels." The thought of millions of Americans losing their jobs never cross their minds.
alvenestthol@reddit
If companies can fuck up their operation and wait for bailouts from the government, then we ought to hope companies operate on a loss in our favour
Many tech companies start out operating at a loss and are sustained by investment, and then they enshittify once they have enough market share and actually start to turn a profit. Being smart means using services that are being run at a loss, and ditching services that are making a profit.
We are hoping for hardware companies to up the production for a demand crunch that will not exist, and then be forced to liquidify a massive inventory of chips for lower than cost.
DXPower@reddit
The AI hardware companies aren't producing these chips, the fabs do. And fabs have so much capital investment that you cannot just "run at a loss" - the company would get strangled by debt. This already happened in the last memory shortsge, and it nearly killed multiple memory fabrication companies. That shortage, of course, was much smaller than this one.
oioioi9537@reddit
sk and samsung's battery subsidiaries also learnt the same lesson scaling up batteries for EV, and now are rotting. these people genuinely think companies should take on losses for the gamers. like i fucking hate the AI bubble but these takes are just so insanely out of touch with reality its crazy
Any-Captain-7937@reddit
Ok, keep hoping lol
SpitneyBearz@reddit
200%?? 500-600% here!
zoson@reddit
Yep. I paid $530 for my 64gb GSkill Zeta R5 ECC 6400MHz 4x16GB kit. The equivalent kit is now $3000.
Gimme_Doi@reddit
nasty
krioru@reddit
7 times? Doubt it.
shermX@reddit
7 times is probably a bit exggerated, but 5 times is unfortuntely real in some cases
https://cdna.pcpartpicker.com/static/forever/images/trends/2026.03.25.usd.ram.ddr5.6000.2x32768.3ebdeff0bbc2468433864cc61459e63d.png
AOEIU@reddit
In the context of memory manufactures it's not an exaggeration. Spot prices for memory modules are ~7.5x of what they were 1 year ago.
https://www.dramexchange.com/ https://web.archive.org/web/20250320015714/https://www.dramexchange.com/
The rest of the stuff that goes into selling a DIMM hasn't changed in price, so looking at a retail product (like your link) will show a smaller increase.
tnoy@reddit
Retail pricing and manufacturer pricing are not always going to be one to one.
SJGucky@reddit
+300% in most cases.
My specific 64GB RAM kit for example was bought for 225€ 16months ago.
It was over 1100€ at some point in the last few months. Now it is around 900€.
Mental-At-ThirtyFive@reddit
There is no artificial scarcity - just demand outstripping supply.
Yeah they can expand capacity, but this is a multi-year planning and execution that the corporate boards to figure out (using management consultants obviously) long term profitability.
Look at Nvidia leveraging the demand with all the new adjacent tech and products - expect memory product shape to be similarly different by the time we hit 2030s Let's say 1% of Saudi Arabia oil fields gets hit by Iran - they will laugh their way to much higher revenue multiples
Creepy-Bell-4527@reddit
Demand backed by money that doesn’t exist.
Mental-At-ThirtyFive@reddit
Tech bros decided that that can do money supply better than the Feds - velocity of money takes on a new meaning
Mental-At-ThirtyFive@reddit
Should buy some gold coins
fatso486@reddit
very true, but It's probably closer to %500. The part I don't understand is the Chinese memory makers like CXMT. Their memory prices are not different than the big three right now and I don't see them flooding the markets with the products. Isn't it a huge lost opportunity for them to be on the good graces of consumers.
JustGotHit@reddit
Theyre not allowed to do business with most Western companies due to threats from the US. If you do business with CXMT, you get put on an entity list that bars any American or American owned company from doing business with those on the list. CXMT is on said list. Apple tried getting permission from the US to use CXMT ram and storage chips in Asia only iPhone but they got told if they went through with it, say goodbye to your entire business. It even applies if its not American owned. If say the chinese/Asia division of a European firm wants to do business with a company on an entity list, the entire company gets put on the entity list.
There is CXMT ram by King Bank in Asia and Australia but they're pricing themselves just as high as everybody else. King Bank could honestly skyrocket their sales by decreasing their prices.
RSACT@reddit
They have halved the price of their DDR4, but for DDR5 they just started up most of their factories in December 2025, and they have enough demand in just China that there's no need for them to lower their price. For RAM, don't see how consumer goodwill matters, people just buy whatever is cheapest while sometimes avoiding certain brands from my experience.
