Walk into the server room of a Nepali bank, ISP or telco and there’s a good chance the virtual machines are running on Proxmox. That’s what I see in most of the infrastructure I’m asked to look at in Nepal, from small branch-office clusters to the platforms ISPs use to sell VPS.
Proxmox earned that place. It’s free to install, it runs on the servers you already own, it doesn’t need a dollar payment approved by the bank, and any good Linux engineer can learn it in a week. When Broadcom ended perpetual VMware licences in February 2024 and moved everyone to per-core subscriptions, Proxmox was the obvious way out.
But “obvious” and “right” are different questions. Many of the clusters I see were built quickly, by one person, to get off VMware or to avoid buying it. They work. The question is whether they’ll keep working when a disk fails at month-end, when the engineer who built them leaves, or when an auditor asks how long recovery takes.
This article is for the people who own those clusters. It lists what to check, explains which answers should worry you, and ends with a 10-question score you can take right now.
Key takeaways
- Most Nepali organisations chose Proxmox for convenience: no licence fee, no dollar approval, existing servers, existing Linux skills.
- Proxmox VE 8 reached end of life in August 2026. Clusters not on version 9 no longer get security fixes.
- Proxmox itself says the free update channel is not recommended for production, and its Ceph, quorum and HA guidance asks for more than many quick builds provide.
- Banks now have to report downtime and recovery limits to Nepal Rastra Bank, so untested restores and failover are a regulatory problem as well as a technical one.
- Proxmox is the right platform for most Nepali estates. It just has to be designed and run as production.
Why did Nepal choose Proxmox?
Most teams I talk to chose Proxmox for four reasons, and all four are about convenience rather than design.
| Reason | Why it mattered in Nepal |
|---|---|
| It’s free to run | Proxmox VE is open source. Paid subscriptions only add the tested update channel and vendor support, from €120 to €1,100 per CPU socket per year. |
| No dollar approval | VMware is now sold per core in US dollars. In Nepal, IT prepaid cards are capped at $3,000 a year, and any single foreign payment above $30,000 needs Nepal Rastra Bank approval. |
| It runs on what you have | It installs on the servers already in the rack, and Ceph storage and backups come in the box. |
| Your team already knows it | It’s Debian Linux underneath. Anyone who has run a Linux server can be productive in days. |
None of these reasons is wrong. But none of them says anything about how the cluster behaves when something breaks, and that’s what the rest of this article is about.
Eight questions your cluster should be able to answer
Each question below comes from Proxmox’s own documentation or from Nepali regulation. If you can’t answer one with confidence, that’s where your risk is.
Which version are you on?
Proxmox VE 8 reached end of life in August 2026. Version 7 ended in July 2024. If your hosts haven’t moved to version 9, they no longer get security fixes. Major upgrades are supported, but Proxmox says they need planning, testing and a fresh backup first, which is exactly why many clusters get left behind.
Where do your updates come from?
Proxmox has two update channels. The free “no-subscription” channel is, in Proxmox’s words, “not recommended to use this on production servers” because its packages are less tested. The enterprise channel is the one it calls suitable for production. In my experience, many production clusters in Nepal run on the free channel, often because nobody decided otherwise.
Was Ceph built the way Ceph needs?
Built-in Ceph storage is one of Proxmox’s best features and one of its easiest to get wrong. Proxmox recommends at least three identical nodes, around 12 disks across them, at least 10 Gbps of network used only for Ceph, separate networks for storage and cluster traffic, plain HBA controllers instead of RAID cards, and about 8 GB of memory per disk.
A Ceph cluster on 1 Gbps links and RAID cards will run fine on a quiet day. It’s during a rebuild after a failed disk, under real load, that it slows everything down.
What happens when one node dies?
A Proxmox cluster needs a majority of votes to keep working. If it loses that majority, it goes read-only. A two-node cluster can’t keep a majority after a failure unless a small third “tie-breaker” machine (a QDevice) is added. Cluster traffic also needs latency under 5 ms, ideally on its own network.
Does high availability actually work?
Proxmox HA needs at least three nodes, shared storage, and a watchdog that switches off a failed node before its VMs start elsewhere. Without a hardware watchdog it falls back to a less reliable software one. Failover takes about two minutes, and the remaining nodes must have enough spare capacity to carry the extra load.
If you’ve never pulled the power on a node to test it, you don’t know whether HA works.
When did you last restore a backup?
Proxmox Backup Server can verify backups automatically, and Proxmox recommends re-verifying all of them at least monthly because disks degrade over time. A verified backup is good; a restored one is proof. Banks have to show their recovery times to Nepal Rastra Bank, so “we have backups” is no longer enough.
How many people really know the cluster?
In many organisations one engineer built the cluster and holds the knowledge. If that person leaves, the configuration, the workarounds and the root passwords leave with them. Ask whether someone else could rebuild a failed node from your documentation alone.
Are you using Proxmox as a cloud?
Proxmox has users, roles, permissions and resource pools. It doesn’t document quotas, self-service projects or billing. That’s fine for running your own VMs. If you’re an ISP or hosting provider selling VMs to customers, those gaps get filled with add-ons and scripts, and that’s usually the point where a cloud platform like OpenStack starts to make more sense.
