[HN Gopher] The RAM shortage comes for us all
___________________________________________________________________
The RAM shortage comes for us all
Author : speckx
Score : 275 points
Date : 2025-12-04 19:16 UTC (3 hours ago)
(HTM) web link (www.jeffgeerling.com)
(TXT) w3m dump (www.jeffgeerling.com)
| jsheard wrote:
| I wonder if Apple will budge. The margins on their RAM upgrades
| were so ludicrous before that they're probably still RAM-
| profitable even without raising their prices, but do they _want_
| to give up those fat margins?
| loloquwowndueo wrote:
| It's 4D chess my dude, they were just training people to accept
| those super high ram prices. They saw this coming I tell you!
| dcchambers wrote:
| I am fully expecting a 20%+ price bump on new mac hardware next
| year.
| Eric_WVGG wrote:
| Not me. It's wildly unusual for Apple to raise their prices
| on basically anything... in fact I'm not sure if its ever
| happened. *
|
| It's been pointed out by others that price is part of Apple's
| marketing strategy. You can see that in the trash can Mac
| Pro, which logically should have gotten cheaper over the
| ridiculous six years it was on sale with near-unchanged
| specs. But the marketing message was, "we're selling a $3000
| computer."
|
| Those fat margins leave them with a nice buffer. Competing
| products will get more expensive; Apple's will sit still and
| look even better by comparison.
|
| We are fortunate that Apple picked last year to make 16gb the
| new floor, though! And I don't think we're going to see base
| SSDs get any more generous for a very, very long time.
|
| * okay I do remember that Macbook Airs could be had for $999
| for a few years, that disappeared for a while, then came back
| suprnurd wrote:
| I read online that Apple uses three different RAM suppliers
| supposedly? I wonder if Apple has the ability to just make
| their own RAM?
| umanwizard wrote:
| Apple doesn't own semiconductor fabs. They're not capable of
| making their own RAM.
| kayson wrote:
| Apple doesn't own any foundries, so no. It's not trivial to
| spin up a DRAM foundry either. I do wonder if we'll see TSMC
| enter the market though. Maybe under pressure from Apple or
| nvidia...
| FastFT wrote:
| There are no large scale pure play DRAM fabs that I'm aware
| of, so Apple is (more or less) buying from the same 3
| companies as everyone else.
| Night_Thastus wrote:
| RAM upgrades are such a minor, insignificant part of Apple's
| income - and play no part in plans for future expansion/stock
| growth.
|
| They don't care. They'll pass the cost on to the consumers and
| not give it a second thought.
| rfmc wrote:
| I know contract prices are not set in stone. But if there's one
| company that probably has their contract prices set for some
| time in the future, that company is Apple, so I don't think
| they will be giving up their margins anytime soon.
| throw0101d wrote:
| > _I wonder if Apple will budge._
|
| Perhaps I don't understand something so clarification would be
| helpful:
|
| I was under the impression that Apple's RAM was on-die, and so
| baked in during chip manufacturing and not a 'stand alone' SKU
| that is grafted onto the die. So Apple does not go out to
| purchase third-party product, but rather self-makes it (via
| ASML) when the rest of the chip is made (CPU, GPU, I/O
| controller, _etc_ ).
|
| Is this not the case?
| jsheard wrote:
| Apple's RAM is on-package, not on-die. The memory is still a
| separate die which they buy from the same suppliers as
| everyone else.
|
| https://upload.wikimedia.org/wikipedia/commons/d/df/Mac_Mini.
| ..
|
| That whole square is the M1 package, Apple's custom die is
| under the heatspreader on the left, and the two blocks on the
| right are LPDDR packages stacked on top of the main package.
|
| https://wccftech.com/apple-m2-ultra-soc-delidded-package-
| siz...
|
| Scaled up, the M2 Ultra is the same deal just with two
| compute dies and 8 separate memory packages.
| mixermachine wrote:
| Sadly everything in the general direction of RAM or SSD chips
| is getting more expensive because a lot of production
| capacity is redistributed to serve AI chips and everything
| around.
|
| Even lower end GPUs are getting more expensive even if they
| are not really useful for AI. But they still contain <some>
| chips and ram which is in high demand.
|
| So yes, Apple will likely also have to pay higher priceses
| when they renew their contracts.
| diabllicseagull wrote:
| I'd like to believe that their pricing for ram upgrades are
| like that so the base model can hit a low enough of a price. I
| don't believe they have the same margin for the base model
| compared to the base model + memory upgrade.
| txdv wrote:
| on one hand they are loosing profit, on the other hand they are
| gaining on market share. They will probably wait a short while
| to assess how much they are willing to sacrifice profits for
| market share
| hyperhello wrote:
| Is this a shortage of every type of RAM simultaneously?
| internetter wrote:
| Essentially yes, not necessarily equivalently but every type
| has increased substantially
| jsheard wrote:
| Every type of DRAM is ultimately made at the same fabs, so if
| one type is suddenly in high demand then the supply of
| everything else is going to suffer.
| bee_rider wrote:
| Wait, really? For CPUs each generation needs basically a
| whole new fab, I thought... are they more able to
| incrementally upgrade RAM fabs somehow?
| selectodude wrote:
| The old equipment is mothballed because china is the only
| buyer and nobody wants to do anything that the Trump admin
| will at some point decide is tariff-worthy. So it all sits.
| mixermachine wrote:
| Not quickly but if somebody puts enough money on the table,
| the fabs change too. All about cost and return. Micron just
| axed their brand crucial (end customer RAM and SSD) because
| they will only sell to database centers from now on.
|
| Crazy times.
| fithisux wrote:
| 32GB should be more than enough.
|
| You can go 16GB if you go native and throw some assembly in the
| mix. Use old school scripting languages. Debloat browsers.
|
| It has been long delayed.
| Night_Thastus wrote:
| 16GB is more than fine if you're not doing high-end gaming, or
| heavy production workloads. No need for debloating.
|
| But it doesn't matter either way, because both 16 and 32GB have
| what, doubled, tripled? It's nuts. Even if you say "just buy
| less memory", now is a horrible time to be building a system.
| Aardwolf wrote:
| I found web browser tabs eating too much memory when you only
| have 16GB
| Night_Thastus wrote:
| That's strange. How many tabs are we talking? Are they
| running something intense or just ordinary webpages?
| prmoustache wrote:
| Use adblock, stop visiting nasty websites and open less
| tabs. Problem is easily solved.
| felixfurtak wrote:
| "640K ought to be enough for anybody." - Bill Gates
| shawndumas wrote:
| there's no reliable evidence he ever uttered that phrase
| felixfurtak wrote:
| Doesn't matter. It's folklore.
| franciscojs wrote:
| I bought a motherboard to build a DIY NAS... takes DDR5 SO-DIMM
| RAM and only 16gb costs more than double the motherboard (which
| includes an intel processor)
| harvey9 wrote:
| Hey, hey 16k What does that get you today?
|
| https://m.youtube.com/watch?v=IagZIM9MtLo
| khannn wrote:
| 16GB is more than enough on Linux, but Win11 eats resources
| like crazy
| anthk wrote:
| I'm using 1GB with TWM, DIllo, TUI tools, XTerm, MuPDF and the
| like. As most tools are small from https://t3x.org,
| https://luxferre.top and https://howerj.github.io/subleq.htm
| with EForth, (and I try to use cparser instead of) clang, my
| requeriments are really tiny.
|
| You can achieve a lot by learning Klong and reading the intro
| on statistics. And xargs to paralelize stuff. Oh, and vidir to
| edit directories at crazy speeds with any editor, even nano or
| gedit if you like them.
| Forgeties79 wrote:
| Sure but 32GB DDR5 ram has just jumped from ~$100 to $300+ in a
| flash. The 2x16GB I have in my recent build went from $105 for
| the pair to $250 _each_. $500 total!
|
| SSD's are also up. Hell I am seeing refurbished enterprise
| HDD's at 2x right now. It's sharp increases basically across
| the board except for CPU's/GPU's.
|
| Every PC build basically just cranked up $400-$600 easily, and
| that's not accounting for the impact of inflation over the last
| few years weakening everyone's wallets. The $1600 machine I
| spec'd out for my buddy 5 weeks ago to buy parts for this Black
| Friday now runs $2k even.
| Loic wrote:
| I think the OpenAI deal to lock wafers was a wonderful coup.
| OpenAI is more and more losing ground against the regularity[0]
| of the improvements coming from Anthropic, Google and even the
| open weights models. By creating a chock point at the hardware
| level, OpenAI can prevent the competition from increasing their
| reach because of the lack of hardware.
|
| [0]: For me this is really an important part of working with
| Claude, the model improves with the time but stay consistent, its
| "personality" or whatever you want to call it, has been really
| stable over the past versions, this allows a very smooth
| transition from version N to N+1.
| codybontecou wrote:
| This became very clear with the outrage, rather than
| excitement, of forcing users to upgrade to ChatGPT-5 over 4o.
| Grosvenor wrote:
| Could this generate pressure to produce less memory hungry
| models?
| hodgehog11 wrote:
| There has always been pressure to do so, but there are
| fundamental bottlenecks in performance when it comes to model
| size.
|
| What I can think of is that there may be a push toward
| training for exclusively search-based rewards so that the
| model isn't required to compress a large proportion of the
| internet into their weights. But this is likely to be much
| slower and come with initial performance costs that frontier
| model developers will not want to incur.
