The situation is absolutely awful for anybody who needs significant memory or storage (or devices incorporating such).
Even people who think they have enough for the next few years already could end up with an unpleasant surprise if they e.g. have a stick or two of RAM go bad or a GPU fry. I have a few machines with aging sticks that I'm hoping will manage to hold out until when/if things improve.
I'm saving a ton of money by simply not upgrading. I used to do upgrades every 1-2 years just to keep up on things, pass the old box down to my son or otherwise donate it. Now I intend to keep my current setup for 7+ years. Its 2 years old next month. I may even stay with it longer. I just see no reason to upgrade with things how they are.
I considered a Mac Studio a couple months ago. Price was up but might have pulled the trigger if not for the delivery date that was almost 4 months out. Sale lost. I suspect a lot of people are like me right now and we'll start seeing significant drops in consumer spending on electronics the next 2-4 years minimum.
There also seems to have been a major shift in strategy from AMD / Intel / Apple / NVidia due to this, as I was used to CPU / GPU / chipset upgrades coming out more regularly with double-digit percentage improvements each year, but that seems to be basically over for the consumer space now that everyone is just trying to build as much AI hardware as fast as possible for enterprise markets.
I also used to rebuild my desktop every 1-2 years and upgrade laptops every 2-3 years, but I'm currently a bit past that and have no plans to refresh anything in the near future. I'll probably wait another CPU and GPU generation or two to even consider it, especially with prices being this ridiculous.
I'm fully anticipating AI hardware demand to dry up once Anthropic and OpenAI start tanking financially in a few years after missing earnings and investors and banks wanting their loans paid back combined with more folks realizing that LLMs cannot achieve AGI. After that, I would hope that the demand that's driving this pricing spike will fall along with it, but I also fully expect hardware companies to be as greedy as possible, so maybe a 32TB HDD will just cost $1000 going forward and the era of "a decent gaming PC made up of the tier of components one step below TOTL costs ~$2000" we've been in since the 1990s is fully over, and it's just going to be more like $5k going forward.
> There also seems to have been a major shift in strategy from AMD / Intel / Apple / NVidia due to this, as I was used to CPU / GPU / chipset upgrades coming out more regularly with double-digit percentage improvements each year, but that seems to be basically over for the consumer space now that everyone is just trying to build as much AI hardware as fast as possible for enterprise markets.
I believe all of them realized nobody really needs more than 16c32t on a desktop, and architectural changes nowadays can only go so far to bump performance. The biggest changes across generations in the last ~5 years were the introduction of specialized instructions for wider vectors (AVX512/AVX2, and even then you need specialized workloads to use them), bigger L3 caches and/or efficiency cores to lower power draw. The double-digits gains recently were in memory bandwidth, frequency and locality, because you can make a 5GHz CPU and make it go 10% faster than the previous generation, but you also need to supply the extra data for the CPU to make useful work so that it's not data-starved.
> I also used to rebuild my desktop every 1-2 years and upgrade laptops every 2-3 years, but I'm currently a bit past that and have no plans to refresh anything in the near future. I'll probably wait another CPU and GPU generation or two to even consider it, especially with prices being this ridiculous.
I built my desktop in 2019 (right before the COVID madness, luckily) and I was also used to 1-3 year cycles. I upgraded the CPU and GPU since then, but the performance is still that of a workstation, even though perhaps a little bit behind in pure single-core benchmarks. I plan to keep this until either CPU, RAM or MB fall out (hopefully another 5y at least), and then I'll make a huge leap, instead of making 4x smaller ones which deliver little uplift anyway.
I'm running a 12 year old i5 3.2ghz CPU with 16 GB of memory, an SSD and an AMD RX 480 (8 GB) GPU. Currently using a 100% scaled 4k monitor and a 2nd 1440p monitor. I keep the machine on 24 / 7. It's great for software development along with video + image editing and casual gaming. Everything feels fast and snappy.
I will use this until the hardware stops. There's a 0% chance I'd buy hardware at the current prices.
