Hetzner (European hosting provider) to increase prices by up to 38%
old.reddit.com260 points by doener 3 hours ago
260 points by doener 3 hours ago
Running a small project on Hetzner from Germany. Got the email this morning. Honestly, even after the increase their dedicated boxes are still absurdly cheap compared to what you'd pay at AWS or GCP for equivalent specs.
The real story here isn't Hetzner being greedy. It's that AI companies are vacuuming up every DRAM chip on the planet and the rest of us get to pay the tax. I priced out a RAM upgrade for my home server last week. Same kit I bought 8 months ago for 90 EUR is now 400+. That's not normal market dynamics.
What worries me more is the second-order effects. Startups that would normally spin up cheap VPS instances to prototype and iterate now face meaningfully higher costs at the exact stage where every euro matters. The "just deploy it" culture that made European indie dev scene so productive was built on sub-10 EUR/month boxes. Those days might be over for a while.
> It's that AI companies are vacuuming up every DRAM chip on the planet and the rest of us get to pay the tax.
DRAM is priced based on supply and demand, like every other market.
When demand goes up, the price goes up for everyone. It’s not a “tax” on the rest of us in any sense. There’s just a lot of demand everywhere.
> That's not normal market dynamics.
This is actually a textbook example of markets functioning in response to a demand shock where supply cannot be increased rapidly.
I do find it interesting that so many people think “market rate” means the opposite of what economics teaches, and that prices should stay stable and not change much when the economic conditions change.
I also find it interesting to read all of the “we shouldn’t let them…” takes in response to this situation. The DRAM market is international. Trying to restrict it in one country would just see the data centers get built in another country.
Factory capacity does not follow market dynamics easily
I think the usefulness of market dynamics is their ability to follow things like factory capacity, which are themselves hard to follow, not the other way around.
Most markets don't have three purchasers trying to corner the entire supply of one product.
Economic history is full of examples of demand shocks. This is not some unique situation that has never occurred before.
This is actually a clean commodity price spike because it’s specifically not for market manipulation or financial engineering. It’s because demand for this product really did explode overnight.
> This is actually a clean commodity price spike because it’s specifically not for market manipulation or financial engineering. It’s because demand for this product really did explode overnight.
Based on how the same 3 billion has been circiling between Anthropic, OpenAI, Nvidia, Google, Microsoft, Amazon, and a few other companies... I really doubt that this is the case, to be honest.
By 3 buyers who have no known plan to finance the purchase orders they have made.
Economic history is also full of examples of bubble bursts.
But aren't those the same startups that think they need to run on AWS EKS instead of using a single cheap server? The cheapest used Hetzner server currently is €39.24 / month:
- Intel Core i7-6700 - 32 GB - 2 x 480 GB Datacenter SSD - 1 GB/s - 20 TB traffic
Their VPS are even cheaper. And you can run a lot on this.
If your only need is a lot of bandwidth with very low server CPU use that’s fine.
That CPU is ancient, though. Over a decade old. That DRAM is 2-channel DDR3.
This could be a good deal for someone, but entrusting your startup’s operations to a 10 year old slow computer in Germany instead of using EKS would be an extremely short sighted move. A startup should be developing software and shipping it quickly to validate the market, not pinching pennies to save the equivalent of a couple hours of developer salary.
Except 40€ a month is extremely poor value for this CPU that's more than a decade old.
No, that's actually a really good deal for dedicated hardware with those specs. For a project sized for hardware like that, the CPU is a lot less relevant than the RAM and storage and transfer.
8 threads at 3.4 GHz, 8MB cache. Seems fine, depending on your use case.
Measuring CPUs by thread count and clock speed is not a good way to gauge performance. A current gen CPU would be several times faster than this old CPU.
Depending on workload, this old CPU might be as slow as a 2 thread or even 1 thread current gen server.
> It's that AI companies are vacuuming up every DRAM chip on the planet and the rest of us get to pay the tax
European victim mentality !1!
/s
When I have looked on Newegg and on Amazon USA last month, I have seen even greater prices than here in Europe, by 30% to 40% greater, which is reversed from previous years, when computers and computer-related components were cheaper in USA than in Europe.
So I think that the victims are all the computer users of the entire world, with the exception of a negligible number of humans tied to the AI companies. Moreover, the US victims appear to be hit by the price hikes even more than in other countries, at least for now.
> That's not normal market dynamics.
It is, in fact, normal market dynamics.
It's normal market dynamics for markets dominated by quasi-monopolies, which is why regulation should have prevented the existence of such markets.
A single company that never made a profit outbiding the entire world is normal?
Yes. In fact that's not just normal, that's very frequent in market.
