They are forbidden to buy foreign equipment beyond their current process node, which is already obsolete, die size is 40% bigger than Samsung, not to mention lithography, the big 3 are using EUV while they are stuck with lobotomized DUV.
They can start making some decent money now, but vastly expanding capacity as is means enormous losses if the cycle went downward a few years later, that's how all previous makers went bankrupt.
They can squeeze out a bit more performance if they are ready to go beyond their current node using only domestic equipment and be blacklisted by the US government.
But the cap is there, unless they can make a working EUV machine in 5 years, they are doomed to be a minor player, if the current cycle even lasts that long.
likely naive question: why it is a problem? I would be fine if RAM in my PC has 10 times larger physical size if it is overall cheaper.
I guess that larger may have more power draw, but given costs of RAM and electricity and power draw of RAM it sounds unlikely to be a problem.
At a high end it would run into real-estate prices - at some point using half of room for computer stops making economical sense, given costs of rent or buying flat space. But just doubling size of PC does not sound like a bad tradeoff if it would be say 20% cheaper. Or 50% cheaper.
Is it about not fitting existing motherboards?
Is there reason why they cannot just make memories physically larger? It is "only" 40%, not 40000%
They don't currently have the tech to compete on HBM3 production, but they can produce DDR5 memory, and they will undoubtedly be scaling up production on this.
Which suits the rest of the world just fine. More for the rest of us, and if the single-digit-percent portion of their market that the US represents wants to lock itself out, no skin off anyone else's nose.
And? That's good enough. My daily driver desktop, which I use to do development, plus play a few games (FC5, Dirt Rally, etc) has 16GB DDR3.
For 90% of computers in use today, including laptops, that RAM you call obsolete is fine.
Users aren't going to complain that a document which takes 3s to open now takes 3.5s
What you need as far as RAM goes, to make the computer perform acceptably is capacity. Users get a bigger performance boost by going from 8GB RAM to 16GB Ram and from 16GB Ram to 32 GB RAM than from DDR3 to DDR4 or from DDR4 to DDR5.
These three have collectively committed what, approaching $50B towards construction of new facilities and fabs in response to the demand?
The memory industry has traditionally projected demand several years out and proactively scheduled construction and manufacturing to be able to meet the projected demand. The last time they did that, in the crypto boom, the boom quickly turned into a bust and the memory makers got burned with a bad case of oversupply for years. With that context, can you blame them for wanting to go a bit more slowly with this boom?
Sure, the new fabs won't be up and at volume production until late 2027 / early 2028, but committing tens of billions of dollars to new production facilities, including to facilities dedicated to DRAM rather than NAND or HBM, is hardly 'abandoning'. They're pivoting to higher profit margin segments - rational behavior for a for-profit corporation - but thanks to the invisible hand of the (not quite as free as it should be) market, this is, partially, a self-solving issue, as DRAM margins soar while HBM margins compress, and we're already seeing industry response to that dynamic, too: https://www.guru3d.com/story/samsung-reallocates-of-hbm3-cap...
Look at what happened to Crucial. Why would Micron axe it's whole consumer RAM division if it was just experiencing a temporary drop in DRAM supplies until new fabs were brought online? Samsung and SK Hynix may have changes in priorities in the coming years, and in the case of Samsung I'm sure they'll still make sure to supply sufficient DRAM chips for the devices it manufactures (phones, TVs, etc...) but Micron has made it's current intentions fairly clear. They'll probably work with OEMs, but they're unlikely to return to selling to the general public any time soon.
Except for RAM from YMTC, which the USA gave a near-death sentence to by placing it on the Dept. of Commerce “Entity List” so no USA-associated business can do business with YMTC now.
One problem with US sanctions is it could hurt US companies too, like in the case of cutting-edge EUV and CXMT. This is when China is actually a hero and not a villain.
If corporations in western (aligned) countries stopped feeding sovereign wealth funds and private equity with profits and actually invested something maybe they could compete with China more closely, even with whatever shenanigans the CPC get up to with state support.
What I found is that it is cheap because the cores, and presumably ram, is old. Like, 2013 era Xeon E3-1275 v3 old. But that's fine! Old hardware like this uses old ram that is less affected by the current shortage. It's good enough for my needs.
Unlike Fourplex.net which uses modern ASRock Ryzen 9000 servers, Qeru.net used older HPE DL360 Gen9 servers.
I gave 3GB of RAM for $3-4/mo then. But these servers weren't very fast. I ended up selling the business, and am happy I did.
It goes on about DSL and dial-up for some reason?
And yes VPS providers are affected by ram shortage. Turns out things that need ram are affected by ram shortages. 5 stars for the insight
IPv4 shortages didn’t kill it, and I don’t think this will either.