I have a baseline iMac Pro and custom built tower with similar specs (3.2Ghz Xeon W-2140B, 32GB RAM, Vega Pro 56 and i7-6700K, 32GB, 5700XT respectively) and they're both perfectly responsive. For many day to day things the difference between them and newer machines is negligible. For other things, my 5950X tower and M5 Max MBP run circles around them.
Both the 5950X tower and M5 Max MBP are deep into diminishing returns territory regardless, however. Aside from some extreme niche workloads like LLMs I can't imagine needing anything more powerful. They both have a ridiculous amount of headroom for the overwhelming majority of anything most people might want a computer for.
Yep, I have an M4 MBP company issued laptop for work and in my day to day there's not a enough of a performance win to feel like I'm missing out for most tasks.
Sure, the Slack web app might initially load half a second faster since there's a lot of CPU bound JS executing but generally speaking I never feel like "wow this sucks" on my main machine. It's kind of the opposite, for a number of things my desktop machine feels faster than the MBP for overall "using the computer", with development heavy tasks (running Docker based web apps, browser tabs, various tools open, streaming music, etc.). It was like that when I ran Windows 10 Pro with WSL 2 and now native Linux with Arch + niri.
Even CPU bound tasks on my 12 year old box that I do a lot, such as editing a blog post on a static site with Hugo takes ~100ms to get live reloading with over 1,000 posts and drafts. I'm sure if I were constantly doing huge CPU bound tasks I'd have issues but I don't have that use case.
After recording 500+ videos, encoding isn't a problem too. If I record a 30 min video, it takes ~20 minutes to render but it's so easy to fit that into other areas of my life, like I'll record + edit a video and then render it while I walk.
I have a 2021 M1 pro and see no reason to upgrade it... ever?
I do web & general java work on it and it's plenty fast even using a memory hog like IntelliJ (which is typically disk bound anyway). Claude tells me that a new macbook would feel about 1.8x faster in my CPU-bound cases, which is not common.
If memory prices returned to normal and if local inference gets over the hump, I could see upgrading for local AI agents, but I'm fine waiting for a good long while on that.
I say all this as someone who, five years ago, was a perpetual upgrader: immediately bought new product and got excited for next product. Wonder if the future will see the end of that culture in tech and, looking back, it was a little silly and wasteful anyway.
I too have an M1 Pro and it does everything I need to do. I bought it in 2023, so it's 'only' 3.5 years old now.
Maybe this is the tipping point where hardware companies can't rely on the upgrade cycle for sales any more. I've always upgraded at least every three years, but with the combination of high RAM prices and good enough performance from an older machine, there is no need to.
There isn't much to upgrade to if all you want is a desktop for gaming, a non-workstation work box or just a decent laptop. Much like smartphones, there just hasn't been much innovation of late. A midrange gaming desktop from mid 2023, something like rtx 4070 and ryzen 7800x3d, can still play pretty much any video game on max settings and high framerate at 1440p, and has sufficient performance for most workstation-esque tasks as long as you have enough ram. Similarly, the only thing my spare macbook m2 lacks compared to m5 is hardware av1 decoding - the performance is worse, but I don't really notice it in normal usage, and macOS 27 runs perfectly well on it.
Local AI will crush any machine, of course, and if you need a 512gb ram box you'll probably need to sell both kidneys, but you're well past the average consumer territory with those.
So, fingers crossed nothing breaks down and that some of the funny money goes into R&D which will eventually trickle down to consumer devices in a few years (16TB gum stick SSDs at sane prices pretty please) instead of endless datacenter GPUs and SSDs that will be tossed into a shredder in x years.
Most devices never test their memory and don’t have ecc, so you generally never know. Random instability can be fixed by restarting and your program not being in an area with bad bits anymore.
Anecdotally, RAM failures are more common than they used to be in previous decades, I assume due to increased capacities and smaller process nodes. Looser manufacturing standards may also play a part.
The component category I've never seen fail is CPU.
