Is the RAM shortage killing small VPS hosts?(fourplex.net) |
Is the RAM shortage killing small VPS hosts?(fourplex.net) |
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.
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.
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.
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.
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
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...
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.
We use ASRock Rack servers, mainly because the only option for our industry are OEMs like Supermicro and ASRock. Dell and HPE are non-starters, except for our "storage" offering.
Back in 2019, HPE was a good midrange option. Then came ASRock Rack who obliterated HPE with the X470D4U, relegating HPE to high-end enterprise servers. But also made Ryzen-based VPS hosts including yours truly, BuyVM, et al.
Edit: Haha, withing a handful of seconds I got a downvote. :-D
For a real world example, check out Tahoe LAFS: https://github.com/tahoe-lafs/tahoe-lafs
What they could do very easily in this market is bring back frickin Optane and hook it up to a modern PCIe bus with modern PCIe performance.
If Intel made an equivalent high bandwidth Optane product, they’d immediately corner the GPU memory market.
In the end it got too much work with that many servers, something is always broken ; we had some virtualisation and failover stuff, but you have to go and repair it. We were basically two fulltime guys + some freelance and that meant getting into a car and driving for an hour every other day, storm or ice on the roads, christmas or birthday, doesn't matter. It made good money and we sold nicely, but servers are loud and we have been so often there for hours that even with ear protection I think it messed something up. Also crawling in small spaces for wiring etc isn't great for your body. I did learn a lot about Linux and the popular packages; we had our own patches and versions of most to shield from 0-days and save processing waste to put more on one machine without it degrading quality.
I would not do it with reselling ; you have no control; when the police called us for illegal/child materials and so on, we were in charge of removing/blocking that; if you resell, they will likely first just shutdown everything you have and then ask you for an explanation. And after a few time delete you and that's it. Or when there is something wrong technically, you are left holding the bag anyway as many will just blame you (and then after 'some time' your service 'suddenly' is back). You can hire much more expensive stuff with much better support but in our experience, that does not help much when there is a LOT of abuse (and with millions of sites, you have a lot of abuse that you cannot check).
It's fun though; just too much work.
If one day you want to write your memories, please let us all know! I am sure a few here would be more than excited to read about them!
You obviously won't host the service at home, but it's a good intro to the hardware side of things.
People renting VPS to do something, running a service, a website, email etc. But there are other ways to achieve the same without the need of a VPS.
If VPS cost increase to a certain level, some people will just host the service on their own Raspberry Pis through Cloudflare Tunnel, or just simply shut the service down.
You cannot use Lightsail to work around EC2 costs, but you can use Lightsail instead of using EC2.
But, hypothetically if they had a ton of previous gen GPUs (so, less efficient) and a ton of intermittent energy (from solar or wind) maybe it could be a good tradeoff to run them intermittently?
Ultimately a workload that can profitably consumer “free” watts (and therefore flops) from renewable overprovisioning would be good for society I guess.
Second: Even solar and wind are not really "free" as the capital costs still depreciate over the lifetime of the plant. You might be getting the power for near-zero or even negative cost for a short while, but the power cost advantage will very quickly be competed away since it's so easy to spend a lot of energy. Even remelting recycled metals would need much less capital investment than even a previous-gen datacentre.
That leaves the GPUs. Even previous gen GPUs will still cost money if you want to buy them at scale, and those too depreciate over time even if you don't use them. So to get the maximum value out of them, you'd want to run them as much as possible, but that contradicts the business idea of utilizing low cost energy from intermittent sources.
Long story short: in might work in very specific circumstances if you can make the numbers work. But the odds are heavily stacked against you because typically energy costs are relatively minor compared to capital costs, especially if you intend to run only a small fraction of the time when electricity is cheap. Do your own math for your own situation of course. If you live in Iceland things might be completely different.
I don't think this makes much sense because the "waste" of hardware infrastructure by going from 99.999% duty cycle to 99% is still only ~1%. It's linear in the fraction of forgone capacity, while the fraction of power costs you save from simply shaving off the costliest peaks and shifting that demand to the lows is superlinear.
3D DRAM is no magic, it will only give them maybe 2 generations' breathing room if they got the required etching equipment figured out. But others will be doing 3D DRAM with EUV by then.
>CXMT has begun mass production of DRAM using a D1z (sub-16nm) process.
https://www.globalsmt.net/advanced-packaging/decoding-cxmt-d...
They call it "10nm class" later in the article.
It's hard to find much concrete info tbh.
Netcup does oversubscribe/overshare but not sooo much. I have a server there and I don't really observe too much but although I haven't really gotten ways to detect that stealing factor but there are definitely scripts to detect it, maybe I will run it some day but oh well the laziness.