| UncleOxidant wrote:
| Or maybe models that are much more task-focused? Like
| models that are trained on just math & coding?
| agoodusername63 wrote:
| isn't that what the mixture of experts trick that all the
| big players do is? Bunch of smaller, tightly focused
| models
| Grosvenor wrote:
| Yeah that was my unspoken assumption. The pressure here
| results in an entirely different approach or model
| architecture.
|
| If openAI is spending $500B then someone can get ahead by
| spending $1B which improves the model by >0.2%
|
| I bet there's a group or three that could improve results a
| lot more than 0.2% with $1B.
| parineum wrote:
| > so that the model isn't required to compress a large
| proportion of the internet into their weights.
|
| The knowledge compressed into an LLM is a byproduct of
| training, not a goal. Training on internet data teaches the
| model to talk at all. The knowledge and ability to speak
| are intertwined.
| thisrobot wrote:
| I wonder if this maintains the natural language
| capabilities which are what LLM's magic to me. There is a
| probably some middle ground, but not having to know what
| expressions, or idiomatic speech an LLM will understand is
| really powerful from a user experience point of view.
| jiggawatts wrote:
| > exclusively search-based rewards so that the model isn't
| required to compress a large proportion of the internet
| into their weights.
|
| That just gave me an idea! I wonder how useful (and for
| what) a model would be if it was trained using a two-phase
| approach:
|
| 1) Put the training data through an embedding model to
| create a giant vector index of the entire Internet.
|
| 2) Train a transformer LLM but instead only utilising its
| weights, it can also do lookups against the index.
|
| Its like a MoE where one (or more) of the experts is a
| fuzzy google search.
|
| The best thing is that adding up-to-date knowledge won't
| require retraining the entire model!
| lofaszvanitt wrote:
| Of course and then watch those companies reined in.
| hodgehog11 wrote:
| I don't see this working for Google though, since they make
| their own custom hardware in the form of the TPUs. Unless those
| designs include components that are also susceptible?
| frankchn wrote:
| TPUs use HBM, which are impacted.
| bri3d wrote:
| Still susceptible, TPUs need DRAM dies just as much as
| anything else that needs to process data. I think they use
| some form of HBM, so they basically have to compete alongside
| the DDR supply chain.
| UncleOxidant wrote:
| Even their TPU based systems need RAM.
| jandrese wrote:
| That was why OpenAI went after the wafers, not the finished
| products. By buying up the supply of the raw materials they
| bottleneck _everybody_ , even unrelated fields. It's the kind
| of move that requires a true asshole to pull off, knowing it
| will give your company an advantage but screw up life for
| literally billions of people at the same time.
| mbesto wrote:
| > By buying up the supply
|
| We actually don't know for certain whether these agreements
| are binding. If OpenAI gets in a credit crunch we'll soon
| find out.
| agoodusername63 wrote:
| Went after the right component too. RAM manufacturers
| _love_ an opportunity to create as much scarcity as
| possible.
| hnuser123456 wrote:
| Sure, but if the price is being inflated by inflated demand,
| then the suppliers will just build more factories until they
| hit a new, higher optimal production level, and prices will
| come back down, and eventually process improvements will lead
| to price-per-GB resuming its overall downtrend.
| malfist wrote:
| Micron has said they're not scaling up production. Presumably
| they're afraid of being left holding the bag when the bubble
| does pop
| Analemma_ wrote:
| Not just Micron, SK Hynix has made similar statements
| (unfortunately I can only find sources in Korean).
|
| DRAM manufacturers got burned multiple times in the past
| scaling up production during a price bubble, and it appears
| they've learned their lesson (to the detriment of the rest
| of us).
| fullstop wrote:
| Why are they building a foundry in Idaho?
|
| https://www.micron.com/us-expansion/id
| delfinom wrote:
| Future demand aka DDR6.
|
| The 2027 timeline for the fab is when DDR6 is due to hit
| market.
| roboror wrote:
| I mean it says on the page
|
| >help ensure U.S. leadership in memory development and
| manufacturing, underpinning a national supply chain and
| R&D ecosystem.
|
| It's more political than supply based
| mindslight wrote:
| Hedging is understandable. But what I don't understand is
| why they didn't hedge by keeping Crucial around but more
| dormant (higher prices, less SKUs, etc)
| daemonologist wrote:
| The theory I've heard is built on the fact that China
| (CXMT) is starting to properly get into DRAM
| manufacturing - Micron might expect that to swamp the low
| end of the market, leaving Crucial unprofitable
| regardless, so they might as well throw in the towel now
| and make as much money as possible from AI/datacenter
| (which has bigger margins) while they can.
|
| But yeah even if that's true I don't know why they
| wouldn't hedge their bets a bit.
| mindslight wrote:
| So position Crucial as a premium brand, raise prices 4x
| instead of 3x, and drastically cut down on the SKUs to
| reduce overhead. If they tried that and kept spiraling
| into fewer and fewer SKUs and sales, I could understand
| it. But the discontinuation felt pretty abrupt.
| mholm wrote:
| Chip factories need years of lead time, and manufacturers
| might be hesitant to take on new debt in a massive bubble
| that might pop before they ever see any returns.
| nutjob2 wrote:
| Memory fabs take billions of dollars and years to build, also
| the memory business is a tough one where losses are common,
| so no such relief in sight.
|
| With a bit of luck OpenAI collapses under its own weight
| sooner than later, otherwise we're screwed for several years.
| lysace wrote:
| Please explain to me like I am five: Why does OpenAI need so
| much RAM?
|
| 2024 production was (according to openai/chatgpt) 120 billion
| gigabytes. With 8 billion humans that's about 15 GB per person.
| mebassett wrote:
| large language models are large and must be loaded into
| memory to train or to use for inference if we want to keep
| them fast. older models like gpt3 have around 175 billion
| parameters. at float32s that comes out to something like
| 700GB of memory. newer models are even larger. and openai
| wants to run them as consumer web services.
| lysace wrote:
| I mean, I know that much. The numbers still don't make
| sense to me. How is my internal model this wrong?
|
| For one, if this was about inference, wouldn't the
| bottleneck be the GPU computation part?
| ssl-3 wrote:
| Concurrency?
|
| Suppose some some parallelized, distributed task requires
| 700GB of memory (I don't know if it does or does not) per
| node to accomplish, and that speed is a concern.
|
| A singular pile of memory that is 700GB is insufficient
| not because it lacks capacity, but instead because it
| lacks scalability. That pile is only enough for 1 node.
|
| If more nodes were added to increase speed but they all
| used that same single 700GB pile, then RAM bandwidth (and
| latency) gets in the way.
| Chiron1991 wrote:
| This "memory shortage" is not about AI companies needing
| main memory (which you plug into mainboards), but
| manufacturers are shifting their production capacities to
| other types of memory that will go onto GPUs. That brings
| supply for other memory products down, increasing their
| market price.
| daemonologist wrote:
| The conspiracy theory (which, to be clear, may be correct) is
| that they don't actually need so much RAM, but they know they
| and all their competitors do still need quite a bit of RAM.
| By buying up all the memory supply they can, for a while,
| keep everyone else from being able to add compute
| capacity/grow their business/compete.
| GistNoesis wrote:
| What they need is not so much memory but memory bandwidth.
|
| For training, their models have a certain number of memory
| needed to store the parameters, and this memory is touched
| for every example of every iteration. Big models have 10^12
| (>1T )parameters, and with typical values of 10^3 examples
| per batch, and 10^6 number of iteration. They need ~10^21
| memory accesses per run. And they want to do multiple runs.
|
| DDR5 RAM bandwidth is 100G/s = 10^11, Graphics RAM (HBM) is
| 1T/s = 10^12. By buying the wafer they get to choose which
| types of memory they get.
|
| 10^21 / 10^12 = 10^9s = 30 years of memory access (just to
| update the model weights), you need to also add a factor
| 10^1-10^3 to account for the memory access needed for the
| model computation)
|
| But the good news is that it parallelize extremely well. If
| you parallelize you 1T parameters, 10^3 times, your run time
| is brought down to 10^6 s = 12 days. But you need 10^3 *10^12
| = 10^15 Bytes of RAM by run for weight update and 10^18 for
| computation (your 120 billions gigabytes is 10^20, so not so
| far off).
|
| Are all these memory access technically required : No if you
| use other algorithms, but more compute and memory is better
| if money is not a problem.
|
| Is it strategically good to deprive your concurrents from
| access to memory : Very short-sighted yes.
|
| It's a textbook cornering of the computing market to prevent
| the emergence of local models, because customers won't be
| able to buy the minimal RAM necessary to run the models
| locally even just the inferencing part (not the training).
| Basically a war on people where little Timmy won't be able to
| get a RAM stick to play computer games at Xmas.
| lysace wrote:
| Thanks - but this seems like fairly extreme speculation.
|
| > if money is not a problem.
|
| Money is a problem, even for them.
| Phelinofist wrote:
| > By creating a chock point at the hardware level, OpenAI can
| prevent the competition from increasing their reach because of
| the lack of hardware
|
| I already hate OpenAI, you don't have to convince me
| bakugo wrote:
| Is anyone else deeply perturbed by the realization that a
| single unprofitable corporation can basically buy out the
| entire world's supply of computing hardware so nobody else can
| have it?