I've worked in and around datacenters for a few decades now and I've seen pretty much everything fail, but power supplies in particular do so at a higher rate than I expected. I've had one literally start smoking in my home office recently too. Luckily, it didn't damage any components, but it was pretty dramatic and abrupt -- just powered on my desktop and got nothing but a fog machine coming out of the rear fan until I yanked the power cord a few seconds later.
My evga 1000w has been going strong for 11 years now. Cheaper power supplies fail often but if you buy quality and oversized they’ll outlast all your other components as they don’t become outdated.
It’s the caps that fail. Electrolytic caps have the highest failure rate of any passive component. My team tries to avoid them in our designs as much as possible. But you’re right, caps in a good design can easily last 20 years, making a power supply the longest lasting component of most builds.
My go to source for reviews was always jonnyguru.com but he closed up shop. Not sure who’s actually doing in-depth performance and component reviews these days.
I recently had part of a 32 gig DIMM go bad on a 128 gig system. One of the larger VMs was spontaneously rebooting for no apparent reason. I was able to find the addresses with MemTest86 and map out 512 megs of the affected DIMM with the Linux kernel memmap option.
I've seen some fail. Might have had to do with the fact that the computers were literally flooded (river came up faster than the sandbaggers could keep up) while running.
I've seen computers brought back to life after flooding by lying in rice for a week, but they weren't turned on when the water hit (and not all survived).
I've had a few RAM sticks fail, but the ECC usually caught it first. Also had one case that somehow the stick slightly unseated itself; dunno how but it was out of service for months until someone opened it up and reseated the stick.
...granted though I've also seen a backpane on a server fail. I didn't think that was even possible until that point.
I would absolutely love if some startup came up right now to manufacture and sell RAM at normal prices to everyone else and put Crucial et al out of business for price-gouging everyone just because they can.
DRAM fabs are multi-billion dollar investments which take years to come online. That's the whole reason we're in this situation, there is no quick fix available.
China is making central decisions for its RAM makers (and who can complain now that the US isn't a legitimate free trade defender). I think their strategy could go in any direction to either get a small portion of still gouging or a large portion of a market with somewhat stale specs.
I noticed that resellers on ebay of used server equipment that I keep an eye on in Germany pulled all their DDR4 from surplus and used server racks and now list it separately for sale, at jacked up prices. Where you used to be able to see a used Gigabyte server listed with "128GB" "256GB" etc listed in the title, they now all say "0GB". They took extra effort, to pull used ram from older servers, and list it separate, just out of the greed of potentially capitalizing on this market.
The problem is... now for anyone looking to pickup a slightly older used/surplus server rack - you have the go and buy the (now marked up) RAM separate.
So... did they list the "0GB" racks with removed ram cheaper you ask?
Nope. Many of them seem to be basically the same price as they were or even higher than they were, except now you're paying for RAM separate.
I bought a surplus Gigabyte server about 2 years ago off ebay for a home project, and I just went and the same seller still has some for sale, even older now than they were when I grabbed one, but they are 2x the price, and they don't come with any RAM, whereas I got one already populated as they were from the factory. Once you add that back in literally looking at 150-200% markup in the last 2 years on what was just surplus computer equipment that is 5+ years old.
> So... did they list the "0GB" racks with removed ram cheaper you ask?
Could this be an indication that there's a supply shortage/price increase of the remaining components in the rack? Or do you interpret it as people just trying their luck out of sheer greed?
While I share your optimism, I still don't fully understand why a crash will necessarily lead to an oversupply and lower prices for regular consumers. The components / RAM you are referencing are for enterprise use / data centers which in my limited understanding is not a 1:1 swap for the kind of components that your smartphone / car / fridge / gaming rig use. I would love for someone more informed than me to spell out the rationale here.
It seems the chokehold started from the very beginning of the supply chain (wafers), so if a crash were to occur I don't see how we wouldn't be doubly screwed. Lots of worthless enterprise gear being shopped on eBay + a scarcity of consumer grade components leading to sustained high prices.
My understanding is that it's fundamentally a fab capacity limit. At a given node size/technology (think 7nm, etc) they can only output so many wafers. If AI data centers are generating enormous demand for memory, then that's what those wafers are going to be used to for, unless they can be outcompeted, hence higher prices.
