Lightbits as Proxmox storage: what the new plugin actually delivers

Proxmox announced a storage partnership on 29 October 2026: Lightbits Labs became an official solution provider, with a native plugin for Proxmox VE. A storage system with its own plugin that handles guest images, live migration and HA without extra configuration had not existed in the Proxmox environment before. The press release names as the target group virtualization environments that want to leave a competitor's proprietary licensing model. Everyone reads VMware there. Judging by Lightbits' price list, the intent is real: the integration list includes a plugin for VMware vCenter.

In the vendor's layout, the storage cluster runs as its own appliance next to the Proxmox nodes and is reached over NVMe over Fabrics. From the Proxmox side, that removes the hand-work an iSCSI target used to require in every configuration, wired into each node by hand.

The plugin is a connection, not built-in storage

The press release calls it a "natively integrated option". Native here means: Lightbits writes the plugin itself, and it shows up in the Proxmox interface like any other storage type. Proxmox VE 9 does not include it. You install it as a separate package on every node, it registers a storage ID, and from there it presents the inventory of the external system. According to the vendor, hand-wired LVM and iSCSI setups fall away. Anyone who runs a Proxmox upgrade will have to update the plugin along with it, because compatibility with the next major release is an open question for a plugin this young.

The repository behind it has been public since June 2026, carries version 1.0.4 since 26 October, and stood at fifteen stars on 4 November 2026. The license, as written in the LICENSE file, is Apache 2.0. Support runs through the project's GitHub page, best-effort, without any commitment. The vendor states only NVMe over Fabrics over TCP as a tested transport. Anyone who wants to lean on RDMA networks has to measure that themselves. A second maturity detail: between the first public commit in June and Proxmox's official statement at the end of October lay a little over four months.

The operations table in the repository covers what a virtualization operation needs daily: create and delete storage, grow volumes, snapshot and roll back, live migration, HA, capacity reporting in the interface. Lightbits promises mirrored copies and thin provisioning for the blocks. The architecture behind it stays simple: the Proxmox nodes talk REST API and NVMe-oF to the storage cluster, two TCP port groups, no sidecar services on the nodes.

The price is an entry figure, not an offer

The price list starts at 17.25 US dollars per terabyte per month, on own hardware in your rack and on cloud storage at the same rate, measured on 4 November 2026. One arithmetic example over the vendor's tariff table, without a contract and without our terms: 10 terabytes of usable capacity come to roughly 2,070 US dollars a year. The page names no support ceiling, no response time, no minimum buy. That entry figure is a signal downward against the list price of classic storage arrays, nothing more.

A second detail from the repository belongs in every purchasing decision: TRIM passthrough for thin provisioning is disabled by default and requires at least storage version 3.16. Once switched on, it cannot be switched off again. Anyone planning to run thin provisioning on the system has to know the arithmetic before the migration, and not merely afterwards.

Resilience class	Usable capacity after protection	From one raw block of 12 drives at 3.84 TB	Extra hardware
Hyperconverged, three copies (our DRBD setup)	15.36 TB	12 drives in the nodes, no extra box	0 %
Mirroring storage cluster, two nodes	23.04 TB	12 drives plus dedicated servers	33 %
Storage cluster with RAID 6 protection, two nodes	23.04 TB	12 drives plus dedicated servers	33 %
Ceph, replication factor 3	15.36 TB	12 drives in the nodes, no extra box	0 %

Unit: terabytes (TB) at 10^12 bytes. Subject: one raw block of 12 NVMe drives at 3.84 TB each, 46 TB gross, as it appears in a mid-size cluster. Date of the calculation: 4 November 2026. Source: our own arithmetic over the four protection classes, not a vendor figure.

This table is the reason we watch the topic instead of rebuilding our setup. The capacity saving every shared storage pitches does not exist automatically: the triple-replication class drops capacity back to the nodes' level, two mirroring storage nodes mathematically gain 33 percent, exactly like RAID 6 on the same drives. Savings start only when a protection class with two copies instead of three is enough and the drives sit centrally instead of in each node.

What we do differently, and why

Since summer the vendor has been running its own Proxmox page and comparison guides aimed at Ceph. The image shows the 2024 wordmark. (Quelle: Wikimedia Commons, public domain)

In our environment, guest storage runs hyperconverged over DRBD and LINSTOR on the nodes, two sites, re-evaluated on 31 July 2026 and left that way on purpose. The pull of shared storage for us is not capacity but decoupling: with DRBD, compute and disk always grow as one package, and both worlds share the same maintenance windows. A storage cluster next door, updated independently of the virtualization nodes, is exactly the split our build does not have. The price is a second product world with its own update cycle and support mirror, which we got rid of in virtualization when VMware left. Add one dependency that DRBD does not carry: when the storage cluster goes down, every Proxmox node stops, no matter how many there are.

Lightbits is not new in the category of vendors selling a storage array with dedicated support, by the way. NetApp published its own Proxmox plugin back in 2020. What is new in the current announcement is that Proxmox itself walks this road through a partner program, with a system that entered the price list at the lowest starting rate we have read in this category.

For the technical background, two neighbouring posts on this channel help: the article on the TrueNAS plugin for Proxmox describes how an NVMe/TCP defect in Linux reads, on the very same transport. And the storage and backup hub holds our complete DRBD, Ceph and Vitastor series. Anyone weighing an own storage array against a subscription model will find the business side of that question on our IT services page.

Four questions to answer before any trial

Does our network match the transport? The vendor names NVMe over Fabrics over TCP as the only tested path, plain Ethernet without RDMA. That is simple, but then latency lives in the network, not in the NVMe drive.
Which protection class is meant contractually? Replication or RAID decide the usable capacity. An offer should state the protection scheme in numbers instead of only terabytes.
Who updates what when the next Proxmox release lands? The plugin comes from the vendor, the upgrade from us. That chain deserves a dated answer.
What is the way back? Moving from shared storage back to node-attached storage is a weekend of migrating every block. Before the trial, know whether the return path exists at all.

A plugin from a specialized storage vendor that serves live migration, snapshots and HA without special configuration is the most grown-up step Proxmox has taken on external storage. Whether it earns its keep in a given environment is rarely decided by the technology. The real question is whether compute and storage maintenance windows should be separate at all. We evaluated that split on 31 July 2026 and chose to keep the windows together.

Addendum 4 November 2026: the vendor's comparison guide

The day after we published, Lightbits' own storage-architecture guide for Proxmox appeared, pitting Ceph against NVMe/TCP systems. Its core argument: integrated Ceph runs on the same machines as the guests and eats their CPU and memory. Broadly documented, but not our configuration – our replicated storage runs on dedicated nodes. The guide sells a separation we already built. Its headline numbers are lab figures with no test conditions, and the real study sits behind an email form. That NVMe/TCP was invented at Lightbits reads differently at the standards body: the protocol came out of a working group Lightbits helped develop. The guide is right on one point – running Ceph on the guest hosts means paying that tax every day.