Just went to market recently for some new kit and I almost fell off my chair when the quotes came in. There appears to have been around a 50% increase in what I buy since August 2025 and I’m told it’s only going to get worse. Seems there’s a chip shortage due to the AI explosion. Was curious to see if anyone else has seen this and what we can expect over the coming year?
Welcome to the AI bubble!
It’s not just chips (esp. RAM): HDDs are also much more expensive than last year.
Yes it was largely HDD that my quote was based on. I was told if it was RAM or SSD heavy it would have been 100% or 200% more.
DDR5 kits that were around 250 bucks half a year ago are not at over 1000… never been a worse time if your system dies and you have to buy new hardware…
Yesterday my main PC suddendly die under usage when i was working…
Freeze and not posting… i started sweat coldy thinking about a ram stick fails ![]()
Instead, was the NVME that die and cause the pc to stuck… ok it was a chinese NVME paied 28€ just 23 months ago, but was pretty performant despite the price and i have used it quite intensive in this period (and being in a mini pc, it runs most of the time hot despite the cooler)… so pretty decent purchase, why not buy another one? Because the price now is 80€!!!
Tricky downgrading to a smaller disk laying around is still in progress ![]()
These increases on new parts are indecent, not the first time happens, but what is worst for me is that even in the second-hand market are full of jackals that are selling 10 years old ram ddr4 kit at 3 times the price they paid to buy it new… and NVME are hard to find
I had to skip RAM upgrades because of this stupid AI nonsense.
I’m convinced that this bubble will burst, the question is when and how badly. Tesla is a car company first and foremost, it has declining sales and yet the market deigns it worthy of a 250+ PE ratio the last time I looked, while Toyota is poking along at 10.
I feel extremely fortunate that when disk prices crashed as part of the post-Chia-bubble era I was able to pick up a replacement set for my NAS at less than $8/TB. Present prices are about 2x that.
But even as I was placing my orders, I was wondering if I wasn’t potentially overpaying, etc. Bottom line, my pool isn’t even 50% full so there is no good reason to dramatically upgrade my HDDs anytime soon and as much as I would love to tinker with a AM4 or AM5 platform to see “how low can you go” re: power consumption, all that can definitely wait given current RAM prices.
Me too. When I do a Google search for something a little more specific and I already know the basics (think something like an elegant way to script something under Linux and you already know the tool you’re going to use and its basics) the AI result is always something that are pretty bad or just incorrect.
On the topic itself from a German perspective: Yes, things are really bad right now. I bought a Ugreen DXP6800 through their Kickstarter in Summer 2024 as my new hardware for Truenas. HDDs went over from the old system but I decided to add 64 GB RAM and two SSDs (mirrored) for Apps and VMs.
The SSDs are Kioxia Exeria Plus G3 with 1 TB. Their price back then was 63 Euros per piece as a deal of the day. Currently their price is at least a 120 Euros per piece.
The RAM is even worse. I bought a kit by Crucial with 2*32 GB for just a little under 200 Euros. The exactly same kit currently costs 520 Euros at the cheapest reputable dealer online.
I have found AI to often confuddle results from multiple sources and apply them incorrectly. For example, it’s not trivial to set up a Multi-WAN Mikrotik setup, principally because enough is changing under the hood / inside the menus that following a recipe usually ends in disaster around step 5 because your version of RouterOS is more current than the one that the AI is giving you advice for.
That said, I have found AI to be quite useful for weird programming challenges, such as reading in large, dynamically-defined arrays, which I found an impossible challenge with regular google but quite manageable via the Co-Pilot helper. I don’t do much VBA and what little C skills I have are only good enough to read and understand the code once Co-Pilot has written it. But my “programs” are less than 1000 lines long, etc. i.e. are simple enough where a decent LLM can do most of the work.
Also concur that RAM prices are totally bonkers at this point, i.e. you spend more on the RAM than the motherboard and CPU combined. That’s nuts. Reminds me of the Raspberry Pi shortage during the pandemic.
Wiskydrinker… kein Wommi oder Bommerlunder? (Trio Live '82) ![]()
It is likely that as AI datacenter growth continues, prices will climb to a peak. As the growth peaks, and later starts to wain, that is when prices will begin to drop. Until then, may all of our storage have plenty of free space and long lived drives.
There’s no peak in sight. Rather, DC growth is below projections/expectations because there’s not enough supply (GPUs, servers, power…) and/or workforce to meet plans.