The most overshared vps provider I know is contabo. Literally search anything on reddit,lowendbox, literally anywhere where there are people and they mention about how ~20-30% figure top of my head could be oversubscribed
I am not exactly sure but my point is that when I first saw them, I found them the cheapest option (with their contabo auctions for something at really scale like 96 gb ram or something) but they are literally out of my book as well even as a frugal guy just because of how unstable they are or how much consistent I have seen people struggle about contabo. It's simply unrecommended imho. Netcup's 10x more pleasant from what I see other people's reaction to. People do mention some stealing factor on netcup but overall its really good and that sort of aligns with my experience with them too ig/
This is also the reason why vps's are said to not use 100% usage 24/7 as it can be noisy for other people.
Also another interesting idea by netcup is that they launche virtual dedicated cores (but still vps at heart) where you actually can use 100% usage 24/7 but to be honest on websites like lowendtalk, I have heard that be described as eerily similar to either bare metal instance or this new cloud metal instance terminology as well but the difference to me feels vdedicores atleast on netcup are focused to be more cheaper than dedicated in many instances but I haven't compared it but I have heard it be described as such in lowendtalk ig.
One of my client had dedicated servers in contabo and had to move to OVH because of it.
Not all of the people who want to avoid KYC are doing something illegal ... but all of the people who are doing something illegal are looking for a no-KYC provider.
Having not used AWS for years, I logged in to check it out, navigated through the Kafkaesque maze of their services until I found what I was looking for:
A lone S3 storage bucket, with one file, "Squirrel.jpg". A 200kB picture of a squirrel that I uploaded 8 years ago and can't remember why.
I wonder what the cost to AWS was for keeping track of that and running your CC. There's no way they made money off you / that 12 cents/year cost them *at least* 12 cents to collect every year
I have $6 in charges and so now my account is locked. Lol. Fuck off AWS.
The provider is still making the same money for the same cost on that pre–existing service, so they might not even choose to raise the price for old customers. They're only losing an opportunity cost, and different businesses weight that differently.
I used to run a small shared hosting business in early 2000s and I’d never raise prices on low consumption users, was never worth the risk of them canceling their subscription
USER RES▽ Command
root 70436 systemd-journald
root 14268 amazon-ssm-agent
root 13508 systemd
root 12160 systemd --user
root 10240 sshd: root@pts/0
root 9088 sshd: root [priv]
root 8944 systemd-udevd
root 8704 systemd-logind
root 8320 nix-daemon --daemon
systemd-ti 8192 systemd-timesyncd
systemd-oo 7808 systemd-oomd
root 6492 -zsh
nscd 6272 nsncd
messagebus 5888 dbus-daemon --system --address=systemd: --nofork --nopidfile -
root 5888 htop
sshd 4904 sshd: root [net]
root 4736 sshd: sshd -D -f /etc/ssh/sshd_config [listener] 1 of 10-100
root 2960 (sd-pam)
root 2816 agetty --login-program login ttyS0 --keep-baud
root 2192 dhcpcd: [privileged proxy]
dhcpcd 1680 dhcpcd: [manager] [ip4] [ip6]
dhcpcd 1468 dhcpcd: [BPF ARP] ens5 172.31.8.86
dhcpcd 1168 dhcpcd: [control proxy]
dhcpcd 1040 dhcpcd: [network proxy]I'm not quite sure what points this makes... That's supposed to fit on 128MB? And it doesn't include memory consumed by the kernel itself (which is not negligible at this scale), and linux needs spare for cache to work remotely decently.
$ awk '{ tot+=$2 } END { print tot /1024 }' < list
214.035
To provide "conclusive proof" that it's possible to run a 128 MB Linux system.
For example, if you remove (or configure smaller memory use for) journald, the Amazon daemon, oomd, timesyncd, you are already at around 128 MB userspace, with more that can be removed if desired.
And this is on a distro that is not at all designed to be minimal in memory usage.
How much cache is good depends entirely on what you want the system to do.
Again: it might work, if the math checks out for your specific source of secondhand GPUs and/or solar panels and/or batteries.
https://lukecyca.com/2025/canadian-vps-review.html
Last year, I moved from DigitalOcean to FullHost (their Vancouver datacentre) for hosting a small SaaS and a bunch of personal projects. It's cheaper and FAR better performance.
Crucial had one of the best marketing positions in the business:
- They had a captive supply chain. There was no risk they'd switch from Samsung to Hynix chips but keep the same SKU, so you could buy a second set later and expect it to have similar timings.
- They had a reputation for being conservative about their offerings. There's a lot of RAM out there that meets rated timings only with a voltage right on the edge of what the memory controllers will start to burn out at.
- They were on a lot of mainboard manufacturer's qualification lists, and yet were easily obtained (at least in the US). There are a fair number of brands and ranges that simply aren't widely distributed.
So they were in a place to say "we can charge 10% more for confidence", and considering enthusiasts willingly pay 30% more for RGB and fancy heatspreaders, that's not a bad message. I mean, I've had competent results with plenty of other brands (I have a Team set in my main rig, and it replaced a G.Skill one before the RAMpocalypse), but I always thought of Crucial as a brand you'd use if you were building a machine for work or a family member and didn't want to deal with surprises.
