HN Debrief

Hetzner Price Adjustment

  • Infrastructure
  • Cloud
  • AI
  • Startups
  • Hardware

Hetzner, a budget-focused European hosting provider, updated its pricing for cloud servers and dedicated machines starting June 15. Existing instances mostly keep their old price unless rescaled, but new purchases in many tiers are far more expensive. The biggest jumps are in cloud and DDR5-heavy offerings, with US cloud plans getting hit especially hard. That matters because Hetzner has long been the default answer for founders, indie operators, and self-hosters who wanted something much cheaper than AWS, GCP, or Azure without going fully DIY.

If your product, hobby service, or startup depends on cheap commodity hosting, treat that assumption as broken and model much higher replacement costs now. Lock in what you have, revisit architecture and provider concentration, and expect smaller hosts and even some consumer hardware segments to keep repricing upward.

Discussion mood

Alarmed and pessimistic. Most comments framed the increase as both a real hardware supply shock and a bad sign for the future of cheap hosting, with extra frustration that AI demand is driving up costs for everyone else while delivering unclear benefits to ordinary users and small operators.

Key insights

  1. 01

    HBM is crowding out ordinary DRAM

    The shortage is not just "memory got expensive". AI systems are pulling wafer starts and packaging capacity toward high-bandwidth memory, or HBM, which begins from the same general DRAM production pipeline but earns far more money per bit. That means manufacturers have every incentive to feed AI accelerator demand first, leaving less ordinary server memory for hosts like Hetzner and making a quick return to old pricing unlikely even if AI spending cools a bit.

    If your infra plans depend on cheap DDR5 boxes coming back soon, do not bank on it. Budget around sustained memory scarcity and prioritize architectures that squeeze more life out of DDR4-era or already-provisioned machines.

      Attribution:
    • benjiro29 #1
    • icyfox #1
    • zettabomb #1
  2. 02

    This is partly a pricing strategy shift

    Some long-time customers did the math and argued the new monthly rates imply much faster hardware payback than before, especially when paired with lower setup fees on some lines. That makes the increase look like more than a raw pass-through of component costs. It looks like Hetzner is using a real shortage to rebalance toward higher recurring revenue and customers who tolerate enterprise-like pricing.

    Do not read provider price changes as neutral reflections of component costs. Re-run vendor comparisons from scratch, because a host that used to be the obvious low-cost choice may now be optimizing for a different customer base.

      Attribution:
    • mschuster91 #1
    • benjiro29 #1
    • pocksuppet #1
  3. 03

    Hyperscalers have not repriced the same way yet

    AWS, GCP, and other large clouds look stable partly because they started from much fatter margins and are still amortizing older hardware generations bought before the current squeeze. Some of their fleets are old enough that near-term replacement cost has not fully hit customer pricing. A few signs of movement are showing up anyway, like new instance generations costing more and cloud providers raising adjacent fees such as egress.

    Do not assume hyperscaler prices are immune or that current quotes reflect the new hardware reality. Watch new instance families, storage, and networking charges, because the increase may arrive there first rather than as a blunt across-the-board compute hike.

      Attribution:
    • Shank #1
    • tjwebbnorfolk #1
    • justincormack #1
  4. 04

    Grandfathering saves incumbents, not new builders

    The immediate damage falls hardest on anyone launching something new, resizing an existing deployment, or replacing failed hardware. Existing Hetzner users can often sit on older pricing, but side projects, indie game servers, and bootstrap experiments now face a much worse starting line. That creates a quiet moat around already-running services while making the next wave of small entrants harder to finance.

    Separate your infra planning into "keep running" and "build new" scenarios. Your existing estate may be fine for a while, but expansion, recovery, and greenfield projects need a very different cost model.

      Attribution:
    • civvv #1
    • littlecranky67 #1
    • AltruisticGapHN #1
  5. 05

    Old hardware becomes strategically valuable again

    Hetzner’s auction servers and used business PCs kept coming up because already-written-off hardware avoids today’s acquisition price shock. That changes the tradeoff. A four- or ten-year-old machine can be economically superior if your workload is memory-heavy, mostly idle, or tolerant of older CPUs. The constraint is no longer elegance. It is replacement cost.

    Audit workloads by RAM need, latency sensitivity, and business value. For non-latency-critical jobs, older dedicated servers, refurb desktops, or colo with used gear may now beat shiny new cloud instances by a wide margin.