|
| How did we get here? What went so wrong?
| zozbot234 wrote:
| They're simply making a bet that they can put the DRAM dies
| to more valuable use than any of the existing alternatives,
| including e.g. average folks playing the latest videogames on
| their gaming rig. At this kind of scale, they had better be
| right or they are toast: they have essentially gone all-in on
| their bet that this whole AI thing is not going to 'pop'
| anytime soon.
| citizenpaul wrote:
| > Micron's killing the Crucial brand of RAM and storage devices
| completely,
|
| More rot economy. Customers are such a drag. Lets just sell to
| other companies for billion dollar deals at once. These AI
| companies have bottomless wallets. No one has thought of this
| before we will totally get rich.
| MengerSponge wrote:
| "I don't want to make a little bit of money every day. I want
| to make a fuck ton of money all at once."
|
| https://www.youtube.com/watch?v=BzAdXyPYKQo
| aynyc wrote:
| Ha! Maybe Javascript developers will finally drop memory usage!
| You need to display the multiplication table? Please allocate 1GB
| of RAM. Oh, you want alternate row coloring? Here is another
| 100MB of CSS to do that.
|
| edit: this is a joke
| phantasmish wrote:
| I do sometimes reflect on how 64MB of memory was enough to
| browse the Web with two or three tabs open, and (if running
| BeOS) even play MP3s at the same time with no stutters. 128MB
| felt luxurious at that time, it was like having no (memory-
| imposed) limits on personal computing tasks at all.
|
| Now you can't even fit a browser _doing nothing_ into that
| memory...
| anthk wrote:
| HN works under Dillo and you don't needs JS at all. If some
| site needs JS, don't waste your time. Use mpv+yt-dlp where
| possible.
| baal80spam wrote:
| Haha, the amount of downvotes of your very true comment just
| proves how many web developers are there on HN.
| TSiege wrote:
| Its funny because the blogpost author makes the same joke
| nish__ wrote:
| Anyone want to start a fab with me? We can buy an ASML machine
| and figure out the rest as we go. Toronto area btw
| jacquesm wrote:
| I hope you have very deep pockets. But I'm cheering you on from
| the sidelines.
| nish__ wrote:
| Just need a half billion in upfront investment. And thank you
| for the support :)
| dylan604 wrote:
| So, playing the Mega Powerball are you?
| nish__ wrote:
| I made a website: https://ontario-chips.vercel.app/
| dylan604 wrote:
| wonder which one will find a winner faster...
| nish__ wrote:
| It's a toss up.
| jvdvegt wrote:
| I suppose it would help if I could read the whole page: I
| cannot see the left few characters on Firefox on Android.
| What did you make this with?
|
| Note that fixing the site won't increase my chances of
| donating, I'm from the ASML country ;)
| nish__ wrote:
| That's annoying... I made it with Next.js and Tailwind
| CSS tho. Hosted on Vercel.
| collinmanderson wrote:
| I'll happily invest $100 into this project in exchange for
| stock. let me know.
| jacquesm wrote:
| Kickstarter time? That would set some records.
| tmaly wrote:
| that is an understatement
| bhhaskin wrote:
| Only if you put up the 10 billion dollars.
| nish__ wrote:
| Machines are less than 400 million.
| pixl97 wrote:
| And the cost of the people to run those machines, and the
| factories that are required to run the machines?
| waynesonfire wrote:
| Dont forget RAM.
| nish__ wrote:
| I'll do it for free. And I'm sure we could rent a
| facility.
| q3k wrote:
| You're just talking about a lithography machine. Patterning
| is one step out of thousands in a modern process (albeit an
| important one). There's plenty more stuff needed for a
| production line, this isn't a 3D printer but for chips. And
| that's just for the FEOL stuff, then you still need to do
| BEOL :). And packaging. And testing
| (accelerated/environmental, too). And failure analysis.
| And...
|
| Also, you know, there's a whole process you'll need to
| develop. So prepare to be not making money (but spending
| tons of it on running the lines) until you have a well
| tested PDK.
| baq wrote:
| > 3D printer but for chips
|
| how about a farm of electron microscopes? these should
| work
| vel0city wrote:
| Canon has been working on an alternative to EUV
| lithography called nanoimprint lithography. It would be a
| bit closer to the idea of having an inkjet printer make
| the masks to etch the wafers. It hasn't been proven in
| scale and there's a lot of thinking this won't really be
| useful, but it's neat to see and maybe the detractors are
| wrong.
|
| https://global.canon/en/technology/nil-2023.html
|
| https://newsletter.semianalysis.com/p/nanoimprint-
| lithograph...
|
| They'll still probably require a good bit of operator and
| designer knowledge to work around whatever rough edges
| exist in the technology to keep yields high, assuming it
| works. It's still not a "plug it in, feed it blank
| wafers, press PRINT, and out comes finished chips!" kind
| of machine some here seem to think exist.
| GeekFortyTwo wrote:
| As someone with no skills in the space, no money, and lives
| near Ottawa: I'd love to help start a fab in Ontario.
| nish__ wrote:
| Right on, partner. I think there's a ton of demand for it tbh
|
| I'll do the engineering so we're good on that front. Just
| need investors.
| mindcrime wrote:
| I think I have a few dollars left on a Starbucks gift card.
| I'm in!
| deuplonicus wrote:
| Sure, we can work on brining in TinyTapeout to modern fab
| Reason077 wrote:
| A dozen or so well-resourced tech titans in China are no doubt
| asking themselves this same question right now.
|
| Of course, it takes quite some time for a fab to go from an
| idea to mass production. Even in China. Expect prices to drop
| 2-3 years from now when all the new capacity comes online?
| dylan604 wrote:
| At that point, it'll be the opposite problem as more capacity
| than demand will be available. These new fabs won't be able
| to pay for themselves. Every tic receives a tok.
| umanwizard wrote:
| China cannot buy ASML machines. All advanced semiconductor
| manufacturing in China is done with stockpiled ASML machines
| from before the ban.
| phil21 wrote:
| Would you really need ASML machines to do DDR5 RAM? Honest
| question, but I figured there was competition for the non-
| bleeding edge - perhaps naively so.
| nish__ wrote:
| Yes. You need 16nm or better for DDR5.
| squigz wrote:
| As someone who knows next to nothing about this space, why
| can China not build their own machines? Is ASML the only
| company making those machines? If so, why? Is it a matter
| of patents, or is the knowledge required for this so
| specialized only they've built it up?
| nish__ wrote:
| Yes. ASML is the only company making these machines. And
| both, they own thousands of patents and are also the only
| ones with the institutional knowledge required to build
| them anyway.
| baq wrote:
| I thought Fermi paradox is about nukes; I increasingly
| think it's about chips
| bluGill wrote:
| They can - if they are willing to invest a a lot of money
| over several years. The US got Nuclear bombs in a few
| years during WWII with this thinking, and China (or
| anyone else) could too. This problem might be harder than
| a bomb, but the point remains, all it takes is a
| willingness to invest.
|
| Of course the problem is we don't see what would be
| missed by doing this investment. If you put extra people
| into solving this problem that means less people curing
| cancer or whatever. (China has a lot of people, but not
| unlimited)
| umanwizard wrote:
| Designing such an advanced lithography machine is
| probably a Manhattan Project scale task. China is indeed
| trying to make progress and they will probably get there
| someday but for now ASML is the only company in the world
| that knows how to build these or anything remotely close.
| tooltalk wrote:
| That restriction is only for the most advanced systems.
| According to ASML's Q3 2025 filing, 42% of all system sales
| went to China.
|
| SK Hynix also has significant memory manufacturing presence
| in China; or about 40% of the company's entire DRAM
| capacity.
| UncleOxidant wrote:
| I think it would be more like 5-7 years from now if they
| started breaking ground on new fabs today.
| nish__ wrote:
| My napkin math:
|
| According to my research, these machines can etch around 150
| wafers per hour and each wafer can fit around 50 top-of-the-
| line GPUs. This means we can produce around 7500 AI chips per
| hour. Sell them for $1k a piece. That's $7.5 million per hour
| in revenue. Run the thing for 3 days and we recover costs.
|
| I'm sure there's more involved but that sounds like a pretty
| good ROI to me.
| Keyframe wrote:
| that's 100% yield which ain't happening
| nish__ wrote:
| What would you expect yield to be?
| mindslight wrote:
| Sounds like the kind of question ChatGPT would be good at
| answering...
| danparsonson wrote:
| With no prior experience? 0%. Those machines are not just
| like printers :-)
| nish__ wrote:
| We'll have to gain some experience then :)
| danparsonson wrote:
| Sure - once you have dozens of engineers and 5 years
| under your belt you'll be good to go!
|
| This will get you started: https://youtu.be/B2482h_TNwg
|
| Keep in mind that every wafer makes multiple trips around
| the fab, and on each trip it visits multiple machines.
| Broadly, one trip lays down one layer, and you may need
| 80-100 layers (although I guess DRAM will be fewer). Each
| layer must be aligned to nanometer precision with
| previous layers, otherwise the wafer is junk.
|
| Then as others have said, once you finish the wafer, you
| still need to slice it, test the dies, and then package
| them.
|
| Plus all the other stuff....
|
| You'll need billions in investment, not millions - good
| luck!
