Proxmox VE arm64

[Tushar Tyagi]

Hi Team, Congrats on the new launch!

I have been doing deep research around ARM-native cloud infrastructure and open-source virtualization ecosystems.

I wanted to understand Proxmox’s long-term vision around ARM64 support, especially in the context of:

• hyperscale public cloud infrastructure
• AI inference workloads
• Kubernetes-native environments
• sovereign cloud initiatives
• edge cloud deployments
• high-density low-power compute clusters

From an architectural perspective, Proxmox + ARM appears strategically very interesting compared to traditional VMware/x86-heavy stacks, especially considering power efficiency and future AI infrastructure trends.

I would love to understand the following:

• whether native ARM64 support is being actively explored internally
• what technical limitations currently exist
• whether there are roadmap discussions around enterprise ARM support
• and how the Proxmox team sees ARM adoption evolving over the next 5–10 years.

Would genuinely love to hear your thoughts.

 

[t.lamprecht]

whether native ARM64 support is being actively explored internally

Yes.

what technical limitations currently exist

From our side no actual blockers for server/enterprise HW.

whether there are roadmap discussions around enterprise ARM support

Yes.

and how the Proxmox team sees ARM adoption evolving over the next 5–10 years

If the server enterprise hardware on the horizon materializes: quite positive.

 

[Tushar Tyagi]

Thomas, thank you for bringing clarity to this .I genuinely appreciate it.
I’m working as a Cloud Solution Architect, and the deeper I explore Proxmox, ARM, and open infrastructure, the more questions and ideas I end up with around scalability, cloud adoption, AI workloads, and future possibilities.
Would it be possible for you to guide me on how I can take this discussion further with the Proxmox team? If feasible, I’d truly value an opportunity for a short virtual discussion with someone from the team to better understand the direction and possibilities around the ecosystem.

 

[abufrejoval]

I explored an ARM fork made by a Chinese developer some time ago (version 8 based).

My interest was at the opposite end, putting a set of services I wanted to run in a super resilient HCI manner on a three node HCI cluster with Ceph using Raspberries, instead of fully passive Atom machines I'd been using for earlier concept trials.

The idea was really to have a bunch of cheap boxes wired into a cluster to provide resilience to a single node failure for things like identity management for Linux and Windows clients using Univention and perhaps a small NextCloud instance.

It worked pretty well, even if live-migration between a Raspberry PI5 (8GB) and an Orange PI 5+ (32GB) failed for lack of abstracted ARM CPU types e.g. "ARM64-V2-AES" or rather "ARM64-V[n]-[dialect]".

Because Proxmox is so modest in terms of resources and overhead due to the simplicity of its design vs. thinks like oVirt, XenServer or VMware, I really see it as a champion of bottom-up grass-roots resilient infras that are as cloud independent as they can be.

These days hardware prices are crazy no matter what, but then I find myself having replaced Android phones with 16GB of RAM, a USB3 port and way more CPU power than most of these ARM SBC. Some aren't even looking that bad running KDE Plasma on an external 4k screen!

And they even come with a UPS included...

So anyway, I believe the Chinese fork never made it to V9 and might have died down from lack of support.

So if Proxmox could ensure that they don't aim for datacenter ARM chips exclusively, that would be very appreciated, indeed.

That's not likely going to deliver direct revenues, but whoever made Proxmox work for personal use, is a much better Proxmox engineer on the job: just saying!

 

[Tushar Tyagi]

Thanks for showing your interest. The cloud market is already huge. I was mainly looking at niche areas where ARM and Proxmox naturally complement each other well.
Your point about keeping ARM support open beyond only datacenter-grade hardware makes complete sense.