Where is each sector exposed?
The same eight questions weigh differently depending on who you are.
| Sector | The question that matters most | Why |
|---|---|---|
| Banks and financial institutions | Restores, failover and documented recovery times | Nepal Rastra Bank’s 2026 risk guidelines require a 24/7 security operations centre, a cyber incident response team, and maximum downtime and recovery times for digital services, reported to the central bank. Its IT guidelines already require a business continuity plan. |
| ISPs and telcos | Using Proxmox as a cloud, and Ceph under load | Subscriber-facing services such as DNS, RADIUS and routing functions can’t wait two minutes for failover, and selling VMs to customers needs quotas, isolation and billing that Proxmox doesn’t provide on its own. |
| Government and its suppliers | Version, updates and people | Since February 2025, data centres and cloud providers in Nepal must register with the Department of Information Technology, hold security certificates and pass annual audits. Data centres holding government data need at least a Tier 3 rating. |
Proxmox can meet every one of these requirements. It doesn’t do so by default; someone has to design and run it that way.
When is Proxmox the right choice?
For most Nepali organisations running their own workloads, Proxmox is the right platform. It just needs to be run like production.
| Situation | What fits |
|---|---|
| Your own VMs on 3 to a few dozen hosts, one IT team | Proxmox, on version 9, on the enterprise channel, with tested HA and restores |
| Proxmox already running but built in a hurry | Keep Proxmox; fix the design (Ceph network, quorum, watchdog, backups) and write it down |
| Selling VMs or VPS to customers who need self-service, quotas and billing | A cloud platform such as OpenStack, often alongside Proxmox for internal systems |
| Many sites or teams that each need their own isolated environment | OpenStack, or Proxmox with a management layer designed for it |
| Still on VMware with a renewal coming | Proxmox for most estates, moved one VM at a time with VMware kept bootable until sign-off |
The difference between a cluster that was chosen and a cluster that was designed is usually a few weeks of work: the right network, the right update channel, a tested failover, a restored backup and a runbook someone else can follow.
Self-score · 10 questions · no sign-up
Score your Proxmox cluster
Answer for your main production cluster. If you’re not sure, answer no.
All hosts run Proxmox VE 9.Version 8 lost security fixes in August 2026.
Production hosts update from the enterprise channel.Not the free no-subscription repository.
Ceph or shared storage has at least 10 Gbps used for nothing else.
Cluster traffic (corosync) runs on its own network.
The cluster has at least three votes.Three nodes, or two nodes plus a QDevice.
In the last 12 months you powered off a node and watched HA move its VMs.
Backups are verified automatically at least monthly.
In the last 6 months you restored a full production VM from backup.
Someone other than its builder could rebuild a node from your documentation.
If you sell VMs, quotas, isolation and billing come from a platform, not scripts.Answer yes if you don’t sell VMs.
Your result and the area to fix first appear here once every question is answered.
Frequently asked questions
Is Proxmox good enough for a bank in Nepal?
Yes, if it’s designed and run as production: a supported version, the enterprise update channel, at least three nodes, tested failover and tested restores. The platform isn’t the risk; an undocumented, untested build is.
Is it safe to stay on Proxmox VE 8?
No. Proxmox VE 8 reached end of life in August 2026, so it no longer receives security fixes. Plan the upgrade to version 9 with a backup and a test run first.
Do I need a Proxmox subscription?
For production, the subscription gives you the tested enterprise update channel and, from the Basic tier up, vendor support tickets. Prices run from €120 to €1,100 per CPU socket per year, which is small next to per-core VMware pricing.
Can an ISP sell VPS on Proxmox?
Many do. Proxmox has permissions and resource pools but no documented quotas, self-service projects or billing, so those come from add-ons or scripts. At scale, a cloud platform such as OpenStack handles them natively.
Should we move from VMware to Proxmox?
For most Nepali estates, yes: it removes per-core dollar licensing. Move one VM at a time, keep VMware bootable until you’ve signed off, and design the Proxmox side properly from day one.
If your score left you with questions, book 30 minutes with me. Bring your answers; I’ll tell you plainly what I’d fix first, even if the answer is “nothing”. Talk to an engineer.
Sources
- Proxmox VE FAQ: support lifecycle
- Proxmox VE: Package Repositories
- Proxmox VE: Deploy Hyper-Converged Ceph Cluster
- Proxmox VE: Cluster Manager
- Proxmox VE: High Availability
- Proxmox VE: User Management
- Proxmox Backup Server: Maintenance
- HorizonIQ: Proxmox subscription pricing (July 2026)
- ThinkOn: Broadcom VMware pricing changes (March 2026)
- B360: NRB foreign currency access for the ICT sector (April 2026)
- Ratopati: NRB raises foreign payment limits (October 2026)
- Pradhan Law: Data Center and Cloud Service Directive 2081
- ShareSansar: NRB IT Guidelines for banks
- NEPSE Trading: NRB Risk Management Guidelines 2026