New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
Ja, viel zu teuer.
As someone said above they are actively engaging into a more high-end market, a market populated by heavyweights, I'm not so sure they'll be able to win a fight against such powerful opponents.
How do you like it ? Which location, package ? Any YABS ?
This wasn't on our 2026 bingo card.
If anyone is looking for a new home for their workloads, we're happy to help with the move - mevspace.com
I shall be waiting for your new sales thread
Can anyone running a DC and selling dedicated servers comment on whether these price rises are to be expected across the board? I've got some servers I don't need but I'm nervous to let them go given they're on relatively cheap pricing.
Eg I would consider this my worst value auction server and I don't really need it currently but I'm worried I won't be able to get something with same NVME for similar price:
Xeon Gold 5412u
128gb RAM
3x 3.84TB NVME
190 euros/mo
It only applies to new orders for now.
My question is more whether we can expect similar price rises across the industry for new services, so I know whether to hold older services I have but could let go. Hetzner's statement seems to suggest that they expect the industry to align to their prices. Personally I don't see RAM and NVME prices coming down in the short to medium term, hopefully it's also not going up much more either, but are the kind of prices we are seeing from hertzner now really what is needed to cover current hardware prices?
Keep it if I were you.
to replace my hetzner cx23 i got the leaseweb 1 x 4 vCPU - 6 GB vRAM - 100 GB NVMe Disk 2 years contract 2 years billing for 2.79€ / month
are there similar offers from big providers?
YABS or didn't happened
Leaseweb is a big provider tbh
Yes besides leaseweb
I have two piko netcup but this leaseweb is even more convenient comparing performance
Bring a $99/year dedi in APAC please
Absolutely keep a server if it’s at a good price. It’s only expected to get worse from here. Everything is through the roof: electricity, real estate, space availability, and of course hardware, which is up 500% with little to no availability, even if you’re willing to accept an absurd price increase and wait six months or more for delivery from manufacturers.
32 GB DDR5 ECC memory is now $1,850 per chip, and most vendors have only a few in stock at any given time.
For everyone saying AI demand is a bubble, ask yourself how much AI has already integrated into the internet ecosystem. Now add medical, vehicle, and robotics control to the equation, and in 5–10 years, data center and compute infrastructure will continue to be among the most in-demand sectors the world has ever seen. Compute is a commodity, just like oil or anything else.
Will new manufacturing capacity come online and alleviate some of this crunch? Sure. But demand will continue, and in many cases, it will outpace manufacturing capacity for the foreseeable future. Compute has always greatly benefited from scale. AI inference will absolutely drive cloud demand to dizzying new heights.
Utilities are refusing Data Center interconnects due to grid constraints, water rights are being heavily fought over, industrial electricians are in high demand and most new data center space is focused solely on hyper scale.
People are saying there's an AI bubble not because AI isn't providing value. It's because none of the AI majors are profitable yet. Some of them don't even have a path to profitability. Yet more and more money is being dumped on it, with cost driven up because of high demand. A whole bunch of these smaller companies, and perhaps even some of the majors, will start BK-ing at one point. This will ease pressure on the market.
The other thing that will likely happen is that when companies will have to start paying the real price for AI, they will be much more restrained and selective with its usage. This will also cool demand. Token maxing is already on its return because of how much it costs, and that's while AI is being offered at like a 90% (or higher) discount right now.
As for Hetzner, I doubt this is just a cost-related increase. I think they also simple want to make more money and aligned their pricing with the big boys (Akamai, DO, Vultr, etc.). And they're free to do so. But personally, I've never seen Hetzner as an A-tier provider, so I will not be spending A-tier money on them.
Yet every single Internet behemoth started in a arguably non-profitable way, Google, Facebook, etc. and all found ways to mega profits by absorbing an important industries market share. AI will automate everything, whats the value on that?
Sure, there will be an equilibrium to be found in service cost, but the demand will not go away.
Cost of operations of this industry have in every aspect increased, it will be passed on to the consumer if any provider wants to stay in business. All we're saying is, if you have a well priced service that works for you now, hold on to it.
The question isn’t if AI will have demand, its if investors will find the level of demand (and in time, profitability) enough to find decent ROI in the huge investments being made.
That AI will be huge is priced in, everyone knows this.
The market is expecting x level of adoption across segments. This is priced in. Will adoption be faster than this? Then bull run continues.
Will adoption be really fast, yet slower than expected? This is where markets may get rough.
To understand which one we’re heading for one doesn’t only have to form an opinion on where we’re going but also consider exactly what the market expects given current CAPEX and that’s what’s impossible for any of us to do, hence, the dicussion is quite dumb and pointless. Any attempt at doing this is just guessing.
usually, i like it very big
not this time
Comes close when I ordered though
Why I'm not even surprised it's you who somehow managed to get the best deals?
Left nothing for the rest of us 😠
As you said, the demand is real. AI is also a commodity — people have moved from capability benchmarks to token pricing. So what shapes token pricing? The cost of building datacenters, electricity, and hardware, divided by how many tokens the hardware stack can produce.
NVIDIA's roadmap shows token generation capability increasing by an order of magnitude within two quarters. That doesn't include LPU/XPU solutions (Groq, Cerebras) that increase production further.
If the denominator increases, the only way the price holds is if the numerator rises proportionally — meaning GPU prices have to compensate. With current BoM, even a 2.5× materials cost increase doesn't offset a 10× jump in token throughput. So token prices will go lower.
The bubble crowd implies AI labs massively subsidize their product. They do — but subsidy now doesn't mean they can't be profitable later. Midjourney is profitable on Y2025. When token costs drop, the gap between cost and price closes, and labs move into margin.
But here's what the bubble crowd misframes: the risk isn't in the labs. It's in the infrastructure layer. When token prices drop, previous-generation hardware becomes economically indefensible — not gradually, abruptly. The right amortization window for a Blackwell cluster isn't 36 months, it's closer to 12. The Anthropic-xAI Colossus contract already shows why: H100/H200 clusters at implied ~$3.65/Mtoken cost, against Vera Rubin's projected floor of $0.02–0.05/Mtoken. That's 73–180× more expensive per token. It's not depreciation — it's obsolescence within the financing window.
If an AI lab walks away — or simply refuses to renew on older hardware — the operator absorbs the loss. The operator effectively subsidized the lab's operation, funded by whatever profits they had or debt they could raise. When the cluster value drops, the debt face value doesn't follow. The credit-default-swap does — in the other direction.
You haven't really been around during .com, have you? Sure starting out operating at a loss sometimes leads to very profitable projects but it's far from a guarantee and in this case it's not even that much about the companies themselves being able to operate economically but if the product they sell manages to provide an economical advantage to their users. The AI support agent is regularly only useful as long as it costs less than the human it replaces. Sure, there's a whole lot of different angles but in a large part the big question comes down to: Will the final product, which highly likely will be x-times as expensive as it is now, still be attractive to potential buyers?
Currently probably mostly how much loss the seller is willing to take for subsidizing it.
Did you use AI to write this ?
In hindsight it's kinda obvious with all the long dashes, formating and how annoying it is to read it.
For grammatical and data augmentation purpose.. 90-95% of words could be traced to earlier version I wrote. Around 2-3% is a phrasing correction while the rest is the augmentation.
If this site logs changes to drafts, the site administrator and moderator could attest my statement.
Any question?
Wrote into notebook
It's not augmentation, it's entropy expansion.
You could figure out the key content using the words in bold.
Those quantitative data are necessary as examples.
You do what you gotta do.
You're absolutely right!