| vel0city wrote:
| Especially when the plan is to just run them in a random
| rented commercial warehouse.
|
| I drive by a large fab most days of the week. A few
| breweries I like are down the street from a few small
| boutique fabs. I got to play with some experimental fab
| equipment in college. These aren't just some quickly
| thrown together spaces in any random warehouse.
|
| And it's also ignoring the water manufacturing process,
| and having the right supply chain to receive and handle
| these ultra clean discs without introducing lots of gunk
| into your space.
| warmwaffles wrote:
| Not with that attitude.
| nish__ wrote:
| My man.
| warmwaffles wrote:
| A man can dream of no waste!
| shadowpho wrote:
| What about the $10b to build the facility (including clean
| air/water/chemicals/etc)?
| nish__ wrote:
| Rent a warehouse.
| ghc wrote:
| It would be cheaper to bulldoze the warehouse and start
| over.
| bigiain wrote:
| Rent a warehouse in one of the non Han dominated areas of
| China, where you can use all you want from the city's
| drinking water supply and pump all your used chemicals
| into the nearby river. Make sure to totally automate your
| robotic production line so you don't need to employ any
| locals.
| shadowpho wrote:
| And then spend the next five years building the actual
| fab around it? Like what's the plan here.
| bluGill wrote:
| The catch is if you started today with plenty of money
| (billions of dollars!) and could hire the right experts as
| you need them (this is a big if!) there would still be a
| couple years between today and producing 150 wafers per
| house. So the question isn't what does the math look like
| today, it is what the math looks like in 2 years - if you
| could answer that why didn't you start two years ago so you
| could get the current prices?
| q3k wrote:
| A photolithography machine doesn't etch anything (well,
| some EUV machines do it as an unwanted side effect because
| of plasma generation), it just patterns some resist
| material. Etching is happening elsewhere. Also, keep in
| mind, you'll need to do multiple passes through a
| photolithography machine to pattern different steps of the
| process - it's not a single pass thing.
| retube wrote:
| yeaaahhh.. these fabs don't just have a button that says
| "punch out a top of the line nvidia GPU"
| Keyframe wrote:
| it's just a bunch of melted sand. How hard can it be?
| asjir wrote:
| But what if it's a bubble driven by speculation?
|
| It wouldn't pay off.
|
| Starting a futures exchange on RAM chips, on the other hand...
| panzagl wrote:
| Hear me out- artisanal DRAM.
| nish__ wrote:
| I'm listening.
| q3k wrote:
| Bathtub semiconductors! Shameless plug: https://wiki.fa-
| fo.de/orga:donate
| layer8 wrote:
| Every transistor carefully hand-wired with love, giving each
| bit its own individual characteristic, for more organic,
| natural data.
| bigiain wrote:
| You've cracked the funding problem!
|
| You need to sell the artisanal DRAM to HiFi hucksters. Get
| Monster Cable on board to say their cables can only reach
| peak warmth, punch, and clarity - when using gear with
| artisanal DRAM. You'll easily sell into that market at $10k
| per GB...
| bigiain wrote:
| Locally sourced artisanal DRAM
| gizmo wrote:
| The big 3 memory manufacturers (SK Hynix, Samsung, Micron) are
| essentially all moving upmarket. They have limited capacity and
| want to use it for high margin HBM for GPUs and ddr5 for servers.
| At the same time CXMT, Winbond and Nanya are stepping in at the
| lower end of the market.
|
| I don't think there is a conspiracy or price fixing going on
| here. Demand for high profit margin memory is insatiable (at
| least until 2027 maybe beyond) and by the time extra capacity
| comes online and the memory crunch eases the minor memory players
| will have captured such a large part of the legacy/consumer
| market that it makes little sense for the big 3 to get involved
| anymore.
|
| Add to that scars from overbuilding capacity during previous
| super memory super cycles and you end up with this perfect storm.
| mixermachine wrote:
| Let's see if this demand truly holds. I'm still unsure.
| Currently nobody makes money with AI. There will be a
| correction (if I can trust my magic crystal ball ;) )
| stevenjgarner wrote:
| My understanding is that this is primarily hitting DDR5 RAM (or
| better). With prices so inflated, is there an argument to revert
| and downgrade systems to DDR4 RAM technology in many use cases
| (which is not so inflated)?
| cptnapalm wrote:
| DDR 4 shot up too. It was bad enough that instead of trying to
| put together a system with the AM4 m/b I already have, I just
| bought a Legion Go S.
| Numerlor wrote:
| DDR4 manufacturing is mostly shut down, so if any real demand
| picks starts there the prices will shoot up
| gizmo wrote:
| No DDR4 is affected too. It's a simple question of production
| and demand, and the biggest memory manufacturers are all
| winding down their DDR4/DDR5 memory production for consumers
| (they still make some DDR5 for OEMS and servers).
| tencentshill wrote:
| It will be hit just as hard, they have stopped new DDR4
| production to focus on DDR5 and HBM.
| segmondy wrote:
| DDR4 prices have gone up 4x in the last 3 months.
| geerlingguy wrote:
| Linked in the article, DDR4 and LPDDR4 are also 2-4x more
| expensive now, forcing smaller manufactures to raise prices or
| cancel some products entirely.
| phendrenad2 wrote:
| I'm way ahead of all of you, I'm hoarding DDR2.
| altmanaltman wrote:
| I mean what's the big deal, can't we just download more ram
| submeta wrote:
| Very happy to have bought a MBP M4 with 64 gb of Ram last year.
| wslh wrote:
| And sharing the RAM between your CPU/GPU/NPU instead of using
| separate memories.
| SchwKatze wrote:
| I just gathered enough money to build my new PC. I'll even go to
| another country to pay less taxes, and this spike hit me hard.
| I'll buy anyway because I don't believe it will slow down so
| soon. But yeah, for me is a lot of money
| rolandog wrote:
| Me too. I had saved up to make a small server. I guess I'll
| have to wait up until 2027-2028 at this rate.
| dylan604 wrote:
| Buy used gear and rip out the guts???
| comeonbro wrote:
| This is ultimately the first stage of human economic obsolescence
| and extinction.
|
| This
| https://cdna.pcpartpicker.com/static/forever/images/trends/2...
| will happen to every class of thing (once it hits energy,
| everything is downstream of energy).
| kalterdev wrote:
| Why should we believe in another apocalypse prediction?
| IAmBroom wrote:
| One of them has to be right, eventually!
| gooseus wrote:
| Because the collapse of complex societies is real - https://g
| ithub.com/danielmkarlsson/library/blob/master/Josep...
|
| Unbounded increases in complexity lead to diminishing returns
| on energy investment and increased system fragility which
| both contribute to an increased likelihood of collapse as
| solutions to old problems generate new problems faster than
| new solutions can be created since energy that should be
| dedicated to new solutions is needed to maintain the layers
| of complexity generated by the layers of previous solutions.
| captainkrtek wrote:
| I'm no economist, but if (when?) the AI bubble bursts and
| demand collapses at the price point memory and other related
| components are at, wouldn't price recover?
|
| not trying to argue, just curious.
| reducesuffering wrote:
| _IF_ a theoretical AI bubble bursts sure. However the largest
| capitalized companies in the world and all the smartest
| people able to do cutting edge AI research are betting
| otherwise. This is also what the start of a takeoff looks
| like
| squidbeak wrote:
| I'm no economist either, but I imagine the manufacturing
| processes for the two types of RAM are too different for
| supply to quickly bounce back.
| benlivengood wrote:
| If your argument is that value produced per-cpu will increase
| so significantly that the value produced by AGI/ASI per unit
| cost exceeds what humans can produce for their upkeep in food
| and shelter, then yes that seems to be one of the significant
| risks long term if governments don't intervene.
|
| If the argument is that prices will skyrocket simply because of
| long-term AI demand, I think that ignores the fact that
| manufacturing vastly more products will stabilize prices up to
| the point that raw materials start to become significantly more
| expensive, and is strongly incentivized over the ~10-year
| timeframe for IC manufacturers.
| kace91 wrote:
| >the value produced by AGI/ASI per unit cost exceeds what
| humans can produce for their upkeep in food and shelter
|
| The value of AGI/ASI is not only defined by its practical
| use, It is also bounded by the purchasing power of potential
| consumers.
|
| If humans aren't worth paying, those humans won't be paying
| anyone either. No business can function without customers, no
| matter how good the product.
| benlivengood wrote:
| Precisely the place where government intervention is
| required to distribute wealth equitably.
| asdfman123 wrote:
| I can't help but be the pessimist angle. RAM production will need
| to increase to supply AI data centers. When the AI bubble bursts
| (and I do believe it will), the whole computing supply chain,
| which has been built around it, will take a huge hit too. Excess
| production capacity.
|
| Wonder what would happen if it really takes a dive. The impact on
| the SF tech scene will be brutal. Maybe I'll go escape on a
| sailboat for 3 years or something.
|
| Anyway, tangential, but something I think about occasionally.
| XorNot wrote:
| Prices are high because no one believes it's not a bubble.
| Nvidias strategy has been careful careful with volume this
| whole time as well.
|
| The thing is it's also not a conventional looking bubble: what
| we're seeing here is cashed up companies ploughing money into
| the only thing in their core business they could find to do so
| with, rather then a lot of over exuberant public trading and
| debt financing.