Bigger DCs are used to run ever bigger and ever more expensive models. Revenues grow at a much slower pace than costs. So all of this is paid by investors… and circular deals which we are told are not “vendor financing” but still furiously look like so.
The drop will come when bankers/investors/accountants will no longer be able to escape the realisation that future profits will NOT meet expectations. Until then, I suggest to watch JC Chandor’s excellent movie Margin Call.
And while US capitalists bet on proprietary AI to lock the future profits, China takes the lead on open AI. Long game… ![]()
For a few months it’s been “anything with (D)RAM/NAND" and it’s spreading to HDDs too.
I was completing a new server build in November and had just gotten in under the wire buying some used Transcend RAM from an IT asset disposal supplier. Got away with 64GB of working ECC DDR4 for less than $300 USD. Last week I was looking at a comparable SKU from a different manufacturer, new, for “only” $800 USD before taxes and shipping.
A friend of mine who lives in my home town was telling me how the local community has been organizing a campaign to pressure local politicians into passing a legal moratorium on issuing permits to construct new data centers or expand existing ones. The noise pollution from generators run at peak grid hours and the instability of the grid as more data centers have opened has been quite poorly received (understandably).
The issue I see here is the disconnect between what growth is being promised, the investments needed to get there, and the revenue / profit in general.
Basically, the faster they grow, the more cash they hemorrhage. Presently, the growth curve associated with investment is still steeper than the revenue curve. That doesn’t math unless investments are miniscule.
Industry CAPEX is allegedly breaking $500BN a year while the incremental revenue is maybe 1/20th of that - on a quickly depreciating asset. NVIDIA is promising vastly faster chips at lower power consumption but that would necessitate ripping out extant infrastructure and writing it off.
Never mind the performance gap, where models are not improving quickly enough, except in still niche applications. At least the dark fiber laid in 2000 had an eventual use.
i was trolling ebay for a bit looking for 4x32 ddr4 kits to max out my board, i bid on a few up to like 250 bucks or so but they kept selling for up to 300. i said meh, theyll always be around, ill just wait until i get one for the price i want. and then…….. ![]()
im just happy i already have 64gb and it was a want rather than a need to try to upgrade.
Until the dust settles from the fascination and shiny new aspect of AI, it will be interesting to watch.
Until prices drop, watching is all I will be doing.
Currently binge watching the Vikings TV series on Plex. ![]()
I think I’m probably as dismayed as everyone else. I have a 10 year old NAS with 4TB drives that is nearing capacity and wanted to see about replacing it. I was hoping by now I could get 8TB or 12TB drives for almost the same price, but nope.
Pre 2010 whenever a new capacity of HDD was introduced it used to knock the lowest capacity off the market and everything would shift down the pricing table. Theoretically by now we should have been seeing 4TB drives for $60 at the entry level, 8TB for $100 and $16 TB for under $200, but it all seems to have stalled and the 4TB drives I bought in 2015 for $135 are virtually the same price today.
Yet, new capacities are still getting added to the top. We just have more than a dozen capacities on the market now instead of like half a dozen.
It’s the crisis’ along with bubbles….
First the floods in Thailand, then the Chia bubble, now the AI bubble. Add to that a 3-way oligopoly with high cost of entry and an uncertain future (remember, we’re supposed to go all flash, like soon, right?) and the case to enter the business is difficult at best.
I don’t see new competitors springing up unless China decides that this is yet another industry they need to have a national champion in.
All flash would be awesome. If they could get the cost of those 60 TB SSDs down to around $1500 instead of $6,000 they might stand a chance of replacing everything.
Right now looks like we are heading more toward $60,000 than toward $1500 per unit ![]()
Also, even with all flash on a working system, it would be nice to have a backup, and backups will still be rotational, I suppose.
The flash / RAM business is even more expensive to enter and stay alive in than the HDD one. Super high fixed costs, massive R&D requirements, combined with a boom-bust cycle. Plenty of dead companies in the wake of that industry.
You’d have to be a big OEM with massive RAM needs like Apple to even begin to make it all pencil out - and that’s after dealing with the retaliation by your supply base. They did with their CPUs, no reason they cannot expand to RAM, except that they are just as constrained by fab capacity as everyone else.
See also how well that investment for Apple has fared re: baseband chips. Isn’t the Chinese industry further along to prevent another ZTE squeeze than Apple is fielding a competitive baseband chip despite paying big bucks to intel to buy that business? AFAIK, Apple is still using Qualcomm radio chips in its iPhones, x years and y $BN later.