Maybe not normal market dynamics, but typical human behavior: https://en.wikipedia.org/wiki/Vincent_Kosuga#Cornering_the_o...
> It's that AI companies are vacuuming up every DRAM chip on the planet and the rest of us get to pay the tax.
That’s a redefinition of the term “tax.” The supply of DRAM is constrained because capacity decisions were made years ago. And these AI companies highly value the DRAM so they’re willing to pay more. It’s not a “tax” on everyone else, it’s just supply and demand.
Man alive, stop being performatively dense. Getting to pay "the tax" in this context is just a colloquialism equating getting to pay "the burden".
A significant part of this is probably just the hockey-stick growth in the price of memory we have seen in the past 6 months. Would be surprised if this wasn't impacting their bottom line for maintenance.
RAM increased the most, but also SSD and HDD prices increased significantly. And it seems there are also supply problems, so you can't even be sure if you get the components you want at higher prices.
There is another factor at play here: EU hosting providers that are not owned lock, stock & barrel are few and far between and Hetzner has a very nice sales representative in the White House.
Can you expound on that? I'm not sure I get what you're implying.
Pretty sure they are implying that the actions of the current president/administration are causing people to re-evaluate US dependencies. I don't really understand the first half
I think in the first part they are implying that there are very few independent companies to turn to.
(I also prefer comments that are clear without insinuations).
What about all of the long-tail providers that are often listed on lowendbox.com and similar sites?
Ahh, the sales rep is Trump, that makes sense, thank you. I thought Jacques meant they had lobbyists somehow.
1. There's no meaningful European competition.
2. Trump is making everyone scared to use US hosting.
So they're leveraging for extra profits.
That the USA is no longer seen as a stable partner for the long term and that Trump with his idiotic policies and tariffs is driving sales for the few EU hosting scale-ups that are not somehow owned by America.
That Trump makes us very motivated to stop relying on American tech.
This doesn't solve the issue that globalism caused. Europe doesn't make DRAM nor has the know-how to quickly bring factories online which usually take 10+ years.
We are tied to American economy and if AI companies start driving prices up not only DRAM but basically everything will become more expensive.
Europe has stopped making DRAM relatively recently (Qimonda).
This should have not been allowed to happen.
America doesn't manufacture DRAM either, this is all South Korea and Taiwan.
??? Micron has DRAM megafabs in both Idaho and New York state.
They don't "have them", they're building them.
https://www.micron.com/us-expansion/id
> Micron has already achieved key construction milestones on its first Idaho fab with DRAM output scheduled to begin in 2027.
https://investors.micron.com/news-releases/news-release-deta...
> Production is expected to start in 2030 with the fabs ramping throughout the decade.
Until they start outputting DRAM in any meaningful quantity, they're not relevant.
It looks like it's still a big difference between how the US and EU are responding to the chip supply wars. The US is actually building their own manufacturing capabilities domestically while the EU is apparently doing nothing, which is unfortunate.
Are those plants still functional after CHIPS act was axed? I thought they mainly produce in Asia now.
Well first of all, the CHIPS Act was not "axed", it is federal law passed by an overwhelming bipartisan majority of the House and Senate. It would take a complete reversal of congress to repeal it and it's still very popular among both parties.
Where do you get your information from?
> after CHIPS act was axed
This is news to a lot of Americans! The 2022 CHIPS and Science Act is codified federal law. I think a lot of states (Arizona, Idaho, New York) would be very interested to learn that the funding for the infrastructure that they are already building has somehow gone poof.
> Currently, 100% of leading-edge DRAM production occurs overseas, primarily in East Asia.[0]
They make DRAM for cars, not computers, in the USA. They've promised they'll bring manufacturing onshore any time soon, which effectively means they'll wait until Trump forgets about it.
0: https://www.nist.gov/news-events/news/2025/06/president-trum...
That's not how it works, DRAM substrate (the actual chip that contains memory cells) is shared. It's only the packaging that differs.
Isn’t there also basically 0 American DRAM?
Micron Technology, Inc. is an American semiconductor company that manufactures computer memory
They don't produce them within the US. They're building some factories to do so in the future, but as of now their output is 0.
The newfound desire to move away from American cloud providers isn’t related to pricing, it’s about the perception of growing instability within the American government, the perception of deteriorating freedom of speech, and the perception of an increasingly non-neutral business environment.
E.g., if I’m running a business in the US and I don’t kiss Trump’s ring (and pay bribes), if he becomes dictator for life in 2028, all bets are off for my business.