But for a new provider like us, we'd have to spend more than an established player like BuyVM or RackNerd who bought most of their servers pre-AI-boom.
It's 12 dedicated cores of a modern EPYC CPU, 32GB RAM, 1TB NVMe.
I got that offer during their Black Friday sale and pay €25/month (price before VAT), plus the offer I got has a 2TB NVMe instead of the 1TB one.
CPU launch price 11000 USD. RAM will likely be another 10000 USD
20000 / 8 customers / 40 USD/mo = 62 months just to recoup CPU and RAM let alone other components.
Weird, whenever I napkin math offers of any HW for renting, I get that I could buy it myself in 1-2 years of rent. Sometimes faster.
Do they not intend to recoup the costs of HW? :)
the most important question is the power, such a setup will blow through easily 500W of power. Granted, a datacenter may not pay 33 ct/kWh but even then, 1/3rd of the monthly income will just go towards naked electricity, not including cooling.
Electricity costs are ridiculous.
so for 3 months, I got 1 month free, another month with the 5 euro voucher and essentially only paid for the last month.
I ended up paying 8$ in total for the 3 months. Although after the 3 months, I have to pay more around ~5-6$ per month but its still a recurring deal so I guess I might as well continue it (given I have setten everything up)
Some of it is definitely lose making as well to lock in. I usually try to see what gets me to buy a deal because i usually don't buy many things especially online and its usually within some deal concept where I feel like "winning" that I do.
But to be honest, I did some napkin math and honestly, at least for my VPS (although Its usually idle but I have hosted many services on it or experimented with it, its honestly pretty fun imo, I definitely feel so much more freedom imo somehow after getting a vps its hard to explain I suppose or maybe not idk)
But my napkin maths suggests that they lose quite a lot of money with atleast my deal in particular so I got better off getting this deal than say if I was netcup myself somehow (because they probably lose net money on my deal if I actually really use my allocated limits I guess) atleast for the 3 months but I feel like netcup's still one of the more cheaper options and I got a recurring discount deal with it too sooo I would still consider them to have only marginal profits over a long term deal as well.
SO I guess overall, this was a really profitable deal for me.
Your deal's pretty cool too, honestly at large scale. I have done scraping of lowendtalk and analyzed and done many decisions and infrastructural decisions and honestly I will say taht for anything somewhat large scale like you mention 32 gb or something, I found hetzner boxes to be the best/most referred to option & OVH kimsufi line is created to be the cheapest dedicated boxes.
But yeah regarding OVH i recommend this website that I found regarding OVH KS line or just in general https://checkservers.ovh
Just checked it right now and I see Xeon E-2274G 4c / 8t 4.4 GHz+ 32 GB DDR4 ECC 2666 MHz 2 x 960 GB NVMe for around 37-38 euros ish
Also Netcup's definitely not small. I looked at anexia group and I wouldn't constitute them as small. For understanding what small means I guess when I mean small, I mean lowendbox single shop providers and others.
Also regarding Netcup, its payment system was really ass from what I felt like. Like It was the first company which made me feel like I wanted to pay it money but I had to make a CCP or some account nad literally had to go through like 3-4 really long things imo which really made me say (wtf) in my head each time and frustrated me only to give me a stripe link in the end.
It had me so pissed that for 2-3 days one idea of simplifying its whole process and other such providers stuck in my head. Its payment / the whole process of I want to pay -> create account -> actually pay had like 5-6 maybe 7 steps or something and I seriously can't tell you how I frustrated I was but the deal was too lucrative lol
I've personally built a smart watch that connects to the internet, helps me perform organisation duties, all whilst running (micro-)Python, in far less RAM. To do the equivalent on desktop would require some form of sandboxed browser engine, JavaScript and some weird client-server stack.
I recently upgraded my servers with RackNerd from 768MB to 1GB because it was near free to do so, but I didn't need that extra memory. Before anybody claims they don't do anything, these servers deal with thousands of users a day.
I'm cautiously hoping that we start to see a period of optimisation again. There is currently zero incentivize to optimise anything any more, even if it negatively affects the end user.
The "did we have X 10 years ago" argument also misses a a lot of modern software requirements. Yes, we had mid-range laptops and budget gaming PCs in 2016, but the memory footprint of everyday applications has exploded since then. Electron-based apps (Slack, Teams, VS Code, Discord) routinely consume 500MB-1GB+ each [5], and it's entirely normal to have multiple running simultaneously. A typical "light" workload in 2026 easily uses 2-3x the RAM that a comparable workflow needed in 2016.