| roadside_picnic wrote:
| I know this is mostly paranoid thinking on my behalf, but it
| almost feels like this is a conscious effort to attempt to
| destroy "personal" computing.
|
| I've been a huge advocate for local, open, generative AI as the
| best resistance to massive take-over by large corporations
| controlling all of this content creation. But even as it is (or
| "was" I should say), running decent models at home is
| prohibitively expensive for most people.
|
| Micron has already decided to just eliminate the Crucial brand
| (as mentioned in the post). It feels like if this continues, once
| our nice home PCs start to break, we won't be able to repair
| them.
|
| The extreme version of this is that even dumb terminals (which
| still require _some_ ram) will be as expensive as laptops today.
| In this world, our entire computing experience is connecting a
| dumb terminal to a ChatGPT interface where the only way we can
| interact with anything is through "agents" and prompts.
|
| In this world, OpenAI is _not_ overvalued, and there is no bubble
| because the large LLM companies _become computing_.
|
| But again, I think this is mostly a dystopian sci-fi fiction...
| but it does sit a bit too close to the realm of possible for my
| tastes.
| ben_w wrote:
| The first "proper" "modern" computer I had, initially came with
| 8 megabytes of RAM.
|
| It's not a lot, but it's enough for a dumb terminal.
| rolandog wrote:
| That's not disproving OP's comment; OpenAI is, in my opinion,
| making it untenable for a regular Joe to build a PC capable
| of running local LLM model. It's an attack on all our
| wallets.
| petre wrote:
| Why do you need a LLM running locally so much that's the
| inflated RAM prices are an attack on your wallet? One can
| always opt not to play this losing game.
|
| I remember when the crypto miners rented a plane to deliver
| their precious GPUs.
| plufz wrote:
| I don't think you need a conspiracy theory to explain this.
| This is simply capitalism, a system that seems less and less
| like the way forward. I'm not against markets, but I believe
| most countries need more regulations targeted at the biggest
| companies and richest people. We need stronger welfare states,
| smaller income gaps and more democracy. But most countries
| seems to vote in the absolute opposite direction.
| BizarroLand wrote:
| The end goal of capitalism is the same as the end goal of
| monopoly.
|
| 1 person has all the money and all the power and everyone
| else is bankrupt forever and sad.
| dangus wrote:
| It's not a conspiracy, it's just typical dumb short term
| business decisions amplified and enabled by a cartel supply
| market.
|
| If Crucial screws up by closing their consumer business they
| won't feel any pain from it because the idea of new competitors
| entering the space is basically impossible.
| SimianSci wrote:
| Wouldn't the easy answer to this be increased efficiency of RAM
| usage?
|
| RAM being plentiful and cheap led to a lot of software
| development being very RAM-unaware, allowing the inefficiencies
| of programs to be mostly obfuscated from the user. If RAM
| prices continue rising, the semi-apocalytic consumer fiction
| you've spun here would require that developers not change their
| behaviors when it comes to software they write. There will be
| an equillibrium in the market that still allows the entry of
| consumer PC's it will just mean devices people buy will have
| less available RAM than is typical. The demand will eventually
| match up to the change in supply as is typical of supply/demand
| issues and not continuously rise into an infinite horizon.
| kalterdev wrote:
| I believe that while centralized computing excels at specific
| tasks like consumer storage, it cannot _compete_ with the
| unmatched diversity and unique intrinsic benefits of personal
| computing. Kindle cannot replace all e-readers. Even Apple's
| closed ecosystem cannot permit it to replace macOS with iPadOS.
| These are not preferences but constraints of reality.
|
| The goal shouldn't be to eliminate one side or the other, but
| to bridge the gap separating them. Let vscode.dev handle the
| most common cases, but preserve vscode.exe for the uncommon yet
| critical ones.
| clusterhacks wrote:
| I share your paranoia.
|
| My kids use personal computing devices for school, but their
| primary platform (just like their friends) is locked-down
| phones. Combining that usage pattern with business incentives
| to lock users into walled gardens, I kind of worry we are
| backing into the destruction of personal computing.
| walterbell wrote:
| If the OpenAI Hodling Company buys and warehouses 40% of global
| memory production or 900,000 memory wafers (i.e. not yet turned
| into DDR5/DDR6 DIMMs) per month at price X in October 2025,
| leading to supply shortages and tripling of price, they have the
| option of later un-holding the warehoused memory wafers for a
| profit.
|
| https://news.ycombinator.com/item?id=46142100#46143535
| Had Samsung known SK Hynix was about to commit a similar chunk of
| supply -- or vice-versa -- the pricing and terms would have
| likely been different. It's entirely conceivable they wouldn't
| have both agreed to supply such a substantial part of global
| supply if they had known more...but at the end of the day -
| OpenAI did succeed in keeping the circles tight, locking down the
| NDAs, and leveraging the fact that these companies assumed the
| other wasn't giving up this much wafer volume simultaneously...in
| order to make a surgical strike on the global RAM supply chain..
|
| What's the economic value per warehoused and insured cubic inch
| of 900,000 memory wafers? Grok response:
|
| _> As of late 2025, 900,000 finished 300 mm 3D NAND memory
| wafers (typical high-volume inventory for a major memory maker)
| are worth roughly $9 billion and occupy about 104-105 million
| cubic inches when properly warehoused in FOUPs. - Economic value
| [?] $85-90 per warehoused cubic inch._
| petre wrote:
| Sounds like the Silver Thursday all over again. I hope OpenAI
| ends up like the Hunt Btothers.
|
| https://en.wikipedia.org/wiki/Silver_Thursday
| enlyth wrote:
| > To save the situation, a consortium of US banks provided a
| $1.1 billion line of credit to the brothers which allowed
| them to pay Bache which, in turn, survived the ordeal.
|
| It seems once you amass a certain amount of wealth, you just
| get automatically bailed out from your mistakes
| radicality wrote:
| Think the article should also mention how OpenAI is likely
| responsible for it. Good article I found from another thread here
| yesterday: https://www.mooreslawisdead.com/post/sam-altman-s-
| dirty-dram...
| diabllicseagull wrote:
| yes. on the moore's law is dead podcast they were talking about
| rumors where some 'AI enterprise company's representatives'
| were trying to buy memory in bulk from brick and mortar stores.
| in some cases openai was mentioned. crazy if true. also
| interesting considering none of those would be ECC certified
| like what you would opt for for a commercial server.
| mschuster91 wrote:
| > The reason for all this, of course, is AI datacenter buildouts.
| I have no clue if there's any price fixing going on like there
| was a few decades ago--that's something conspiracy theorists can
| debate--but the problem is there's only a few companies producing
| all the world's memory supplies.
|
| So it's the Bitcoin craze all over again. Sigh. The bubble will
| eventually collapse, it _has to_ - but the markets can stay
| irrational longer than you can stay solvent... or, to use a more
| appropriate comparison, have a working computer.
|
| I for myself? I hope once this bubble collapses, we see actual
| _punishments_ again. Too-large-to-fail companies broken up,
| people getting prosecuted for the wash trading masquerading
| itself as "legitimate investments" in the entire bubble (that
| more looks like the genetic family table of the infamously
| incestuous Habsburg family), greedy executives jailed or, at
| least where national security is impacted due to chip shortages,
| permanently gotten rid of. I'm sick and tired of large companies
| being able to just get away with gobbling up everything, killing
| off the economy at large, they are not just parasites - they are
| a cancer, killing its host society.
| otherayden wrote:
| I think we kiss of deathed the article haha. Here's an archive
| https://archive.is/6QD8c
| carlCarlCarlCar wrote:
| Pricing compute out of the average persons budget to prop up
| investment in data centers, stocks, ultimately control agency
|
| If an RTX 5000 series price topped out at historical prices no
| one would need hosted AI
|
| Then it came to be that models were on a path to run well enough
| loaded into RAM... uh oh
|
| This is in line with ISPs long ago banning running personal
| services and the long held desire to sell dumb thin clients that
| must work with a central service
|
| Web developers fell for confidence games of old elders hook line
| and sinker. Nothing but the insane ego and vanity of some tech
| oligarchs driving this. They cannot appear weak. Vain aura
| farming, projection of strength.
| zoobab wrote:
| "dig into that pile of old projects you never finished instead of
| buying something new this year."
|
| You don't need a new PC. Just use the old one.
| kevin_thibedeau wrote:
| I just bought some 30pin SIMMs to rehab an old computer. That
| market is fine.
| fullstop wrote:
| I have a bag of SIMMs that I saved, no idea why, because I
| clearly wrote BAD on the mylar bag.
|
| At time time I was messing around with the "badram" patch for
| Linux.
| ogogmad wrote:
| Every shortage is followed by a glut. Wait and see for RAM prices
| to go way down. This will happen because RAM makers are racing to
| produce units to reap profits from the higher price. That
| overproduction will cause prices to crash.
| barnas2 wrote:
| They aren't overproducing consumer modules, they're actively
| cutting production of those. They're producing datacenter/AI
| specific form factors that won't be compatible with consumer
| hardware.
| baq wrote:
| _somebody_ will step up to pick up the free money if this
| continues.
| ok_dad wrote:
| I am very excited for a few years when the bubble bursts and all
| this hardware is on the market for cheap like back in the early
| to mid 2000's after that bubble burst and you had tons of old
| servers available for homelabs. I can't wait to fill a room with
| 50kW of bulk GPUs on a pallet and run some cool shit.
