The way it's written in the annoucement, this will be applicable only to new Cloud orders. I belive the existing cloud infrastructure will be priced the same as started in April 2026.

I suppose the bigger question to ask, is does it have to do anything with AI forcing hardware prices to go up for everyone, or is this related to the 'green energy' transformation in Germany.

But still. Probably one of the cheapest and most reliable cloud providers (Hetzner Cloud)?

Dear ...

This mail is to inform you about an unfortunate, but necessary price change that we must make.

There have been drastic price increases in various areas in the IT branch recently. That is why, unfortunately, we must also increase the prices of our products.

The costs to operate our infrastructure and to buy new hardware have both increased dramatically. Therefore, our price adjustment will affect both your existing products and new orders.

...

* It would be very easy for Cloudflare to implement such a filter. *

What you are stating, the way I understand it, is that in order to implement your original idea (EU browser), one would need the second thing as well, which is force Claudeflare by EU regulations to expose the IP of the server behind a proxy? Maybe it would be easy to implement for Claudflare, but how would you otherwise convince them to do that?

Looks like a pretty big scope creep to me.

1. Do ASN scan of the IP where the DNS entry for that webpage points to

2. Analyze the web resources the page is referring to (much the way urlsca.io does)

If I was implementing that, then with 1. I would probably immediately hit the issue of some/most of the pages being behind a proxy (cloudflare, etc.). With 2. if you had Google Analytics tag on it (and most of the pages do?), then it would show a lot of references to the US.

My point is, that it might be hard to implement not only becasue of whether it makes sense, but also because of: how would you do it?

If you were thinking the way for instance broadcasting companies restrict their content based on where you try to watch a movie from (they only allow certain countries), then I think that's a totally different setup.

Actually I started thinking about the idea you are proposing a lot, but in a more general way. With all the recent development in geopolitics, on whether I can have all the data and technology in EU. The natural move was to verify how much of the solution I already have, ie. host the data itself on Hetzner Cloud. But I think EU is still far behind when it comes to the glue, ie. the software part and the analytical part. Practically every company needs some sort of tracking and most of those solutions that we currently have immediately put you outside of EU.

I am currently experimenting for instance with umami to swap out Google Analytics. They have a solution that you can self-host. But again, this is some effort compared to ready off the shelf GA that 99% of companies probably would use.

After reviewing your updated customer information, we have decided to deactivate your account because of some concerns we have regarding this information. Therefore, we have cancelled all your existing products and orders with us.

Best regards

Your Hetzner Online Team

---

This is the email that I received, after I mistyped my credit card data, when creating an account on Hetzner Cloud. You don't have to be a big cloud company to enforce the KYC rules to defend against fraud.

I wonder how big/small wa pissmail? Would be good to know where is the threshold of sanity:

1. staying with smaller providers is cheaper (and you usually get non-AI customer support)

2. at the risk of stepping into something that makes you lose your data (like you did)

I wonder if anything changed WRT to user management in OVHCloud and how does it compare to other platforms.

I believe what makes it work is the glue, ie. what it integrates with: connect with application logging (log4j), notification with slack.

A few years back I was working with splunk (but this is another galaxy when it comes to cost) and ELK stack. But this was only for logging, not full observability.

If I could turn back the clock a year back, and evaluate the tool you are proposing against what happened, here's how it would look like.

The first thing we had problems with was performance. We moved to smaller Hetzner Cloud machines and split a multi-tenant bare metal systems to fine-grained virtual machines. Having no metrics, meant that we were absolutely blind time-wise. We could log into console and issue 'top', but we couldn't do this after the fact. Decision: self-hosted graphite. I see you have metrics in tracewayapp -> +1.

Now 2 months fast forward. The second problem we had was stability. Because we now moved from a single stable machine to around 200 unstable cloud machines, we had no idea which system is up and which isn't. We did a research of how to outsource uptime. We had online meetings with sales teams of uptime.com and uptimerobot. The initial cost was doing two of those 30-minute sales/fit calls. But that's marginal. The real cost would be something that they price if I remember correctly 1USD/probe/mo. We'd need 200 probes by their definition. The pricing is what killed the deals. Decision: self-host uptimekuma. Initial cost of in-house setup and then just the cost of the smallest hetzner machine which is 2.99EUR/mo. We heavily rely on uptimekuma->slack integration for notifications. I see no uptime tool in tracewayapp -> -1.

Another 3 months have passed. We stopped looking at graphite dashboards on day-to-day basis. Natural human optimization. Systems started going down because running out of disk space, or bugs that exhausted connection pools to the database (twice a month for one of the biggest customers). We quickly realized we need threshold notifications based on metrics. Decision: self-host moira. Heavily rely on slack integration. It's hard to find whether there's something similar in tracewayapp -> -1 (correct me if I'm wrong).

Some few months fast forward. Some of the deployments had a bug that resulted in a flood of exceptions. Even though the system was up (uptimekuma green), some critical services were not working. We did a tricky hibernate upgrade and only found out 4 hours after the deployment that the system is not working. Decision: integrate logging with metrics (graphite) and moira to trigger slack notifications when say #errors > threshold. It's hard to say whether this workflow is easily configurable in tracewayapp -> -1.

Can you elaborate more on the points where I perhaps might have rated your app negatively, but you actually actively support these scenarios and the information is burried somewhere deep in your docs (a few pointers would be helpful if this is the case).

What I like about, what looks like a very complex stack I ended up with, is the fact that it works much the same way as UNIX pipes: I can pretty much change one piece in this flow - ie. I am avoiding vendor lock-in.