Jeep-Eep@reddit
Yeah, their lines are still bootstrapping - while they can match the best of the western RAM suppliers if those 8000 rated kits are anything to go by, they're still building up capacity.
But they will build up capacity because the PRC will not tolerate being held hostage by the RAM cartel.
StarbeamII@reddit
Because the scarcity is not artificial and the Chinese produce into the same pool of DRAM supply, so they can command the same prices.
Pixelgordo@reddit
Good grace of consumers Vs good grace of benefits. Which one do you think they prefer?
Proglamer@reddit
The free market will self-correct in 1...2... /s
odrea@reddit
here it's 600-700%
reddit_reaper@reddit
They're learning from the diamond market
GraXXoR@reddit
200% lol. Oh you sweet summer child.
Awkward-Candle-4977@reddit
They do rush to maximize profits of their factories
Yodl007@reddit
You know how you maximize profits from existing factories ? You don't produce as much so supply is even lower than if you ran at 100%.
Strazdas1@reddit
You maximise profits by running your factory at 100% when demand is above your manufacturing capacity.
Cheap-Scarcity-1621@reddit
You can't just expand production. It'll only take a year or two to just build the plant. And what about the lithography equipment for it? There's a waiting list for a couple of years ahead today. And what about training the people?
red286@reddit
There's also the issue that the second the AI bubble pops, any expanded capacity will go to waste. If that happens within 3 years, that's 100% wasted capital investment.
cheapseats91@reddit
Everyone here hates on AI and wants it to go away but expects DRAM manufacturers to make a hundreds of millions of dollar bet on AI still being around and paying through the nose for memory.
nosurprisespls@reddit
AI is not going away. Even if OpenAI collapses, there is still huge demand by other company.
elkond@reddit
the exact bet they are making is that the bubble will pop and that's why they arent rushing to investments that would be done in a time frame, that too compare, is as far from today as NFT craze
StarbeamII@reddit
Closer to tens of billions.
gorion@reddit
3+ Years, not one.
ajrf92@reddit
Not to mention bureaucracy.
loinclothsucculent@reddit
No guarantee consumers will have the cash to spend in 4 years anyway, so no point investing in a losing proposition.
elkond@reddit
yeah 3+ years is a "everybody locked tf in" kind of a semi fab spin up
finance is a mental disorder that ideas like "well just get more production capacity???" is not too preposterous to them to speak out loud
UpsetKoalaBear@reddit
A lot of people don’t know how much DRAM/NAND production is effectively commoditised.
These companies are incredibly wary of building capacity because they don’t want overproduction.
To give you an idea of how commoditised this industry is, these companies frequently cut and increase their DRAM production to prevent inventory build up.
The current situation was predicted by trendforce earlier last year because of cuts in production.
no6969el@reddit
Exactly, it's not like they can't do it..it's cause they do not want to either.
Vushivushi@reddit
Micron straight up acquired a pre-existing fab from PSMC and it still takes about 18 months to retrofit and start production.
This is basically as fast as they can do it.
ajrf92@reddit
Not to mention that if people don't buy as much as before...
LingonberryGreen8881@reddit
A year ago, the industry jabber was about creating an ecosystem of consumers with 128GB graphics memory - that's impossible now, a year later.
Nvidia's entire line of "Super" 2026 GPUs was cancelled.
Level 4 self driving cars will need a minimum of 300GB each which is price prohibitive.
Agents and AGI will unlock a magnitude more demand.
There is such a hole in the market right now that a drop in price opens up entirely new product possibilities. in 2027, memory makers could still demand strong margins if they had 10x the bit production they have today.
You can think of it like the invention of electricity. People in 1900 might have been able to trade some potatoes for enough lamp oil to satisfy their lighting needs. They might call electricty overhyped and say there is too much investment in it. Those people would be failing to appreciate the posibilities that electricity will unlock.
Celodurismo@reddit
And then you go invest to accomplish everything you listed, and then supply returns to normal.
moashforbridgefour@reddit
This headline is nonsense. Literally all memory suppliers are currently expanding production capability. It takes years and many billions of dollars for the new fabs to come online, but Hynix and Samsung both have new fabs that are, at this moment, ramping up production. And micron has like 8 new fab announcements, with ground broken on around 5 of them, 2 of which will be ramping up production in 2027.
tecialist@reddit (OP)
I get the feeling you didn’t actually read the article, so quick recap because the headline isn’t wrong: yes fabs are being built but they take years to show up, most of the new capacity is going into HBM and other high-margin parts instead of regular DRAM, and after getting burned in 2023 the big players are being careful not to flood the market again, so the point isn’t “no expansion” it’s that the expansion isn’t fast enough or aimed at the parts that would actually fix the shortage right now
Vushivushi@reddit
If you have to explain that the headline doesn't mean what it actually says, then the headline is wrong.