I'm not actually sure if enterprise designated memory is equivalent or not to consumer designated. Could be they are the same just packaged differently and/or with better binning. Either way, it's the same supply/demand dynamic at work.
Datacentre NVMe disk is still usable in a standard desktop. Sever memory less so, but essentially if there were no 'new' server market whatsoever then you could redirect 100% of memory production to consumer chips without enormous upheaval.
By the way I personally think this scenario is very unlikely to happen.
I’m even more confused by the economics of the frontier model businesses. Even if they got preferential prices, they are still paying an inordinate amount for their infrastructure.
If it wasn’t economically feasible without VC/Nvidia money 3 years ago, how is it possibly economically feasible now at 10x prices for things like memory?
And watch … $5 says they have nothing to do with the ram because even if there was enough power there’s no use case for this much spend
If I had to guess they’re just buying up ram to keep others from having it same way meta, Google, etc hired up engineers to keep others from having them
I’ve been wondering this if you buy up all the memory you can ensure the competition cannot compete with huge investment raising the bar for competition. Of course this all sounds all conspiracy theory. Though without a place to power this or put it makes me wonder this.
It's pulling the ladder up behind them. OpenAI bought up more co tracts for wafers than they could even use. But now that means everyone else has to fight over what's left. This makes it too expensive to compete with OpenAI. No upstart can really afford to build infrastructure.
I bought an RX 6600 for $450 during the GPU shortage a few years ago. I know the AI boom is different than the crypto/COVID boom, but still. Don't be me.
Even with AI pressure I'm willing to bet there's no need for prices to be this high. There has to be some price gouging/fixing going on. Same thing for nand and GPUs. Or else they're doing it to try to pressure people not to use local compute instead of the "cloud"
Been long past that time, probably 15-20 years past. It's not even the languages its the 500 different layers of abstraction and abominations like Electron.
"The best good ever is to maximise shareholder profits" is a tired and cliched caricature that some people unfortunately take seriously. But this is too on the nose.
The situation is absolutely awful for anybody who needs significant memory or storage (or devices incorporating such).
Even people who think they have enough for the next few years already could end up with an unpleasant surprise if they e.g. have a stick or two of RAM go bad or a GPU fry. I have a few machines with aging sticks that I'm hoping will manage to hold out until when/if things improve.
I'm saving a ton of money by simply not upgrading. I used to do upgrades every 1-2 years just to keep up on things, pass the old box down to my son or otherwise donate it. Now I intend to keep my current setup for 7+ years. Its 2 years old next month. I may even stay with it longer. I just see no reason to upgrade with things how they are.
I considered a Mac Studio a couple months ago. Price was up but might have pulled the trigger if not for the delivery date that was almost 4 months out. Sale lost. I suspect a lot of people are like me right now and we'll start seeing significant drops in consumer spending on electronics the next 2-4 years minimum.
There also seems to have been a major shift in strategy from AMD / Intel / Apple / NVidia due to this, as I was used to CPU / GPU / chipset upgrades coming out more regularly with double-digit percentage improvements each year, but that seems to be basically over for the consumer space now that everyone is just trying to build as much AI hardware as fast as possible for enterprise markets.
I also used to rebuild my desktop every 1-2 years and upgrade laptops every 2-3 years, but I'm currently a bit past that and have no plans to refresh anything in the near future. I'll probably wait another CPU and GPU generation or two to even consider it, especially with prices being this ridiculous.
I'm fully anticipating AI hardware demand to dry up once Anthropic and OpenAI start tanking financially in a few years after missing earnings and investors and banks wanting their loans paid back combined with more folks realizing that LLMs cannot achieve AGI. After that, I would hope that the demand that's driving this pricing spike will fall along with it, but I also fully expect hardware companies to be as greedy as possible, so maybe a 32TB HDD will just cost $1000 going forward and the era of "a decent gaming PC made up of the tier of components one step below TOTL costs ~$2000" we've been in since the 1990s is fully over, and it's just going to be more like $5k going forward.