RAM from China already works and as Chinese RAM manufacturers scale up there's no reason they can't be competitive on price. Furthermore, it's more than possible for China to catch up technologically as well. Micron is likely to regret pissing off the majority of their customers if Chinese RAM manufacturers start aggressively targeting the market Micron have left behind, how bad this gets for Micron depends on how quickly they pivot back to the general consumer market.
It might be an eye popping number that it costs $3k to run optic fiber to a rural house but the service costs $100 a month and brings in $3600 in revenue in 3 years. The fiber is likely to be good for 50+ years so they only way this is not "financially viable" is that capital is saving it's powder to invest in bubbles. That is, we are already living in an "Atlas Shrugged" world where capital demands unreasonably high rates of return on investment for physical world investments.
I was reading an article in The Economist about how China would like to have its own Starlink but has the problem that it was not shy to invest in fiber so it has already cannibalized its own domestic market and foreign markets either don't trust China or are too poor.
I'm also sure they can go as far as 5nm like SMIC if they really wanted to, since it's strategic for China, but the cost would only be justified if the current cycle lasts long enough.
plus, the ram manufacturer cycle moves and does this all the time.
Atrioc does a really good job explaining these cycles[0]
But the point is that AI demand peaked when the supply was at its lowest which is why we are caught up in this messed up timeline that we live in. And this has sort of happen in the past too and this industries notorious for it (again watch the video, definitely worth it imo)
But still it feels like we are in this atleast for a year or two hard. Micron is iirc like suggesting what hundreds of billions of $ in factory investment right now and saying that the fastest might open in early 2027
Some estimates 2028 idk, I do feel like the chances of AI bubble popping around this time are likely too.
But still for atleast 1-2 years, we either as consumers or as small vps providers (yes the people who create vps providers are same people like you and me) are absolutely f*ed and the question is around that imo.
[0] The AI Tax Has Started: https://www.youtube.com/watch?v=nipeaKC3dWs
And that will ultimately buy China a lot of time to shove their ram into the market cutting ram manufacturers out of most non-US markets.
I think the major memory manufacturers are simply banking on their ability to flood the market if worst comes to worse. That or I could see some standards trickery around DDR6 (or some new BS standard). It'd not shock me if they coordinated with AMD/Intel to keep the standard secret as long as possible simply give themselves a lead in production.
https://github.com/orgs/community/discussions/10539
For countries, if you meaning connecting to VPS, lot of countries have good IPv6 connectivity now. For me both ISPs I use have native v6. This will differ from person to person.
That is, if being unreachable below "presentation layer" is acceptable in the first place, but I guess the question kind of presupposes this.
This destroys all the AI companies, and more.
Micron Fab 6 currently makes about 2% of global memory production in Virginia.
The plant is currently being expanded and upgraded to Micron's 1a node.
The ram pricing is coming directly from manufacturers.
Eh, I don't think we were on the same planet then. Even post crypto pre-AI GPUs were far more expensive than they were before said crypto. We just got used to paying $1000 for a mid tier cards.
If you're using AWS/GCP/Azure to just host a couple of VMs for a small group you're massively overpaying.
Not including the faster SSD & included traffic
Personally, the only thing I know of that is a true deal vs. competition is cold storage of data. Using the s3 glacier tiers for long term data that is saved solely for emergencies is really cheap, something like $1/100GB a month or less.
AWS is usually not the cheapest EVER when it comes to offerings like EC2. If you aren't doing cloud-native or serverless at AWS, you're probably spending too much.
But if all you really do with cloud stuff is "ssh into a server I have" (which covers a ton!) then you'll find much cheaper/more performant elsewhere.
A lot of people do it.
People feel the battleship is safe and familiar. For most businesses the extra cost is not even noticed. Even a small business spending $500/month on hosting instead of $50 is not going to notice.
Also, if something goes wrong (e.g. your AWS region goes down) its far easier to explain to a manager or client that "its Amazon's fault and lots of stuff is down", rather than "its Digital Ocean's fault".
I view a large percentage of "cloud" usage like Teslas stock price: it's completely detached from reality by people who have drunk the kool aid and can't get out.
My theory: Microsoft is extremely interested in tracking you via your IPv4.
There may indeed be some tracking that MS does via IPv4, but it's not a good way to do it.
I suspect any such tracking is essentially just some cruft that snuck in (either their own or legislative) in the early 2000s, and nobody thinks it's their problem to make go away.
That said, that IPv4 is a poor way to do tracking doesn't guarantee there's no manager demanding it: any corporation eventually gets someone with no technical knowledge demanding bad solutions.
My point is small businesses tend to be owner managed and therefore more cost sensitive and the costs are still not a significant factor. AS businesses get bigger its even less significant and its other people's money.
And at least here in Germany, they are nickel an dimming on the hosting cost.
The problem, however, is IO controller support has been dropped, many new CPUs don't even support DDR4 any more, especially mobile ones.