| JKCalhoun wrote:
| > maybe it's a good time to dig into that pile of old projects
| you never finished instead of buying something new this year.
|
| Always good advice.
| haunter wrote:
| panem et circenses
|
| But what will happen when people are priced out from the circus?
| mikelitoris wrote:
| Companies are adamant about RAMming AI down our throats, it
| seems.
| AceJohnny2 wrote:
| > _And those companies all realized they can make billions more
| dollars making RAM just for AI datacenter products, and neglect
| the rest of the market._
|
| I wouldn't ascribe that much intent. More simply, datacenter
| builders have bought up the entire supply (and likely future
| production for some time), hence the supply shortfall.
|
| This is a very simple supply-and-demand situation, nothing
| nefarious about it.
| teamonkey wrote:
| That makes it sound like they are powerless, which is not the
| case. They don't _have_ to have their capacity fully bought
| out, they could choose to keep a proportion of capacity for
| maintaining the existing PC market, which they would do if they
| thought it would benefit them in the long term.
|
| They're not doing that, because it benefits them not to.
| baq wrote:
| $20B, 5 years and you can have your own DDR5 fab to print
| money with.
|
| jokes aside, if the AI demand actually materializes, somebody
| will look at the above calculation and say 'we're doing it in
| 12 months' with a completely straight face - incumbents'
| margin will be the upstart's opportunity.
| MPSimmons wrote:
| This reminds me of the recent LaurieWired video presenting a
| hypothetical of, "what if we stopped making CPUs":
| https://www.youtube.com/watch?v=L2OJFqs8bUk
|
| Spoiler, but the answer is basically that old hardware rules the
| day because it lasts longer and is more reliable of timespans of
| decades.
|
| DDR5 32GB is currently going for ~$330 on Amazon
|
| DDR4 32GB is currently going for ~$130 on Amazon
|
| DDR3 32GB is currently going for ~50 on Amazon (4x8GB)
|
| For anyone where cost is a concern, using older hardware seems
| like a particularly easy choice, especially if a person is
| comfortable with a Linux environment, since the massive droves of
| recently retired Windows 10 incompatible hardware works great
| with your Linux distro of choice.
| fullstop wrote:
| If everyone went for DDR4 and DDR3, surely the cost would go
| up. There is no additional supply there, as they are no longer
| being made.
| MPSimmons wrote:
| Undoubtably the cost would go up, but nobody is building out
| datacenters full of DDR4, either, so I don't figure it would
| go up nearly as much as DDR5 is right now.
| fullstop wrote:
| https://pcpartpicker.com/trends/price/memory/
|
| You can see the cost rise of DDR4 here.
| MPSimmons wrote:
| Awesome charts, thanks! I think it bears out that the
| multiplier for older hardware isn't as extreme as the
| newer hardware, right?
| fullstop wrote:
| ~2.8x for DDR4, ~3.6x for DDR5. DDR5 is still being made,
| though, so it will be interesting to see how it changes
| in the future.
|
| Either way, it's going to be a long few years at the
| least.
| oskarkk wrote:
| Unless the AI bubble pops.
| bombcar wrote:
| One possible outcome is the remaining memory
| manufacturers have dedicated all their capacity for AI
| and when the bubble pops, they lose their customer and
| they go out of business too.
| bigiain wrote:
| I wouldn't be too surprised to find at least some of the
| big ram foundries are deeply bought into the fake money
| circles where everybody is "paying" each other with
| unrealised equity in OpenAI/Anthropic/whoever, resulting
| in a few trillion dollars worth of on-paper "money"
| vanishing overnight, at which stage a whole bunch of
| actual-money loans will get called in and billion dollar
| companies get gutted by asset strippers.
|
| Maybe larger makerspaces and companies like Adafruit,
| RasPi, and Pine should start stockpiling (real) money,
| and pick themselves up an entire fab full of gear at
| firesale prices so they can start making their own ram...
| parineum wrote:
| That's the Avarage price for new DDR4 which has dwindling
| supply. Meanwhile used DDR4 is being retired in both
| desktops and data centers.
| fullstop wrote:
| DDR4 production is winding down or done. Only "new old
| stock" will remain, or used DDR4 modules. Good luck
| buying that in quantity.
| acdha wrote:
| At some point that's true, but don't they run the n-1 or 2
| generation production lines for years after the next
| generation launches? There's a significant capital investment
| there and my understanding is that the cost goes down
| significantly over the lifetime as they dial in the process
| so even though the price is lower it's still profitable.
| fullstop wrote:
| Unless plans have changed, the foundries making DDR4 are
| winding down, with the last shipments going out as we
| speak.
| my123 wrote:
| The specialty DRAM vendors (Nanya notably) will keep
| making DDR4 and earlier
| mlyle wrote:
| This is only true as long as there's enough of a market
| left. You tend to end up with oversupply and excessive
| inventory during the transition, and that pushes profit
| margins negative and removes all the supply pretty quickly.
| phil21 wrote:
| Currently trying to source a large amount of DDR4 to upgrade
| a 3 year old fleet of servers at a very unfortunate time.
|
| It's very difficult to source in quantity, and is going up in
| price more or less daily at this point. Vendor quotes are
| good for hours, not days when you can find it.
| fullstop wrote:
| We're looking to buy 1TB of DDR5 for a large database
| server. I'm pretty sure that we're just going to pay the
| premium and move on.
| georgeburdell wrote:
| And that's why everybody's B2B these days. The decision-
| making people at companies are not spending their own
| money
| derefr wrote:
| Have you considered dropping by in person to your nearest
| computer recycler/refurbisher? As a teen I worked at one,
| and the boxes and boxes of RAM sticks pulled from scrapped
| machines (usually scrapped due to bad main boards) made a
| strong impression. They tend to not even put the highest-
| spec stuff they pull into anything, as refurbished machines
| are mostly sold/donated in quantity (to schools and the
| like) and big customers want standardized SKUs with
| interchangeable specs for repairability more than they want
| performance. Workers at these places are often invited to
| build personal machines to take home out of these "too
| good" parts -- and yet there's so much that they can't
| possibly use it all up that way. If someone showed up
| asking if they could have some DDR3, they might literally
| just let you dig through the box and then sell it to you by
| weight!
| christkv wrote:
| A friend built a new rig and went with DDR4 and a 5800x3d just
| because of this as he needed a lot of ram.
| wkat4242 wrote:
| I have exactly this and was planning to upgrade to AM5 soon
| :'(
|
| Looks like AM4 will live on for many more years in my flat
| diabllicseagull wrote:
| a 2x16 ddr4 kit I bought in 2020 for $160 is now $220. older
| memory is relatively cheap but not cheaper than before at all.
| shoo wrote:
| I wondered how much of this is inflation -- after adjusting
| for CPI inflation, $160 in 2020 is worth $200 in today's
| dollars [$], so the price of that ddr4 kit is 10% higher in
| real terms.
|
| [$] https://www.usinflationcalculator.com/
| fullstop wrote:
| USD is also weaker than it was in the past.
| bullen wrote:
| Yes, DDR3 is the lowest CAS latency and lasts ALOT longer.
|
| Just like SSDs from 2010 have 100.000 writes per bit instead of
| below 10.000.
|
| CPUs might even follow the same durability pattern but that
| remains to be seen.
|
| Keep your old machines alive and backed up!
| mrob wrote:
| CAS latency is specified in cycles and clock rates are
| increasing, so despite the number getting bigger there's
| actually been a small improvement in latency with each
| generation.
| bullen wrote:
| Not for small amounts of data.
|
| Bandwith increases, but if you only need a few bytes DDR3
| is faster.
|
| Also slower speed means less heat and longer life.
|
| You can feel the speed advantage by just moving the mouse
| on a DDR3 PC...
| mrob wrote:
| RAM latency doesn't affect mouse response in any
| perceptible way. The fastest gaming mice I know of run at
| 8000Hz, so that's 125000ns between samples, much bigger
| than any CAS latency. And most mice run substantially
| slower.
|
| Maybe your old PC used lower-latency GUI software, e.g.
| uncomposited Xorg instead of Wayland.
| bullen wrote:
| I only felt it on Windows, maybe tht is due to the
| special USB mouse drivers Microsoft made? Still motion-
| to-photon latency is really lower on my DDR3 PCs, would
| be cool to know why.
| CyberDildonics wrote:
| You are conflating two things that have nothing to do
| with each other. Computers have had mice since the 80s.
|
| _Still motion-to-photon latency is really lower on my
| DDR3 PCs, would be cool to know why._
|
| No it isn't, your computer is doing tons of stuff and the
| cursor on windows is a hardware feature of the graphics
| card.
|
| Should I even ask why you think memory bandwidth is the
| cause of mouse latency?
| kbolino wrote:
| While it has nothing to do with how responsive your mouse
| feels, as that is measured in milliseconds while CAS
| latency is measured in nanoseconds, there has indeed been
| a small regression with DDR5 memory compared to the 3
| previous generations. The best DDR2-4 configurations
| could fetch 1 word in about 6-7 ns while the best DDR5
| configurations take about 9-10 ns.
|
| https://en.wikipedia.org/wiki/CAS_latency#Memory_timing_e
| xam...