> There also seems to have been a major shift in strategy from AMD / Intel / Apple / NVidia due to this, as I was used to CPU / GPU / chipset upgrades coming out more regularly with double-digit percentage improvements each year, but that seems to be basically over for the consumer space now that everyone is just trying to build as much AI hardware as fast as possible for enterprise markets.
I believe all of them realized nobody really needs more than 16c32t on a desktop, and architectural changes nowadays can only go so far to bump performance. The biggest changes across generations in the last ~5 years were the introduction of specialized instructions for wider vectors (AVX512/AVX2, and even then you need specialized workloads to use them), bigger L3 caches and/or efficiency cores to lower power draw. The double-digits gains recently were in memory bandwidth, frequency and locality, because you can make a 5GHz CPU and make it go 10% faster than the previous generation, but you also need to supply the extra data for the CPU to make useful work so that it's not data-starved.
> I also used to rebuild my desktop every 1-2 years and upgrade laptops every 2-3 years, but I'm currently a bit past that and have no plans to refresh anything in the near future. I'll probably wait another CPU and GPU generation or two to even consider it, especially with prices being this ridiculous.
I built my desktop in 2019 (right before the COVID madness, luckily) and I was also used to 1-3 year cycles. I upgraded the CPU and GPU since then, but the performance is still that of a workstation, even though perhaps a little bit behind in pure single-core benchmarks. I plan to keep this until either CPU, RAM or MB fall out (hopefully another 5y at least), and then I'll make a huge leap, instead of making 4x smaller ones which deliver little uplift anyway.
Are you willing to short anthropic and openai on your beliefs? Would love to be on the other side of this trade.
"Sometime in the next few years" is not anywhere near narrow enough of a window to start putting shorts on stock prices.
I'm running a 12 year old i5 3.2ghz CPU with 16 GB of memory, an SSD and an AMD RX 480 (8 GB) GPU. Currently using a 100% scaled 4k monitor and a 2nd 1440p monitor. I keep the machine on 24 / 7. It's great for software development along with video + image editing and casual gaming. Everything feels fast and snappy.
I will use this until the hardware stops. There's a 0% chance I'd buy hardware at the current prices.
It really depends on what the task at hand is.
I have a baseline iMac Pro and custom built tower with similar specs (3.2Ghz Xeon W-2140B, 32GB RAM, Vega Pro 56 and i7-6700K, 32GB, 5700XT respectively) and they're both perfectly responsive. For many day to day things the difference between them and newer machines is negligible. For other things, my 5950X tower and M5 Max MBP run circles around them.
Both the 5950X tower and M5 Max MBP are deep into diminishing returns territory regardless, however. Aside from some extreme niche workloads like LLMs I can't imagine needing anything more powerful. They both have a ridiculous amount of headroom for the overwhelming majority of anything most people might want a computer for.
Yep, I have an M4 MBP company issued laptop for work and in my day to day there's not a enough of a performance win to feel like I'm missing out for most tasks.
Sure, the Slack web app might initially load half a second faster since there's a lot of CPU bound JS executing but generally speaking I never feel like "wow this sucks" on my main machine. It's kind of the opposite, for a number of things my desktop machine feels faster than the MBP for overall "using the computer", with development heavy tasks (running Docker based web apps, browser tabs, various tools open, streaming music, etc.). It was like that when I ran Windows 10 Pro with WSL 2 and now native Linux with Arch + niri.
Even CPU bound tasks on my 12 year old box that I do a lot, such as editing a blog post on a static site with Hugo takes ~100ms to get live reloading with over 1,000 posts and drafts. I'm sure if I were constantly doing huge CPU bound tasks I'd have issues but I don't have that use case.
After recording 500+ videos, encoding isn't a problem too. If I record a 30 min video, it takes ~20 minutes to render but it's so easy to fit that into other areas of my life, like I'll record + edit a video and then render it while I walk.
I have a 2021 M1 pro and see no reason to upgrade it... ever?
I do web & general java work on it and it's plenty fast even using a memory hog like IntelliJ (which is typically disk bound anyway). Claude tells me that a new macbook would feel about 1.8x faster in my CPU-bound cases, which is not common.
