M
Ask HN
Ask HN: Cloud providers are losing in favor of bare-metal?
Lately, I’ve noticed a new trend on X: Devs (and indie hackers in particular) are ditching cloud providers and jumping straight to bare-metal servers like Hetzner.

Honestly, I think the big cloud companies just haven’t kept up. Their services feel clunky compared to the standalone alternatives. Just try comparing Vercel’s dev experience to Amplify’s, and you’ll see what I mean. On top of that, AWS has gotten way stingier with startup credits.

Put those two together, and it’s no surprise fewer people are hosting their MVPs on AWS. It’s tough to stay under $150/month with a database and a server, while on bare metal you can grab 16 GB RAM for around $20/month.

- Do you think the cloud is actually losing ground? - And for those using bare-metal: how do you handle DB backups, CI/CD, and pulling logs? - Would you scale something using bare-metal servers?

[Carlos](https://github.com/clostao)

Given that AWS, Azure and GCP are all recording 20-40% YoY growth, no, I don’t think they’re losing ground.

As for startup credits, they’re still handing out $100-200k like candy if they deem you a serious startup. There was a lot of abuse in the past so they started putting up filters.

I don’t think so. Hetzner sees some growth but they are still very small.

DB can have replicas and upload dumps to some S3 compatible object stores. Like in the cloud. CI/CD with stuff like gitlab and Argo CD is mostly the same. You can install some log monitoring stuff like the classic Elk stack. Or be old fashioned and use ssh and journalctl.

I wouldn’t attempt to automatically scale to the moon on bare metal, that won’t happen. But a few beefy servers running k3s or similar can get you pretty far.

Interesting that you call hetzner “bare metal”. Most of their product offerings are cloud based VMs (and certainly for $20 a month you are getting shared hardware VMs). Do you think more devs are are actually buying hardware/servers or do you just mean devs are using VMs provided by smaller companies instead of the hyper scalers, when you say “bare metal”.
True, more than bare-metal I meant dedicated servers on-premise
Again that’s a confusing line. Hetzner does not provide (as its main business) “on premise” options as id use that term.

When you get dedicated servers from them, typically it’s VMs on their machines, in their data centers and the hyper scalers have options just like that.

Er, sorry to interject... incorrect: https://www.hetzner.com/dedicated-rootserver/

Not interested in arguing a pointless distinction like 'main business'. They apparently do everything, even colo. Want a third of a rack or full? If 'retail' metal is too pricey, try their 'server auction' with older equipment.

I just want to know what the questioner means is all. I view on-prem as _not_ meaning in the providers data center.

I view bare metal as meaning not virtualized.

Another related but not identical dimension is dedicated or shared.

Hetzner offers options in all those dimensions, but not in the price the question mentions.

> Hetzner offers options in all those dimensions, but not in the price the question mentions.

Again, incorrect (sorry). $30/m as opposed to $20. Their mistake, I suppose. /s Call it a typo and move on. Rhetorical: how that confuses the question for you, confuses me.

Before we bemoan 'a 50% cost increase', consider: this does quadruple the memory. An additional 48GB on the dedicated non-virtualized box. See: https://www.hetzner.com/sb/

IMO, the $10 on top of $20 is a pittance. Beyond the memory, bonus points: a spare SSD. A small price to reduce administrivia and gain redundancy!

Anyway, pedantry aside, all of this is the question: what does your value judgement look like? You're clearly somewhat familiar with the offerings. Hopefully more-so, now.

While I'm rambling/have the floor, some tangents:

    1. it hasn't made a lot of sense/cents to offer a dedicated *physical* machine with 16GB of memory in quite some time. Now, Cloud? Absolutely. Simplifies Tetris. Physical, though? A knock-off Tamagotchi might make more money after considering power, IPs, and module density for the DDR generation/platform.

    2. I can see reason for hesitation with Hetzner. While I'm a happy customer, they're well-known for being *very* thorough with identification during on-boarding. To the degree that many are uncomfortable.
When someone says “bare metal” to me, it implies that virtualization is a problem for their workloads. I’d be very surprised if there was a mass increase in workloads that was the case for. I’d be surprised if most of Hetzners existing non-virtualization customers even care.

When someone says “on premise” I assume they mean they are running some part of the premise. Whether they are leasing rack space in a broader data center or not, I agree that it’s a continuum, but I’d expect on premise to include negotiating energy and network as part.

To me, the more interesting question for this conversation is a)how are you paying for compute, are you doing long term lease/purchase of servers or are you getting short term provisioning of a standard sku and b) are you using “value added” services. Are you provisioning your own data warehouses, specialized database clusters etc, or are you using a managed service?

The distinction between a vm on a hyper scalar and per hour provisioning of hetzner compute (virtualized or not) seems pretty straightforward and I’d be shocked if there was some mass migration of those workloads away from the big clouds.

I have done workloads where virtualization very much matters and the physical locality of the servers makes a big difference and we are seeing the opposite happen, the cloud providers are dipping their toe in that market which has been closed to them.

I think a lot of people would consider having a rack at a colo to be "on-prem". I usually thought of that term meaning you're renting space and the services provided are redundant/reliable power, a public internet drop (or multiple), and physical security. Where you need to rack it yourself, run your own networking gear, etc.

I don't see how that's too functionally different from having a suite at a commercial office building, save for the fact the power and data isn't going to be nearly as reliable. It's just your "suite" is only a bit more than 19" wide.

  • gkoz
  • ·
  • 12 hours ago
  • ·
  • [ - ]
When you get dedicated servers from them that's exactly what you get: you're renting physical servers, which you manage yourself. And they've been in this business for decades. The VMs offering is much more recent.
They have a cloud product called dedicated as well. I’m not sure what the questioner is arming about. I shouldn’t have said it’s their main business though.
It's just the meme of the year. There's nothing that changed really for quite a long time. Some cloud services provide things on a simplistic and easy way, some in complex and complete ecosystem. The same choice and the tradeoffs have always been there.

The only change is that it's popular to write a "how we saved $$$" blog posts. Which actually could be read as "how we failed to do proper analysis and kept losing $$$ for years".

  • jll29
  • ·
  • 8 hours ago
  • ·
  • [ - ]
Not necessarily "failure": for a startup, it makes sense to prioritize speed of development over cost optimization. Since you don't know just how fast your MVP will grow, remaining elastic early on is a plus, and sticking with hyperscalers gives you that (you never know, you may land on the front page of Slashdow and HN at the same time..).

For solo indie developers, in contrast, it makes more sense to seek out the cost optimal setup straight away.

I totally get the tradeoffs, but that's not what I'm talking about. For example the popular Hey article was 10 years into their operations and they were spending multiple millions with AWS. That's not a startup anymore, and they don't need to optimise for unexpected 10x growth. At some point for new companies it transforms from "good decision" to "total cost is comparable" to "we should notice cloud doesn't make sense now" to "we're years late and failed at planning".
99% of all “startups” will never grow past the need of one VM and one database - that includes all of the dumb “AI startups” that YC is funding.
I would say it depends on the effort to cost savings ratio. I moved my most expensive servers there because that’s where it made the most sense. If it totals considerable amounts of money it is absolutely worth the extra effort, but it is not worth it if you only have a handful of t4g.mediums.
If you are using either Amplify or Vercel, you’re a really small developer that isn’t moving the needle.
[dead]
This would be quite funny since _I’ve been using bare metal the whole time_. But I don’t think it’s as simple as that.
dhh spent 3 years moving from AWS to bare mental to save $2M and in the process completely missed the AI wave. Opportunity cost is real.