| hajile wrote:
| CAS latency doesn't matter so much as ns of total random-
| access latency and the raw clockspeed of the individual RAM
| cells. If you are accessing the same cell repeatedly, RAM
| hasn't gotten faster in years (around DDR2 IIRC).
| 0manrho wrote:
| > 100.000 writes per bit
|
| per cell*
|
| Also, that SSD example is wildly untrue. Especially with the
| context of available capacity at the time. You CAN get modern
| SSD's with mind boggling write endurance per cell, AND has
| multides more cells, resulting in vastly more durable media
| than what was available pre 2015. The one caveat there to
| modern stuff being better than older stuff is Optane (the
| enterprise stuff like the 905P or P5800X, not that memory and
| SSD combo shitshow that Intel was shoveling out the consumer
| door). We still haven't reached parity with the 3DXpoint
| stuff, and it's a damn shame Intel hurt itself in it's
| confusion and cancelled that, because boy would they and
| Micron be printing money hand over fist right now if they
| were still making them. Still, Point being: Not everything is
| a TLC/QLC 0.3DWPD disposable drive like has become standard
| in the consumer space. If you want write endurance, capacity,
| and/or performance, you have more and better options today
| than ever before (Optane/3DXPoint excepted).
|
| Regarding CPU's, they still follow that durability pattern if
| you unfuck what Intel and AMD are doing with boosting
| behavior and limit them to perform with the margins that they
| used to "back in the day". This is more of a problem on the
| consumer side (Core/Ryzen) than the enterprise side
| (Epyc/Xeon). It's also part of why the OC market is dying
| (save for maybe the XOC market that is having fun with LN2),
| those CPU's (especially consumer ones) come from the factory
| with _much_ less margin for pushing things, because they 're
| already close to their limit without exceedingly robust
| cooling.
|
| I have no idea what the relative durability of RAM is tbh,
| it's been pretty bulletproof in my experience over the years,
| or at least bulletproof enough for my usecases that I haven't
| really noticed a difference. Notable exception is what I see
| in GPU's, but that is largely heat-death related and often a
| result of poor QA by the AIB that made it (eg, thermal pads
| not making contact with the GDDR modules).
| bullen wrote:
| Maybe but in my experience a good old <100GB SSD from
| 2010-14 will completely demolish any >100GB from 2014+ in
| longevity.
|
| Some say they have the opposite experience, mine is ONLY
| Intel drives, maybe that is why.
|
| X25-E is the diamond peak of SSDs probably forever since
| the machines to make 45nm SLC are gone.
| zozbot234 wrote:
| What if you overprovision the newer SSD to a point where
| it can run the entirety of the drive in pseudo-SLC
| ("caching") mode? (You'd need to store no more than 25%
| of the nominal capacity, since QLC has four bits per
| cell.) That should have fairly good endurance, though
| still a lot less than Optane/XPoint persistent memory.
| thedangler wrote:
| Nice, My current PC uses DDR4 Time to dust off my 2012 PC and
| put Linux on it.
| xboxnolifes wrote:
| Unfortunately, older RAM also means an older motherboard, which
| also means older socket and older CPUs. It works, but it's not
| usually a drop in replacement.
| maximilianthe1 wrote:
| Can't you use DDR3 in DDR5 compatible board?
| forthefuture wrote:
| Unfortunately not, each version of RAM uses a different
| physical slot.
| layer8 wrote:
| No, the different generations have fundamentally
| incompatible interfaces, not just mechanically, but in
| terms of voltage and signaling.
| NexRebular wrote:
| > works great with your Linux distro of choice.
|
| ...or you could go with FreeBSD. There's even a brand new
| release that just came out!
|
| https://www.freebsd.org/releases/15.0R/announce/
| tucnak wrote:
| I bought 384 GB of DDR5-4800 RDIMM a few months back for a Zen
| 4c system with lots of I/O like dual 25G NIC's and ten MCIO x8
| ports... So far it has been the best value for money compared
| to any memory before it. The bandwidth is nuts. Power
| consumption went DOWN compared to DDR4. Doesn't matter much if
| you got two sticks, but as soon as you get into 6+ territory,
| it does matter a lot. The same goes for NVMe's.
| jl6 wrote:
| Can someone explain why OpenAI is buying DDR5 RAM specifically? I
| thought LLMs typically ran on GPUs with specialised VRAM, not on
| main system memory. Have they figured out how to scale using
| regular RAM?
| fullstop wrote:
| They're not. They are buying wafers / production capacity to
| make HBM so there is less DDR5 supply.
| jl6 wrote:
| OK, fair enough, but what are OpenAI doing buying production
| capacity rather than, say, paying NVIDIA to do it? OpenAI
| aren't the ones making the hardware?
| fullstop wrote:
| Because they can provide the materials to NVIDIA for
| production and prevent Google, Anthropic, etc from having
| them.
| snuxoll wrote:
| Just because Nvidia happily sells people discrete GPU's,
| DGX systems, etc., doesn't mean they would turn down a
| company like OpenAI paying them $$$ for just the packaged
| chips and the technical documentation to build their own
| PCBs; or, let OpenAI provide their own DRAM supply for
| production on an existing line.
|
| If you have a potentially multi-billion dollar contract,
| most businesses will do things outside of their standard
| product offerings to take in that revenue.
| jl6 wrote:
| Got it, thank you.
| vel0city wrote:
| > doesn't mean they would turn down a company like OpenAI
| paying them $$$ for just the packaged chips and the
| technical documentation to build their own PCBs
|
| FWIW, this was the standard state of affairs of the GPU
| market for a long while. nVidia and AMD sold the chips
| they paid someone to produce to integrators like EVGA,
| PNY, MSI, ZOTAC, GIGABYTE, etc. Cards sold under the AMD
| or nVidia name directly were usually partnered with one
| of these companies to actually build the board, place the
| RAM, design the cooling, etc. From a big picture
| perspective, it's a pretty recent thing for nVidia to
| only really deliver finished boards.
|
| On top of this, OpenAI/Sam Altman have been pretty open
| about making their own AI chips and what not. This might
| point to them getting closer to actually delivering on
| that (pure speculation) and wanting to ensure they have
| other needed supplies like RAM.
| baq wrote:
| > OpenAI aren't the ones making the hardware?
|
| how surprised would you be if they announced that they are?
| Thegn wrote:
| They didn't buy DDR5 - they bought raw wafer capacity and a ton
| of it at that.
| christkv wrote:
| I grabbed a framework desktop with 128GB due to this. I can't
| imagine they can keep the price down for the next batches. If you
| bought 128GB of ram with "close" specs to the one used just that
| would be 1200 EUR at retail (who are obviously taking advantage).
| RachelF wrote:
| Perhaps we'll have to start optimizing software for performance
| and RAM usage again.
|
| I look at MS Teams currently using 1.5GB of RAM doing nothing.
| 650REDHAIR wrote:
| I truly hate how bloated and inefficient MS/Windows is.
|
| My hope is that with the growing adoption of Linux that MS
| takes note...
| vanviegen wrote:
| That's a very indirect way of enjoying the benefits of Linux.
| bluGill wrote:
| Can we get web sites to optimize as well? I use a slower
| laptop and a lot of sites are terrible. My old Sparc station
| (40mhz) had a better web experience in 1997 because than
| people cared about this.
| conception wrote:
| So you've hated Windows for the last 30 years?
| alliao wrote:
| now that's a word i haven't heard in a while...optimising
| cube00 wrote:
| > optimising
|
| Usually followed by "stock holder value"
| bigiain wrote:
| And to the tech staff, some out of context and
| misunderstood quote from semi technical management which
| starts with the word "Premature"
| irae wrote:
| RIP electron apps and PWAs. Need to go native, as chromium
| based stuff is so memory hungry. PWAs on Safari use way less
| memory, but PWA support in Safari is not great.
| bigiain wrote:
| I, for one, would not miss a single one of the electron apps
| I'm forced to use.
|
| Every single one of them makes me feel like the vendor is
| telling me "we can't be bothered employing half decent
| developers or giving the developers we have enough time and
| resources to write decent software, so we're just going to
| use cheap and inexperience web developers and burn another
| gigabyte or two of your memory to run what could easily be a
| sub 100MB native app."
|
| At least now I'll have significant numbers to tell my boss:
| "Sure, we can continue to use
| Slack/VSCode/Teams/Figma/Postman/ - but each of those is
| going to require an additional GB or two of memory on every
| staff member's computer - which at today's pricing is over
| $500 in ram per laptop which are all on a 18-24 month
| replacement cycle. So that's maybe a million dollars a year
| in hardware budget to run those 5 applications across the
| whole team. We'll need to ensure we have signoff on that
| expenditure before we renew our subscriptions for those
| apps."
| BobaFloutist wrote:
| Your laptops have an 18-24 month replacement cycle? What
| are you guys _doing_ to the poor things?
| jjcm wrote:
| Likewise, Slack. Currently at 2.3GB.
|
| I'm very curious if this AI wave will actually lead to more
| native apps being made, since the barrier to doing so will be
| lower.
| UltraSane wrote:
| PCIe 5.0 NVMe drives used as page cache can help a lot.
| hathawsh wrote:
| The article suggests that because the power and cooling are
| customized, it would take a ton of effort to run the new AI
| servers in a home environment, but I'm skeptical of that. Home-
| level power and cooling are not difficult these days. I think
| when the next generation of AI hardware comes out (in 3-5 years),
| there will be a large supply of used AI hardware that we'll
| probably be able to repurpose. Maybe we'll sell them as parts. It
| won't be plug-and-play at first, but companies will spring up to
| figure it out.