Radiant-Sherbet-5461@reddit
No, the headine is 100% correct.
SK fabs are increasing capacity but not investing ridiculous amount on it since they did exactly that during COVID and lost billions.
Explanation is simply because Redditors comment without reading TFA.
Vushivushi@reddit
They are rushing to expand, but the two years prior to the past 6 months, they weren't rushing to expand.
This is a better headline:
https://koreajoongangdaily.joins.com/news/2026-03-12/business/industry/Samsung-and-SK-are-expanding-fast-but-why-is-memory-still-in-short-supply/2540153
moashforbridgefour@reddit
Literally everything you and the comment you are replying to said is wrong. No one started or finished a new fab around COVID. The CHIPS act passed in 2022, which was right when everyone started making new fabs. Prior to that, there was a long period where no one was expanding.
Vushivushi@reddit
Both SK Hynix (M16) and Samsung (Pyeongtaek P2) completed in 2021... during COVID. Those were announced during the 2018 memory cycle. That cycle was indeed a long period where there hadn't been enough expansion. As a result of those new fabs finishing during COVID, there was an expansion of new wafer starts.
New fabs were announced in 2023, after the CHIPS Act, though neither SK Hynix or Samsung have memory fabs in the US.
The last memory cycle ended in 2023, with COVID. That's what I mean by, "the two years prior to the past 6 months, they weren't rushing to expand." The fabs/expansions announced in 2023 would be delayed in 2024 and even halted. That's why Samsung's Pyeongtaek P5 won't be completed until 2028. They had barely started it.
Samsung is rushing to complete it now in addition to expansions at P4.
We can go back and forth on the history of the memory industry, but what the headline says is still wrong.
ThirdPartyCrap@reddit
lol every time I see HBM I think of Human body model ESD testing
moashforbridgefour@reddit
You're right, I didn't read the article. It wouldn't load for me, maybe I'm region locked out since I'm in SE Asia. That is why I specifically mentioned the headline.
The article according to your recap is still wrong though. Yes, manufacturers are hesitant to create an oversupply situation again since the up and down cycle has plagued the industry forever. For those of us within the industry, it very much seems like they are expanding at an irresponsibily fast pace. The message we have been hearing unanimously is that new markets are opening and years of stagnation in demand are finally breaking. But even with historic investments around the time of the CHIPS act, they have all announced new fabs globally since then.
YourVelourFog@reddit
This sounds like BS from the Korean Herald considering SKHynix just placed an $8 billion order for ASML machines to produce more HBM and DRAM...
https://www.tomshardware.com/tech-industry/semiconductors/sk-hynix-places-record-8-billion-order-for-asml-euv-lithography-machines
tecialist@reddit (OP)
The article isn’t saying SK hynix and Samsung are sitting on their hands. It’s saying that “we ordered a bunch of gear” is not the same thing as “the market is about to be flooded with DRAM.” An EUV order is basically buying more kitchen equipment while the restaurant is already slammed, useful, yes, but it still has to be delivered, installed, tuned, and worked into actual production, and that takes time.
Then there’s the bigger point the article makes: even the new capacity that does come online is being aimed mostly at HBM and other premium memory, because that’s where the money is and because nobody wants to repeat the 2023 crash by overbuilding commodity DRAM. So the headline isn’t wrong at all. They are expanding, just carefully, slowly, and in a way that does not quickly relieve the broader shortage
moashforbridgefour@reddit
The 2023 crash had nothing to do with overbuilding commodity DRAM. It was from enterprise customers stockpiling memory following a chip shortage. The chip shortage around 2020 was catastrophic for many of these customers, so they overreacted and bought way more than they needed as soon as it became available. When these customer's respective markets softened, memory demand fell through the floor since everyone had a pile of memory to work through.
Memory manufacturers are currently attempting to combat this sort of unmanageable behavior by leveraging their current, strong position to push customers into strategic long term agreements. It will guarantee dependable supply and prices to large customers in return for dependable consumption of that supply. This gives memory manufacturers more safety to make large investments like building fabs.