So we're getting squeezed from both sides: applications demand more memory whilst memory itself costs more. An 8GB laptop was perfectly serviceable for office work in 2016; in 2026, it's borderline unusable with a few Chrome tabs and Teams open. The same kit that cost $55 in June 2025 now costs $150 [2, again]. That's a 150%+ increase in six months, pushing capable systems out of reach for budget-conscious users precisely when software bloat is making RAM more essential than ever.
[1] https://gamersnexus.net/industry/3212-ram-price-investigatio... [2] https://gamersnexus.net/news/ram-wtf [3] https://www.newegg.com/p/pl?d=ddr4+ram+8gb [4] https://pcpartpicker.com/trends/price/memory/ [5] https://josephg.com/blog/electron-is-flash-for-the-desktop/
I do think we're more efficient and matrix operations are better (again, GPU/TPUs), but by and large, the computer hardware world has stopped exponential growth.
The period from 1990 to 2005 was amazing. Both in transistor counts and clock speed, it seemed like we were nearly *doubling* performance with every new generation. Memory and disk space had similar gains.
Today, I can buy a "Corsair Vengeance LPX 32GB (4x8GB) DDR4 PC4-28800C18 3600MHz Quad Channel Kit" for £259.99 [2]
[1] https://web.archive.org/web/20150303141114/http://www.overcl... [2] https://www.overclockers.co.uk/corsair-vengeance-lpx-32gb-4x...
Some dirt-cheap VPS maybe too unreliable to run anything serious, that's why they sell that for dirt cheap. And their consumers generally won't complain about sudden server reboots, because that's what expected for the price.
If they increase their prices, then many of their customers maybe better off just use Linode or DigitalOcean etc instead, as these vendors provides better guaranty on stability.
Not sure what the lift might be, but in theory everything should be relatively similar in future state, just more expensive. This is basically a form of inflation.
The comparison to 2000s telcom market does not seem similar at all to me.
One of the hard rules we learned pre-pandemic was that services attached to usage based billing should really exit on error. It's a lesson I'm keeping in mind working with agents and routing (and the main reason I'm local-first).
If I need to host something small, I don’t want to mess around with the many permissions and quirks that are required to deal with AWS. It is often much easier to just setup the server on a standalone service.
When I worked at Microsoft, I seldom used Azure for personal use due to it being expensive and complicated.
Whereas I have plenty of Fourplex.net servers because even on half the salary, it's affordable enough for 16 Tor exit relays and two personal web/email/Mastodon servers.
There's also the human touch in terms of who you talk to: a lot of the smallest VPS hosts are 1-2 people, both technical, so customer support = sysadmin = contact for everything.
The set of companies that sell VPS and the set of companies that sell dedicated servers aren't the same set.
Does it have a billion 9's of reliability? No, but I don't care, it has literally never not worked when I've used it
Customer Service so far has been human, but that will vary greatly for the provider
I also use a different provider for work related hosting, and the reduced latency of being within 20 ms of the DC has been probably the single biggest (perceived) perf improvement my users have ever seen, specially on the legacy webforms platform we recently decomissioned (We're a bit too geographically far for most Datacenters of most large providers)
Outbound data pricing is a potentially huge saving.
AWS is as much as $90/TB outbound with 1GB free. Hetzner is $1.20/TB (in EU and US) with 1TB/20TB (US/EU) free.
(Good) Smaller places are more likely to have actual technical staff you can talk to.
That is a nice way to have a static IP on the internet and enough resources to do small things like host a nameserver and/or OpenVPN/Wireguard.
I may have had 4 hours of downtime in one year, always announced days in advance.
Meanwhile, small service providers might not actually need those premium features, and just want something that is cheap and makes economical sense. They don't need the state-of-the-art hardware and just want something that works.
That's why while the AAGO (AWS, Azure, GCP and Oracle) attracted a lot of big corpo, that is, almost all of Forbes 500s used them, DigitalOcean and Vultr, with their $5 plan, is those who won the small businesses.
VPS services are usually really, really simple and fairly cheap.
I'd say that actually VPS prices is where we actually see computing prices going down rather than on the big 3.
AWS used to optimize further and pass down the savings to the customers back in the day, now they don't do it anymore.
Never had problems with downtime and I payed, like, 40 bucks a year over 3 years. I think I had to restart the thing once because of something dumb I did on my end.
AWS does offer Lightsail which is similar pricing.
If you’re a government agency or a company you don’t care about saving $14/month, you want a secure provider. And these hosts are not secure, you’re basically just on your own.
I replaced with a home server and it costs way more just in power hahaha.
It’s weird because in most of this industry scale results in lower prices. Not in cloud.