|
| If not, what would these AI companies do with the huge supply of
| hardware they're going to want to get rid of? I think a secondary
| market is sure to appear.
| dist-epoch wrote:
| A single server is 20 KW. A rack is 200 KW.
|
| These are not the old CPU servers of yesterday.
| geerlingguy wrote:
| At minimum, you'd need to wire in new 240V circuits, and you
| could only run one or two of these servers before you'd need
| a service upgrade.
|
| Then you'd have to deal with noise from a literal wall of
| fans, or build a separate high capacity water cooling system
| (and still deal with dumping that heat somewhere).
| quickthrowman wrote:
| A typical residential electrical service is 48kW (240V 200A
| single-phase), with a maximum continuous load of 38.4kW. (80%
| of 48kW)
|
| Assuming 20kW per server, you could power _one_ server and
| _one_ 5-ton minisplit (5 kW, assuming a COP of 4) to cool it.
| Now you have 13kW left to power everything else in your home,
| keep in mind that the mini-split is strictly a CRCU, you'll
| need a separate condenser to cool the rest of your house.
| topaz0 wrote:
| I wonder how much of that RAM is sitting in GPUs in warehouses
| waiting for datacenters to be built or powered?
| bibimsz wrote:
| time to stop using python boys, it's zig from here on out
| QuadrupleA wrote:
| Only a matter of time before supply catches up and then likely
| overshoots (maybe combined with AI / datacenter bubble popping),
| and RAM becomes dirt cheap. Sucks for those who need it now
| though.
| seanalltogether wrote:
| I wonder if these RAM shortages are going to cause the Steam
| Machine to be dead on arrival. Valve is probably not a big enough
| player to have secured production guarantees like Sony or
| Nintendo would have. If they try to launch with a price tag over
| $750, they're probably not going sell a lot.
| pja wrote:
| Yeah, I think (sadly) this kills the Steam Machine in the short
| term if the competition is current consoles.
|
| At least until the supply contracts Sony & Microsoft have
| signed come up for renewal, at which point they're going to be
| getting the short end of the RAM stick too.
|
| In the short term the RAM shortage is going to kill homebrew PC
| building & small PC builders stone dead - prebuilts from the
| larger suppliers will be able to outcompete them on price so
| much that it simply won't make any sense to buy from anyone
| except HP, Dell etc etc. Again, this applies only until the
| supply contracts those big PC firms have signed run out, or
| possibly only until their suppliers find they can't source DDR5
| ram chips for love nor money, because the fabs are only making
| HBM chips & so they have to break the contracts themselves.
|
| It's going to get bumpy.
| fullstop wrote:
| > At least until the supply contracts Sony & Microsoft have
| signed come up for renewal, at which point they're going to
| be getting the short end of the RAM stick too.
|
| Allegedly Sony has an agreement for a number of years, but
| Microsoft does not: https://thegamepost.com/leaker-xbox-
| series-prices-increase-r...
| pja wrote:
| Eesh.
|
| The fight over RAM supply is going to upend a lot of
| product markets. Just random happenstance over whether a
| company decided to lock in supply for a couple of years is
| going to make or break individual products.
| delfinom wrote:
| If anyone that has hidden cash reserves that could buy out even
| Apple. It would probably be Valve.
|
| Lol, wacky reality if they say "hey we had spare cash so we
| bought out Micron to get DDR5 for our gaming systems"
| quesera wrote:
| To save my fellow ignorami some math, Valve is estimated to
| be valued at approximately 3.2% of Micron, or 0.2% of Apple.
| :)
| ActorNightly wrote:
| Value =/= cash reserves. Valve has ran VERY lean for quite
| some time.
| quesera wrote:
| Sure, but valuation should always exceed cash reserves.
| It's very odd when it does not. I think I recall that
| that SUNW (by then JAVA probably) was at one point valued
| below cash, prior to ORCL acquisition.
|
| If the argument is that Valve's secrecy is so good that
| they have (substantially more than) 30-500x cash stashed
| away in excess of their public valuation estimates, then
| perhaps I underestimate Valve's secrecy!
|
| Or it was a humorous exaggeration which I lacked the
| context to realize. I am quite ignorant of the games
| industry.
| quickthrowman wrote:
| Valve is worth _maybe_ 0.1% (single digit billions is what
| I'd guess) of Apple, which made $112B in _net income_ in
| 2025. That's profit, not revenue.
|
| Apple could probably buy Valve for 30 days of its net income,
| which is around $9.3B ($306M per day in profit, including
| weekends)
|
| There's zero chance that Gaben has squirreled away _Four
| Trillion Dollars_ in cash.
| notatoad wrote:
| anybody care to speculate on how long this is likely to last? is
| this a blip that will resolve itself in six months, or is this
| demand sustainable and we are talking years to build up new
| manufacturing facilities to meet demand?
| boxedemp wrote:
| Pure speculation, nobody can say say for sure, but my guess is
| 2-3 years.
| philsnow wrote:
| Red chip supply problems in your factory are usually caused by
| insufficient plastic bars, which is usually caused by oil
| production backing up because you're not consuming your heavy oil
| and/or petroleum fast enough.
|
| Crack heavy oil to light, and turn excess petroleum into solid
| fuel. As a further refinement, you can put these latter
| conversions behind pumps, and use the circuit network to only
| turn the pumps on when the tank storage of the respective reagent
| is higher than ~80%.
|
| hth, glhf
| dbbr wrote:
| You don't happen to play Foxhole [0] do you?
|
| Because if not, the logistics Collies in SOL could make good
| use of a person with your talents. :-)
|
| [0] https://store.steampowered.com/app/505460/Foxhole/
| philsnow wrote:
| Hadn't heard of it before..
|
| > Foxhole is a massively multiplayer game where
|
| nope, not my cup of tea, but thanks for the "if you like this
| you might like this" rec :)
| wpm wrote:
| Yup, and stockpiling solid fuel is not a waste because you need
| it for rocket fuel later on. Just add more chests.
| phkahler wrote:
| Called it! About a year ago (or more?) I thought nVidia was
| overpriced and if AI was coming to PCs RAM would be important and
| it might be good to invest in DRAM makers. As usual I didn't do
| anything with my insight, and here we are. Micron has more than
| doubled since summer.
| f055 wrote:
| Bullwhip effect on this will be funny. At least, we are in for
| some cheap ram in like... a dozen months or so.
| Forgeties79 wrote:
| I'm really hoping this doesn't become another GPU situation
| where every year we think it's going to get lower and it just
| stays the same or gets worse
| leeoniya wrote:
| 1. Buy up 40% of global fab capacity
|
| 2. Resell wafers at huge markup to competitors
|
| 3. Profit
| mastax wrote:
| > And those companies all realized they can make billions more
| dollars making RAM just for AI datacenter products, and neglect
| the rest of the market.
|
| > So they're shutting down their consumer memory lines, and
| devoting all production to AI.
|
| Okay this was the missing piece for me. I was wondering why AI
| demand, which should be mostly HBM, would have such an impact on
| DDR prices, which I'm quite sure are produced on separate lines.
| I'd appreciate a citation so I could read more.
| txdv wrote:
| Just like the GPUs.
|
| NVIDIA started allocating most of the wafer capacity for 50k
| GPU chips. They are a business, its a logical choice.
| loeg wrote:
| It's kind of a weird framing. Of course RAM companies are going
| to sell their limited supply to the highest bidder!
| levkk wrote:
| Not a bad time for the secondary market to be created. We keep
| buying everything new, when the old stuff works just as well.
| There is a ton of e-waste. The enthusiast market can benefit,
| while the enterprise market can just eat the cost.
|
| Also, a great incentive to start writing efficient software. Does
| Chrome really need 5GB to run a few tabs?
| tverbeure wrote:
| I updated a $330 new HP laptop (it flexes like cardboard) from
| 8GB to 32GB in May. Cost back then: $44. Today, the same kit
| costs a ridiculous $180.
|
| https://tomverbeure.github.io/2025/03/12/HP-Laptop-17-RAM-Up...
| mixermachine wrote:
| Got to say, those prices were quite cheap. I also upgraded my
| home server to 32GB RAM and paid something like 55EUR. Now we
| can just wait for some bubble to pop...
| mindslight wrote:
| > _But I 've already put off some projects I was gonna do for
| 2026, and I'm sure I'm not the only one._
|
| Let's be honest here - the projects I'm going to do in 2026, I
| bought the parts for those back in 2024. But this is definitely
| going to make me put off some projects that I might have finally
| gotten around to in 2028.
| almosthere wrote:
| I think I paid like $500 ($1300 today) in 1989ish to upgrade from
| 2MB to 5MB of ram (had to remove 1MB to add 4MB)
| internet2000 wrote:
| Hopefully Apple uses their volume as leverage to avoid getting
| affected by this for as long as possible. I can ride it out if
| they manage to.
| wkat4242 wrote:
| Also I feel that when this bubble bursts, we'll have much bigger
| problems than some expensive ram. More like the big 2007 crisis.
| layer8 wrote:
| So, buy RAM now or gold? ;)
___________________________________________________________________
(page generated 2025-12-04 23:00 UTC)