You say they are expanding slowly and carefully as opposed to some imaginary way that would magic away current shortages. Such a solution does not exist. Fabs take a tremendous amount of time to spin up. They are all, at this very moment, expanding at a breakneck pace and dumping mountains of cash onto it to move even faster. The way you and many people in this thread are framing the issue is categorically wrong.
tecialist@reddit (OP)
You're spot on that the 2023 crash was a classic "bullwhip effect" driven by pandemic-era inventory stockpiling rather than just a raw overbuild of factories, and your point about the shift toward Long-Term Agreements (LTAs) is the most accurate way to describe how the industry is trying to kill off its historical volatility.
However, the article’s "rushing" framing isn't entirely "imaginary" — while Samsung and SK hynix are indeed spending record capital, that cash is being swallowed by the massive HBM die-size penalty (which effectively shrinks total bit supply) and the soaring costs of EUV lithography for the 1c node. They are expanding at a "breakneck pace" for their AI-centric premium customers, but they are demonstrating calculated, strategic restraint toward the commodity DRAM market to ensure they never return to the margin-killing oversupply of previous cycles.
moashforbridgefour@reddit
EUV costs are not a bottleneck. Tool supply is.
HBM and commodity DRAM use the same tooling. HBM just requires advanced packaging in addition. If they are expanding supply for HBM, they are expanding supply for all DRAM. And whatever customers they are aiming new supply at does not change the fact that they are expanding supply at an unbelievable rate. There is no restraint being exercised from my point of view, and I challenge you to provide empirical evidence of any restraint.
ShareACokeWithBoonen@reddit
he's using AI to autogenerate responses, don't get your hopes up
EarlMarshal@reddit
That just sounds like a very very very slight increase that will still leave us with a massive shortage for DRAM.
LAwLzaWU1A@reddit
What do you base the assertion that 30 EUV machines is a "very very very slight increase" on?
EarlMarshal@reddit
The prior forecast was 26. They increased to 30. That's an increase of roughly 15% while they, shift most production capacity away from consumer market.
LAwLzaWU1A@reddit
1) Do you have a source for that? I know that David Dao once speculated about them buying 26 EUV machines, but that was never a confirmed order or anything. It was just "I expect them to buy 26 machines. Oh, it was actually 30". That does not mean the 26 were already planned and part of their usual growth.
2) How do we know that the 26 that they were supposedly already planning wasn't a big expansion to begin with? It feels a bit dishonest to just ignore all the the supposed expansion they were already planning and only focus on the additional one.
3) 15% is in my opinion not "very very very slight increase". I would say it seems like quite a lot. According to TrendForce, which I trust quite a lot, the additional demand from AI datacenters is about 20% of the global DRAM supply. So if it's 20% increase in demand that has caused all of this havoc, an additional 15% extra capacity to the already planned capacity increase seems pretty good to me.
4) It's still about 8 billion dollars of investment. I am not sure how much more you can reasonably expect them to invest in the EUV machines alone. Then there is all the other stuff surrounding the machines. The staff, the clean rooms, the facilities and so on. It is a MASSIVE investment being made and I feel like you are downplaying it quite a lot.
5) A big issue is that ASML has a massive backlog. Even if they were to order more machines, they wouldn't be delivered in a reasonable timeframe.
Lazy_Association_847@reddit
Do you have any info how many do they have now?
Particular-Novel4963@reddit
Everyone acting like this is a “shortage” when it’s really samsung and hynix just dumping capacity into HBM where the money is…. if you’re buying regular DRAM you’re basically an afterthought I guess
tecialist@reddit (OP)
korean memory giants learned the hard way that more supply = instant price collapse. so now demand is insane and they’re still like nope, not falling for that again
FlyingBishop@reddit
The thing is, like, I would love to buy a GPU with 1TB of RAM but I would pay maybe $4k for it. There's a gradient too, I might pay $4K for a GPU with 90GB of RAM. The cost/benefit is complicated, it's gut-based but it's also based on how much money I have to spend and what I can get. At these prices I'm not buying anything. Point is demand goes up a lot as price goes down, nobody needs RAM, not even Sam Altman, but a lot of people, especially anyone who is doing anything with AI would love to have 5x as much RAM for the same price they're paying now.