I use them to run wireguard to evade geoblocks when I'm travelling, a few redundant monitoring scripts alerting me of reachability issues of more critical stuff I care about, they serve as contingency access channels to my home (and home assistant) if my primary channels are down.
I get no support, no updates, it's all on me - which is fine, it allows me to stay current and not lose hands-on practice on skills which I anyway need for my job (and which are anyway my passion). I don't even get an entire IPv4 - I get.... 1/3000th of it? (21 ports, the rest are forwarded to other customers). Suits me fine.
Can't get a Linux box to idle (or even install) under 512M these days.
Can't find a web developer worth a shit who doesn't think he needs a Python backend application server to print "Hello, world" when you could do this with a static page served with something like OpenBSD with two-digit RAM requirements.
It's not the RAM that's changed; it's everyone around the RAM.
A coddled generation who were taught that AWS is the Internet and live in abstractions certainly hasn't helped.
why
Windows NT was routinely run with 32 MB of RAM TOTAL and the event log is basically unchanged 30 years later.
Wait till they rewrite it in Rust!
I still have 64 MB VPS (OpenVZ) which I use in production since 2012. It runs DNS, VPN, some logging stuff.
Install can be tricky indeed, but if you have installed system, it's easier.
But this does highlight the fact that most of our hardware is produced (and thus can be restricted) by a few cartel-like players, just like what's happened in the Internet industry.
https://www.syracuse.com/micron/2025/11/micron-chip-factorie...
- first graphics cards
- then RAM
- now NVME drives
The result? PCs become hyper-expensive and also hard to assemble for mere lack of components, then we get to VPS and keep going. The computing that remains is mobile and cloud, "the only integrated platforms," meaning the absence of digital ownership for most people, total control by the giants.
Think of it in these terms: nowadays GNU/Linux is starting to be a sought-after desktop, nowadays we can self-host with levels of accessibility even for newbies that we didn't have in the past, and the interest in doing so for those who know something is stronger than ever. And here come the reasons not to do it, economic and also physical.
Does it sound like "flat-earthism"? Well, it also sounds very realistic, at least net of the effects, we can say it's conspiracy theorizing, but thinking ill is a sin but you often guess right, said a prime minister a long time ago with a list of extraordinary scandals behind him.
as always: imho (!)
idk ... the article is like ... comparing apples with oranges:
telecommunication-networks and similar kind of infrastructure are also called "natural monopolies"
* https://en.wikipedia.org/wiki/Natural_monopoly
and often those monopolies where (initially) build with public funding.
i don't think these characteristics apply to DRAM/semiconductor related facilities as we have them today.
i think the only "thing" which could "save us" from our own and the DRAM manufactures "greed" are new factories ... anywhere, but right now china looks "the most promising" at least to me.
additionally: i think these articles could be seen as somewhat related
"TSMC Risk"
* https://stratechery.com/2026/tsmc-risk/
and
"The Benefits of Bubbles"
* https://stratechery.com/2025/the-benefits-of-bubbles/
in a nutshell: the "upside" / result if the ongoing AI bubble pops could be
1. more semiconductor facilities
2. more power-generation facilities
cheers a..z
Update: Fourplex (this host) uses a 1GB minimum.
For context, I think I have recently been part of the lowendbox community and discussing things with providers etc.
Recently one of their providers actually cited the price increase directly to ram increase.
I was in talks with other VPS providers one of whom literally shared me a photo in the forum of what it costs them to get ram in their area.
I have been in talks with many many people within the VPS community and I have said this so many ocuntless times in here or lowendtalk (you can read my comments although I used to say this ~1 month ago iirc)
Heck, I wanted to build my own cloud. My father works in the bandwidth business (essentially you can say broadband/wifi)for ~10-15 years/ has very strong ties in it.
I on the other hand like/love tinkering with software side of things including virtualization and have made projects about it. I shared some request one month ago to talk to my dad and he said he was interested in datacenter idea
But I shit you not, it literally doesn't make sense even though A) my father has its own office, b) he can get essentially lots of bandwidth for free, C) we live in country with very cheap electricity comparative to many others.
And even if you have lots of hardwares, it doesn't just make sense to still have it if the price of replacement has increased so much far far.
There is also a forum called lowendspirit and I saw someone shut down their service citing this as well.
There are many posts on lowendbox (in which I have talked to) talking about the same thing as well
Simply I don't know how to state it, but the 4x price increase within ram simply literally didn't make sense. No thanks. I (or we?) are gonna wait the AI bubble or do something else/work more on software sides/just relax instead of making this.
I haven't even read the post but this is the most connected post I have felt in a long time!