      Attribution:
    • ralferoo #1
    • svelle #1
    • amoshebb #1
  6. 06

    Serverless only helps when usage is truly spiky

    A few people pointed to Cloudflare Workers and other pay-per-use models as an escape hatch, but the more grounded comments narrowed that claim fast. If your app is bursty and can fit the platform, serverless can dodge idle VM spend. If you run steady-state compute, batch jobs, databases, or bandwidth-heavy services, it does not solve the underlying economics and may just move you into a different premium pricing trap.

    Use this shock to classify workloads, not to declare one universal migration. Move only the genuinely intermittent pieces to serverless and keep the always-on or data-heavy parts on infrastructure you can cost out clearly.

      Attribution:
    • dllrr #1
    • pocksuppet #1
    • noodlesUK #1
  7. 07

    German hosting KYC is a real adoption tax

    A separate but high-signal complaint was Hetzner’s identity verification flow. Several people said passport or ID upload requirements were enough to push them to OVH or DigitalOcean before price even entered the picture. One comment noted Germany imposes know-your-customer style rules on hosting and telecom services, which explains the requirement more than any company-specific weirdness. Still, for privacy-conscious users, it is a real friction cost that compounds the pricing hit.

    When evaluating non-US hosting, include onboarding and compliance friction in the decision. A cheaper host is not actually cheaper if account creation, identity policies, or regional rules block adoption in your org.

      Attribution:
    • tengwar2 #1
    • nozzlegear #1
    • jpalomaki #1

Against the grain

  1. 01

    Existing customers are still getting the old deal

    The bleakest takes flatten an important distinction. Most current Hetzner customers are grandfathered at prior prices unless they rescale or order new capacity. For anyone who already migrated and sized correctly, the economics are still much better than they would have been on AWS-class providers, so the savings from earlier moves remain real.

    If you already run on Hetzner, avoid panic migrations. First check which instances are actually affected, then preserve your current footprint and postpone resizing until you have a deliberate alternative.

      Attribution:
    • pocksuppet #1
    • ramblurr #1
    • omnimus #1
  2. 02

    Shortages can attract real new supply

    Not everyone bought the idea of permanent hyperscaler domination. Some argued that if AI demand stays durable, suppliers and new entrants will chase the margin. Large buyers may secure better contracts, but they also help justify fab expansion and more hardware investment. The lag is painful, but the market does have a route to eventually relieve scarcity.

    Plan for bad prices in the next few years, not forever. Keep optionality in your architecture so you can benefit if component markets loosen instead of hard-coding today’s worst-case costs into long-term commitments.

      Attribution:
    • Gasp0de #1
    • Dylan16807 #1
    • whatever1 #1
  3. 03

    Hetzner may still be cheap enough

    A few commenters pushed back on the idea that Hetzner is now pointless. Depending on the product line and the alternative, it can still beat providers like Scaleway, OVH, or hyperscalers on price-performance. The bigger change is that it is no longer an absurd outlier. It has moved from "obviously cheapest" to "must be compared carefully."

    Do not assume the answer is to leave Hetzner. Benchmark the exact instance types, regions, storage, and bandwidth you need before migrating, because switching costs can erase a small nominal savings elsewhere.

      Attribution:
    • wiether #1
    • larodi #1
    • speedgoose #1

In plain english

AWS ↩
Amazon Web Services, Amazon’s cloud computing platform.
Azure ↩
Microsoft Azure, Microsoft’s cloud computing platform.
DDR5 ↩
Double Data Rate 5, the current mainstream generation of standard computer memory with higher speed and capacity than DDR4.
DRAM ↩
Dynamic Random-Access Memory, the main working memory used in servers, PCs, and many other computing devices.
GCP ↩
Google Cloud Platform, Google’s cloud computing platform.
HBM ↩
High Bandwidth Memory, a type of very fast memory used on advanced accelerators and datacenter GPUs.
hyperscaler ↩
A very large cloud computing provider such as Amazon Web Services, Microsoft Azure, or Google Cloud that operates massive data centers at global scale.
NAND ↩
A type of flash memory used in solid-state drives and other storage devices.
Serverless ↩
A cloud execution model where you deploy code or services without managing servers directly and typically pay based on usage.
SSD ↩
Solid-state drive, a storage device based on flash memory that is much faster than a mechanical hard disk for many workloads.
VM ↩
Virtual machine, an isolated software-based computer used to run code separately from your main system.

Reference links

Primary pricing references

Hardware market and supply context

Alternatives and migration options

Provider documentation and alternatives