Caffdy@reddit
companies are paying $40K, $50K for gpus with way less vRAM, that's the thing. Why would Nvidia or AMD market a $5K gpu for consumers with 1TB of ram?
FlyingBishop@reddit
The Nvidia GeForce 256 was released in 1999 with 32 megabytes of VRAM and cost $250, which is about $500 adjusted for inflation. They made a profit doing it because nothing existed like it.
They would market a $5k GPU with 1TB of ram for essentially the same reason, which is if that's the market price and they can do it profitably.
dztruthseek@reddit
This....is a pretty dumb comment.
ghostsilver@reddit
so you just describe that DRAM is having a shortage?
CookieEquivalent5996@reddit
Everyone acting like it's a shortage when it's just a lack of supply because of a reason.
moashforbridgefour@reddit
Isn't that exactly what a shortage is, though? If they were dumping capacity into DIMMs, then the data centers would not be getting their HBM. There is a shortage of capacity compared to total market demand. Just because they are choosing to direct capacity away from the sector you prefer doesn't mean that there isn't a shortage.
redisprecious@reddit
Why rush? They're literally profiting off "shortage" by not purposefully making enough.
StarbeamII@reddit
Because they would profit more if they produced more, since prices are way higher than cost of production right now.
Cheeze_It@reddit
Because chaebols are fucking horrible and will always be in power?
jknvv13@reddit
Expand. Demand goes low again because all the datacenters and fairy dust are built. No demand.
Then what.
GestureArtist@reddit
This should be a crime.
SireEvalish@reddit
“Not manufacturing something should be a crime.” - Reddit legal experts
GestureArtist@reddit
Are you dumb?
here let me educate you.
https://en.wikipedia.org/wiki/DRAM_price_fixing_scandal
DehydratedButTired@reddit
Why expand to accommodate a bubble of investor driven experimentation?
jenny_905@reddit
Company likes making big profits, shocker.
-CynicalPole-@reddit
Because they have same profits by artificially inflating price
Euler007@reddit
Collusion is the word you're looking for.
Ants_r_us@reddit
Because it's a cartel and this situation was their ultimate wet dream.
Aragil@reddit
Well, as a result of that Chinese firms will slice a huge part of the consumer market, and when the next time the industry will hit oversupply, guess who will be squeezed out
GestureArtist@reddit
Dont worry, all foreign routers are banned now so... i'm sure all chinese ram would be banned as well should that ever happen
ledfrisby@reddit
Yeah, I guess that makes sense if completely ignore that the security risk of Chinese routers, which is the reason for the ban, doesn't apply to RAM at all. Or maybe it makes sense if you assume no connection between cause and effect.
TruthHistorical7515@reddit
"security risk" = economic protectionism
braiam@reddit
Name a router not manufactured in China. Come on, I will wait. Ubiquity, Mitrokit, Cisco, Tp link, Asus, etc... all of them made in china.
BileBlight@reddit
Just gonna get rebranded and sold on eBay. It’s not a car you have to register, very easy to fit in your pocket and cross the border or ship by mail
MedicBuddy@reddit
Where do you think ram comes from? It's foreign too. There's going to
GestureArtist@reddit
Yeah but so far it’s from us, Taiwan and Korea. Should China threaten that. Our government is so fucked it would probably block Chinese ram and then fuck over consumers that would have benefited
hackenclaw@reddit
they even tried to even ban DJI, after notice it is too late DJI has like 80%+ drone market. Banning them is effectively banning drones. lol.
GestureArtist@reddit
DJI is still banned. Any new models of DJI drones are banned. Those who have existing DJI drones can keep them, fly them etc but no more DJI drones will be sold it seems.
zerGoot@reddit
because it's artificial scarcity, and because they are a cartel
superSmitty9999@reddit
They see the bubble lol
irrealewunsche@reddit
This. They aren't expecting demand to be this high for long, so they aren't willing to invest in new capacity that will probably unnecessary by the time it's ready.
Scooter30@reddit
I would guess limited supply means they can charge more for it. $$$
whaletosser@reddit
Price fixing with korean characteristics
RJsRX7@reddit
Why the DRAM cartel is happy DRAM is in unlimited demand
nariofthewind@reddit
Tight supply = very high margins but the wild card here is China which this year may be silent but next one will most likely enter the market in force. After that is just a game of boom and bust and who survives in between. Historically Samsung always came on top, SK Hynix and Micron are playing a very risky game because bust sure will come.