Both the EU and USA import the majority of their computer equipment, and the USA is placing heavy and unpredictable tariffs on those goods. It’s hard to argue that a business should bet that data centers will be cheaper in the US than in the EU if Trump is the last democratically elected president.
The most stable places to do business in 2026 are probably the EU and China.
The strongest reaction of EU would be to subsidize RIPE small LIR fees to 0€ and embrace decentralization.
Here's to hoping the IOU purchase orders for RAM and SSDs get cancelled... Though I think folks are hedging that this will happen and limiting new suppliers.
The post seems to indicate this is just for VPSs, which doesn't seem true, the email I just received from Hetzner mentions price increases for dedicated servers too.
The ones I'm affected by seemingly:
Product -> previous price -> New price as of 1 April 2026
EX42-NVMe (FSN1) -> € 49.65 -> € 51.13
AX41 (FSN1) -> € 49.73 -> € 51.22
AX41-NVMe (FSN1) -> € 49.73 -> € 51.18
Server Auction -> € 65.22 -> € 67.18
Here is the full list of the updated prices: https://docs.hetzner.com/general/infrastructure-and-availabi...
Seems it's because of increased cost of hardware, and they seemingly tried to avoid increasing the prices but they couldn't. From the email:
> The underlying causes of the increased costs are, among others, the exploding demand for AI-related computing power and for cloud services. In addition, raw material prices and production costs have also generally risen for manufacturers. The costs for RAM and SSDs especially have risen by a large amount. For example, the cost for DRAM memory has increased up to 500% since September 2025. And according to market researchers like TrendForce, this price trend will continue throughout the year.
> We have genuinely tried hard to optimize our costs and to prevent increasing our prices for as long as possible. But we can no longer compensate for the strain that it has placed on our operations. We want to continue to deliver quality products that meet both our standards and your expectations, so we must take this step.
It seems we will run out of hardware by March?
"Hard drives already sold out for this year" - https://www.theregister.com/2026/02/20/ai_blamed_again_as_ha...
Time for an AI tax on the hyperscalers.
> It seems we will run out of hardware by March?
What happens when an unstoppable force (building everything in Electron because hardware is cheap) meets an immovable object (oh no hardware is expensive now)?
We go back to the demoscene days, being creative with what we have instead of shipping Electron junk.
Maybe we need to let go of our auto-scaled 100 pod service mesh for a todo list app, and just deploy it bare metal on 2 servers.
consumer RAM is not what's creating shortage. Data centers doesn't run electron to train the model or for inference
Sure, consumer ram isn't causing a shortage, but it's affected by the shortage.
Every RAM producer is stopping their consumer grade RAM production to provide ECC-RAM and VRAM now. Micron discontinued and closed down Crucial brand as a whole.
So, getting systems with higher RAM capacity is getting harder (from laptops to smartphones). So, for a couple of years, we need to stop using Electron so much and use what we have efficiently.
Data centers, esp. AI hyperscalers do not care about efficiency for now, because they can suffocate consumer-grade part of the hardware marketplace and get anything and everything they want. When their bubble pops, or the whole capacity ends, they need to learn to be efficient, too.
For reference, a well-optimized cluster runs at ~90% efficiency even though they have thousands of users. AI hyperscalers are not there. Maybe 60% efficient, at most. They waste a lot of resources to keep their momentum.
I have a silent hope that because of this change we all will get ECC ram and that consumer CPUs will get proper support for them.
AMD's RYZEN already supports it. ASUStor's latest generation of NAS devices come with AMD x86_64 processors and ECC RAM as a standard, but ECC RAM in SODIMM format was not cheap, even when the RAM was cheap, either.
They effectively do. They’re trained by brute forcing 100TB of training data through them, rather than any logical learning technique.
A human doesn’t need 100TB of books to learn the alphabet.
> A human doesn’t need 100TB of books to learn the alphabet.
A human does need 16ish hours per day of audio/video content for several years to learn the alphabet.
I used a single letter stencil to learn the alphabet, actually, and nobody strapped me to a chair to watch or listen something 16h a day.
Living inside a normal home with my parents was enough for the audio part.
The 16 hours of audio/video per day was a reference to being alive and hearing/seeing things for years before you actually could learn the alphabet.
It was not meant as literally sitting at a screen with audio/video for 16 hours a day.
I know, but the density of the data is much less in human case.
IOW, humans still learn more effectively with less information, because there are innate mechanisms which process this data continuously and extract new meanings from the same data. This is part of both intelligence and consciousness.
LLMs lack both.
> humans still learn more effectively with less information
> because there are innate mechanisms which process this data continuously and extract new meanings from the same data
To me, these statements strongly contradict each other, but I also really do not care enough to debate it.