This mail is to inform you about an unfortunate, but necessary price change that we must make.

There have been drastic price increases in various areas in the IT branch recently. That is why, unfortunately, we must also increase the prices of our products.

The costs to operate our infrastructure and to buy new hardware have both increased dramatically. Therefore, our price adjustment will affect both your existing products and new orders.

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.

The price changes will take effect on 1 April 2026 for both new orders and existing products. For orders placed before 1 April 2026, but delivered after 1 April 2026, the adjusted prices will apply.

The following existing products, listed under the customer number specified in the subject line, are affected by the price increase as follows.

...

We know that price changes are always a challenge for everyone. However, we believe that our products' prices and conditions still have a competitive price-to-performance ratio.

We have prepared a list of all new prices for you on Hetzner Docs at https://docs.hetzner.com/general/infrastructure-and-availabi.... Starting on 1 April 2026, the new prices will also be on our website.

If you do not wish to continue your contract, you can cancel the products within the regular cancellation periods via your administration interface.

We hope that you will be able to understand and accept our decision, and that we are sorry that we have to change our prices.

If you have any questions, we will be happy to help. Please log onto your account and go to "Support" in the menu and create a support request.

Kind regards,

Hetzner Online

Our solutions (from simplest to the most sophisticated).

Use a different datacenter, for instance instead of Falkenstein, use Helsinki.

If that does not work, then instead of using web GUI, use either terraform or hcloud in console. Usually what was shaded in the GUI, worked through console.

If that does not work, try to loop terraform or hcloud in the console with a 5 minutes sleep time in between retries (mind hetzner api rate limit). Usually we could get up to four machines overnight. What we discovered (might not be true any more) is usually around 4-5AM european time, some machine types were going back to the pool.

But the real problem is IMHO the pricing change Hetzner release around 6 months back. This is my guessing here, so do not hold me accountable. All those cheap servers are now called 'Cost Optimized' and are mostly not available. The new group is called 'Regular Performance' - they are twice that expensive as 'Cost Optimized'. Also, notice that Hetzner has around 2 months ago released an official email to all customers that they have to raise their pricing, claiming they cannot keep up with the cost of labour/hardware to keep it running.

If you try to connect the dots, it looks like they just want to force people to move to the 'Regular Performance' group (2x cost) over time, to balance out the earnings.

To comment directly on your 'cpx11' problem. This is 'Regular Performance' group, so you are already in Hetzner's 'target'. I believe if you use one of the techniques I gave you above, you should be able to acquire that machine.

Where I live and work people get 'attribution' from the managerial staff by showing more spending on anthropic graph. Our founder had a hard time convincing people just to use claude or copilot. He fought for 6 months. The adoption is only happening. He measures it by looking at how much did each team member spent in claude console. I asked my other colleague at a different company. He said that one of the developers got recognized by the managers, because he spent 200USD last month on claude tokens - that's more than anybody else in the team.

I have to admit that in my company for some of the people it is really disappointing - the technology itself. We start seeing that what is being promissed by the marketing is not delivered. I think we may get to the point when the AI bubble starts popping, before the technology is widely adopted.

Having said that, from my subjective point of view, I wouldn't incorporate this cost breakdown technique in my company for a very simple reason: it's not worth it. It has a negative ROI at this point for us. But maybe it's different in other companies.

I dropped it quickly in favour of claude in the terminal for various reasons: - that github's copilot license of 19USD/seat/mo was kind of doggy? It had some limit on the tokens after which they'd start to charge per token usage? - it didn't ask me for permission to run certain commands - security risk

I wonder if with the link you've sent anything changed with that regard. I believe I'm still staying with claude, but maybe it would be interesting to know if github's tools are improving.

Story 1

When I started going into avalanche terrain, I quickly realized that I need to get some professional training before something happens. I went to a 2 day course. The lecture for 8 hours in a classroom, even though was done by a really good professional, was hard to understand. Especially when the lecturer introduced something called 'professional method of assesing avalanche risk'. Nobody understood. But at the evening, we went to the bar, ordered a few beers and the lecturer pulled out something called 'Snow Safety Cards' (https://www.snowsafety.nl/wp-content/uploads/2015/11/Snow-Safety-Cards-Additional-Cards-v7.3-EN.pdf). And even though we were getting more drunk with each beer, we started to understand that method.

Story 2

I joined a new IT project. One of the veterans in our team convinced the founder that it would be great to integrate the team that just grew x3 and meet at a company 'Christmas Party'. We had an official dinner, and then we went to the pub. I pulled out these cards after one round (or two): https://punkx.org/unix-pipe-game/ext-0.1/. We started playing. To my surprise, none of the 15+ programmers knew what is a unix pipeline! Even the veteran. But people started learning it on the go. I don't have to tell you that I wasn't the one winning the game (probably because it is not balanced and it wasn't created with that in mind - mostly to teach kids). It was so interesting to see that it was the same story as with avalanche cards. People had no idea what the subject is, they learned easily on the go even with their frontal cortex numbed.

Even though these happened a few years back, I still keep thinking. What is the phenomenon of games that help you go into a pretty abstract/technical field, even when you are in the no-teaching mode? It kind of makes me think that the scientists that study dolphins say that they play 50% of the time, opposed to humans. I wonder if anyone tried to make a dolphin drunk and study how well they acquire knowledge whilst playing.

Comments URL: https://news.ycombinator.com/item?id=48250814

Points: 6

# Comments: 0