If memory prices returned to normal and if local inference gets over the hump, I could see upgrading for local AI agents, but I'm fine waiting for a good long while on that.
I say all this as someone who, five years ago, was a perpetual upgrader: immediately bought new product and got excited for next product. Wonder if the future will see the end of that culture in tech and, looking back, it was a little silly and wasteful anyway.
I'm also using an iPhone mini 12 and it's fine.
I too have an M1 Pro and it does everything I need to do. I bought it in 2023, so it's 'only' 3.5 years old now.
Maybe this is the tipping point where hardware companies can't rely on the upgrade cycle for sales any more. I've always upgraded at least every three years, but with the combination of high RAM prices and good enough performance from an older machine, there is no need to.
There isn't much to upgrade to if all you want is a desktop for gaming, a non-workstation work box or just a decent laptop. Much like smartphones, there just hasn't been much innovation of late. A midrange gaming desktop from mid 2023, something like rtx 4070 and ryzen 7800x3d, can still play pretty much any video game on max settings and high framerate at 1440p, and has sufficient performance for most workstation-esque tasks as long as you have enough ram. Similarly, the only thing my spare macbook m2 lacks compared to m5 is hardware av1 decoding - the performance is worse, but I don't really notice it in normal usage, and macOS 27 runs perfectly well on it.
Local AI will crush any machine, of course, and if you need a 512gb ram box you'll probably need to sell both kidneys, but you're well past the average consumer territory with those.
So, fingers crossed nothing breaks down and that some of the funny money goes into R&D which will eventually trickle down to consumer devices in a few years (16TB gum stick SSDs at sane prices pretty please) instead of endless datacenter GPUs and SSDs that will be tossed into a shredder in x years.
I've been using computers for more than 40 years now and the one component I have never seen fail is RAM.
All the other bits in a computer I've seen fail.
Most devices never test their memory and don’t have ecc, so you generally never know. Random instability can be fixed by restarting and your program not being in an area with bad bits anymore.
Anecdotally, RAM failures are more common than they used to be in previous decades, I assume due to increased capacities and smaller process nodes. Looser manufacturing standards may also play a part.
The component category I've never seen fail is CPU.
I've worked in and around datacenters for a few decades now and I've seen pretty much everything fail, but power supplies in particular do so at a higher rate than I expected. I've had one literally start smoking in my home office recently too. Luckily, it didn't damage any components, but it was pretty dramatic and abrupt -- just powered on my desktop and got nothing but a fog machine coming out of the rear fan until I yanked the power cord a few seconds later.
My evga 1000w has been going strong for 11 years now. Cheaper power supplies fail often but if you buy quality and oversized they’ll outlast all your other components as they don’t become outdated.
It’s the caps that fail. Electrolytic caps have the highest failure rate of any passive component. My team tries to avoid them in our designs as much as possible. But you’re right, caps in a good design can easily last 20 years, making a power supply the longest lasting component of most builds.
My go to source for reviews was always jonnyguru.com but he closed up shop. Not sure who’s actually doing in-depth performance and component reviews these days.
You didn't hold on to your Pentium 4 long enough. :-)
My main machines were PowerPC G3s/G4s/G5s back then with the P4 boxes I had taking secondary/tertiary roles, so that lines up.
Anecdatum, but mine still works :D
Well, you’re lucky. I had to change laptop memory sticks a few times. Though not since they are soldered directly to the motherboard
I recently had part of a 32 gig DIMM go bad on a 128 gig system. One of the larger VMs was spontaneously rebooting for no apparent reason. I was able to find the addresses with MemTest86 and map out 512 megs of the affected DIMM with the Linux kernel memmap option.
Same, it's either bad to begin with or it's basically immortal.
I've seen some fail. Might have had to do with the fact that the computers were literally flooded (river came up faster than the sandbaggers could keep up) while running.
I've seen computers brought back to life after flooding by lying in rice for a week, but they weren't turned on when the water hit (and not all survived).