To be really honest, I will admit that right now, personally I am either looking at mac 512 gb ultra or something which are still not ramflated from what I can tell or something or to be really honest, most cloud providers are burning money to not create a frenzy or still have pricing the same.
Personally, I love small cloud providers but recent times indiciate to me to favour stability. I have a netcup server myself but hetzner is an absolute love too from what i can tell and I can recommend any & OVH is another brilliant solution. All European.
Because these have/probably have large supplies. Soo like I know that OVH literally creates their own machiens, buys land and does everything and is a stock company so they are kind of a capital expenditure -> getting maximal profits machine in some sense (like yknow eliminating as much as middleman as possible kind of thing?) soo I guess the chances of OVH has lots of money to burn through to survive & brand reputation is crucial too and usually many of these run Black friday sometimes too
I guess another part is that most cloud providers can offset the losses right now for the hardware which can get cheap when the AI hardware bubble bursts but I odn't really know. I guess OVH or others might internally discuss this possibility too.
Atleast for me, I am gonna build software regarding the projects. I have found a huge disprecancy where whmcs and others have an almost monopoly so I am probably gonna spend a few months creating an open source project alternative (technically I already have a preview working with gvisor which I used with my netcup vps itself because gvisor can work in kvm machines itself not requiring nested kvm)
To be honest, shit's really hit the fan in this industry in ways that I can't tell. I have/had been in talks with an UK provider (https://xhosts.uk) [promoting because seems like a good guy] & he said that he's suffering the losses/taking the hits (for additional context he is disabled but iirc doesn't take disability checks) & he is from what I can tell actually impacted by such rising costs.
So being honest, this mad fuckery of an AI bubble is hurting an disabled person who wants to make a genuine living doing things he like.
I really want this bubble to pop right now. Like please. Just pop it and have the prices of hardware low because I talk to these providers and they are normal people like you and me and their business is impacted too all the while these large companies like AWS,GCP,Azure is what people use to rent or some shit in the first place with the returns/rent seeking that they are while you can get things for cheaper in lowendbox and other places in the first place but yeah of course you won't get the same stability (but will get pretty damn close imo) for much cheaper but what many people want is the ability to discount that when half of internet shuts down because of AWS east 1 just feeling like it, the customers wouldn't blame them directly for picking AWS but I guess if they use something even reliable lik ehetzner and say hetzner might have one downtime (which is rare but yea lets assume so) then maybe customers or business owners or anyone higher or lower in the team might push on the original person who might've suggested hetzner and they might get (fired?) or these are the justifications I see as to when people suggest that hey aws is bloated and burns so quickly, why not use alternatives like hetzner and many many other awesome services.
Thanks for listening to my (rant) & have a nice day (as one might have after the AI bubble given that no matter how harshly I write, no matter what I do, quite frankly I have no say in the AI bubble, so I personally what I like to do is actually use AI for free or the cheap and burn so much money of these AI companies to actually build open source cloud solution or something when the bubble pops but its a bit of an ideal but something I am working on I guess with the gvisor thing.
I don't know how to point it out but the tech behind AI is cool. There are gonna be some persisting changes (open weights model like kimi k2.5) etc. but I just don't know.
I really don't like the current timeline I live in regarding AI man. And I would consider myself to be sort of a person who actually uses AI many times but I do feel as a net, I feel as if AI might have been negative or atleast AI in the past year for some reason. Claude 3.5 sonnet or Deepseek v3 & things were going okay. Beyond that something definitely changed I mean there was already a bubble during DSv3 days but somehow it was still realist (the stock market). Right now I feel a fundamental disconnect between reality and market and the longer it continues, the worse its aftermath.
I really don't know how to describe what I am feeling but hopefully I have done a good job (with this lengthy comment) to explain all the things that i felt like talkjing about
Thank you to the author for creating this article so that we can have this relevant discussion. I haven't read it so I am gonna go ahead and do it just now.
How can you trust Gary from GaryHosting not to just steal all your data? How can you trust him to have redundant networks? You just can’t.
On the other hand, I can trust Gary. Gary's personally responsible for GaryHosting, and he obviously takes that role seriously, given he slapped his name on the front. And if Gary fails, I can just switch to a different provider. Gary doesn't have a moat, he sells a commodity. His only advantage in this world is treating his customers well enough that they don't leave.
A. rendezvous services so clients can connect to one another,
B. storage/retrieval of encrypted data where the host does not have the key to decrypt,
C. transport of encrypted data which cannot be known by the host due to B above.
> How can you trust him to have redundant networks
You can't, so abstract that away at the application layer. Make it not dependent on a single host or network.