This isn’t true anymore beginning with DDR4 speeds.
I've had a few RAM sticks fail, but the ECC usually caught it first. Also had one case that somehow the stick slightly unseated itself; dunno how but it was out of service for months until someone opened it up and reseated the stick.
...granted though I've also seen a backpane on a server fail. I didn't think that was even possible until that point.
I bought a prebuilt with a 5090 a little over than one month ago, whole system was on sales.
The cost of the prebuilt is now lower than today 5090 price.
before the memory cartel, I was automatically putting in 64 or 128GB for my work flow as the prices just seems absurd not to.
I would absolutely love if some startup came up right now to manufacture and sell RAM at normal prices to everyone else and put Crucial et al out of business for price-gouging everyone just because they can.
DRAM fabs are multi-billion dollar investments which take years to come online. That's the whole reason we're in this situation, there is no quick fix available.
China is making central decisions for its RAM makers (and who can complain now that the US isn't a legitimate free trade defender). I think their strategy could go in any direction to either get a small portion of still gouging or a large portion of a market with somewhat stale specs.
Any new fabs on the horizon or do the powers that be expect the music to stop sooner than that?
I believe they've been building them like crazy since covid, but it takes a looong time for them to come online (like 5-10 years?)
I noticed that resellers on ebay of used server equipment that I keep an eye on in Germany pulled all their DDR4 from surplus and used server racks and now list it separately for sale, at jacked up prices. Where you used to be able to see a used Gigabyte server listed with "128GB" "256GB" etc listed in the title, they now all say "0GB". They took extra effort, to pull used ram from older servers, and list it separate, just out of the greed of potentially capitalizing on this market.
The problem is... now for anyone looking to pickup a slightly older used/surplus server rack - you have the go and buy the (now marked up) RAM separate.
So... did they list the "0GB" racks with removed ram cheaper you ask? Nope. Many of them seem to be basically the same price as they were or even higher than they were, except now you're paying for RAM separate.
I bought a surplus Gigabyte server about 2 years ago off ebay for a home project, and I just went and the same seller still has some for sale, even older now than they were when I grabbed one, but they are 2x the price, and they don't come with any RAM, whereas I got one already populated as they were from the factory. Once you add that back in literally looking at 150-200% markup in the last 2 years on what was just surplus computer equipment that is 5+ years old.
> So... did they list the "0GB" racks with removed ram cheaper you ask?
Could this be an indication that there's a supply shortage/price increase of the remaining components in the rack? Or do you interpret it as people just trying their luck out of sheer greed?
I still hope all this will come crashing down, there will be an oversupply of components and RAM will be cheaper than paper.
While I share your optimism, I still don't fully understand why a crash will necessarily lead to an oversupply and lower prices for regular consumers. The components / RAM you are referencing are for enterprise use / data centers which in my limited understanding is not a 1:1 swap for the kind of components that your smartphone / car / fridge / gaming rig use. I would love for someone more informed than me to spell out the rationale here.
It seems the chokehold started from the very beginning of the supply chain (wafers), so if a crash were to occur I don't see how we wouldn't be doubly screwed. Lots of worthless enterprise gear being shopped on eBay + a scarcity of consumer grade components leading to sustained high prices.
My understanding is that it's fundamentally a fab capacity limit. At a given node size/technology (think 7nm, etc) they can only output so many wafers. If AI data centers are generating enormous demand for memory, then that's what those wafers are going to be used to for, unless they can be outcompeted, hence higher prices.
I'm not actually sure if enterprise designated memory is equivalent or not to consumer designated. Could be they are the same just packaged differently and/or with better binning. Either way, it's the same supply/demand dynamic at work.
Typically (not always!) server memory is 'registered' so won't work with consumer CPUs which run on un-registered.
Datacentre NVMe disk is still usable in a standard desktop. Sever memory less so, but essentially if there were no 'new' server market whatsoever then you could redirect 100% of memory production to consumer chips without enormous upheaval.
By the way I personally think this scenario is very unlikely to happen.
> If there were no 'new' server market whatsoever then you could redirect 100% of memory production to consumer chips without enormous upheaval.
> By the way I personally think this scenario is very unlikely to happen.
Can you explain why? I imagine pivoting to a much less lucrative consumer market when the time comes would lead to financial ruin?
I’m even more confused by the economics of the frontier model businesses. Even if they got preferential prices, they are still paying an inordinate amount for their infrastructure.
If it wasn’t economically feasible without VC/Nvidia money 3 years ago, how is it possibly economically feasible now at 10x prices for things like memory?
OpenAI (Stargate) kicked off the price increases a year ago by locking up 40% of the global supply of RAM output.
https://www.tomshardware.com/pc-components/dram/openais-star...
And watch … $5 says they have nothing to do with the ram because even if there was enough power there’s no use case for this much spend
If I had to guess they’re just buying up ram to keep others from having it same way meta, Google, etc hired up engineers to keep others from having them
>even if there was enough power there’s no use case for this much spend
According to them, demand for tokens already exceeds supply. Coding agents are very popular but burn through millions of tokens per hour.
The reasoning models that have powered breakthroughs in AI for math, coding, etc require substantially more compute than base LLMs.
I’ve been wondering this if you buy up all the memory you can ensure the competition cannot compete with huge investment raising the bar for competition. Of course this all sounds all conspiracy theory. Though without a place to power this or put it makes me wonder this.
next they will become successful business by reselling RAM
There is a reason NVIDIA and the hyperscalers are injecting so much money into them
It's pulling the ladder up behind them. OpenAI bought up more co tracts for wafers than they could even use. But now that means everyone else has to fight over what's left. This makes it too expensive to compete with OpenAI. No upstart can really afford to build infrastructure.
I bought an RX 6600 for $450 during the GPU shortage a few years ago. I know the AI boom is different than the crypto/COVID boom, but still. Don't be me.
I got a 3060 ti for like $900 at that time <melting-face-emoji>
Even with AI pressure I'm willing to bet there's no need for prices to be this high. There has to be some price gouging/fixing going on. Same thing for nand and GPUs. Or else they're doing it to try to pressure people not to use local compute instead of the "cloud"
Storage has more than doubled as well. The same SSD (4TB M.2) I bought in September 2025 for £229 is now £500.
I also had another order where I paid £30 for a 1TB SSD (sata though) and it is now more than £100...
Not just SSDs, HDDs are also way more expensive. End of 2024 I bought 4 14TB drives for ~$250 CAD each, $400+ for even a refurb drive now.
The possibility that these fail does concern me.
I hate that I am no longer certain I don’t need to buy a new machine just in case.
The fake economics of the AI sector are damaging the real economy. This is madness.
I guess it's high time to stop using memory inefficient programming languages for any serious work.
Been long past that time, probably 15-20 years past. It's not even the languages its the 500 different layers of abstraction and abominations like Electron.
Good. The more expensive memory is, the more everyone will be pushed into using cloud AI from hyperscalars.
They should each buy out a RAM manufacture, and make this the new normal.
It really is the best thing for AI adoption and the shareholders.
No, this is not good, in any way, except for AI shareholders (of which I assume you are one, else not rational to support this position).
Paperclipping here, greedily optimizing for shareholder value above all else. Please stop.
Some people really so need the /s huh
To be fair, the AI boosters are saying the exact same thing but mean it seriously. There is no way to distinguish without a tone indicator
> There is no way to distinguish without a tone indicator
And yet many of us aren't having any trouble distinguishing.
sadly, Poe's law scales exponentially on HN
what the heck
did you drop a '/s'?
> and the shareholders.
Your comment is an illustration of exactly what is wrong with this website. Given his preceding statements, I sincerely do not believe it is sarcasm.
I'm quite sure he's being sarcastic.
"The best good ever is to maximise shareholder profits" is a tired and cliched caricature that some people unfortunately take seriously. But this is too on the nose.
The fact that, on HN, you can't be sure of that, tells you all you need to know.
I have the innate right to take what is said literally. Moreover, his preceding statements attempt to legitimize his concluding statement.