The advantages of OpenStack are no licence fees, control over your own hardware and data, and one set of open APIs that behaves the same on three servers as on a million cores. The disadvantages of OpenStack are the people and time it costs: a production cloud is ten or more cooperating services plus a database cluster and a message queue, it needs engineers who know Linux networking and storage, and it must be upgraded at least once a year. Is OpenStack still relevant in 2026? Yes, for the jobs it was built for. The OpenInfra Foundation's latest estimate is that more than 55 million cores run it in production, an all-time high, and the 34th release, Hibiscus, shipped on 30 September 2026.
If the name is new to you: OpenStack is free, open-source software (Apache 2.0 licence) that turns racks of ordinary servers, disks and switches into a self-service cloud. Users start virtual machines, build private networks and attach storage from a dashboard or an API, much as on AWS or Azure, but on hardware that belongs to you or to a provider you pick. Rackspace and NASA started the project in July 2010.
Where OpenStack stands in October 2026:
- More than 55 million cores in production, up from 10 million in 2018. Walmart, Workday and CERN each run 1 million or more.
- A new release every six months. Hibiscus (2026.2) is the newest, and Indri (2027.1) is planned for around 24 March 2027.
- The OpenInfra Foundation, OpenStack's home, became part of the Linux Foundation in June 2025, keeping the Apache 2.0 licence and carrying its governance model over.
- The real costs have not moved: complexity, scarce skills and constant upgrades.
- Our rule of thumb: with one team and fewer than about ten hosts, something simpler wins.
OpenStack advantages and disadvantages at a glance
Most pros and cons lists give the downsides one line. They deserve equal space, because they decide whether a project works out.
| Area | Advantage | Disadvantage |
|---|---|---|
| Cost | No licence fee per core, socket or VM | You pay in salaries, control-plane hardware and often a support contract |
| Control | Your hardware, your building, your rules on where data lives | Every outage, patch and capacity plan is yours as well |
| Scale | The same APIs from a three-node lab to clouds of over a million cores | The control plane is heavy for a small cluster |
| Multi-tenancy | Projects, quotas, roles and isolated networks are built in | Keystone policies and Neutron take real study to get right |
| Flexibility | You pick the hypervisor, storage back end and network design | More choices mean more ways to build it badly |
| Releases | New features every six months and a supported once-a-year upgrade path | Each release is maintained for about 18 months, so upgrading never stops |
| Services | VMs, networks, block, object and file storage, load balancers, DNS, Kubernetes, bare metal | A fraction of the managed services a hyperscaler sells |
| Community | About 600 contributors on the latest release, vendor-neutral governance | The hype is gone: the 2025 European summit drew about 1,200 people, a fraction of the mid-2010s crowds |
What is an OpenStack cloud made of?
OpenStack is a family of services, each with its own REST API, and they call each other through those APIs. Inside each service, the parts talk through a message queue (usually RabbitMQ) and keep their state in a SQL database (usually MariaDB). There are more than 40 official project teams, but a typical cloud runs ten or so services. These are the ones you will meet, with the nearest AWS idea:
| Service | What it does, in plain words | Nearest AWS idea |
|---|---|---|
| Keystone | Logins, projects, roles and the list of service endpoints | IAM |
| Nova | Creates, resizes, moves and deletes virtual machines | EC2 |
| Neutron | Private networks, routers, floating IPs and security groups | VPC |
| Cinder | Block volumes that attach to VMs and outlive them | EBS |
| Glance | The library of images that VMs boot from | AMIs |
| Swift | Object storage over HTTP | S3 |
| Octavia | Load balancers | Elastic Load Balancing |
| Manila | Shared file systems such as NFS and CephFS | EFS |
| Heat | Templates that build a whole stack in one go | CloudFormation |
| Magnum | Kubernetes clusters on demand | EKS |
| Ironic | Physical servers handed out through the same API as VMs | EC2 bare metal instances |
| Horizon, Skyline | Web dashboards; Skyline is the newer one | Management Console |
Asked for the three core components, name Nova for compute, Neutron for networking and Keystone for identity, because nothing runs without them. Glance is just as essential, since a VM needs an image to boot from, and Placement, which Nova uses to keep track of free capacity, is required too. Cinder nearly always comes along for disks that outlive the VM. Many clouds put Cinder, Glance and object storage on one Ceph cluster, as our article on Ceph block and object storage for cloud infrastructure explains, and Octavia gets its own walk-through in OpenStack load balancing as a service with Octavia.
What are the advantages of OpenStack?
Nobody can reprice your hypervisor
OpenStack is Apache 2.0 software with no licence server, no per-core count and no vendor who can change the terms at renewal. That sounded theoretical until Broadcom bought VMware in November 2023 and moved customers from perpetual licences to subscriptions. In its October 2025 summit announcement, the OpenInfra Foundation cited CloudBolt and Wakefield Research figures in which 73% of VMware customers expected price rises of 100% or more. OpenStack's own home is now the Linux Foundation, a move OpenInfra declared complete on 3 June 2025, with the project's governance model carried over. Be clear about what you save, though: the licence fees go, the payroll stays.
One API from a lab to a million cores
The openstack client, the Terraform provider and the Ansible modules talk to a three-node test cloud exactly as they talk to the giants. The foundation's October 2025 Flamingo announcement names Walmart, Workday and CERN as users with 1 million cores or more each, and the 2023 survey report put the messaging company LINE at 6.9 million. The small end is growing too: the same announcement says the biggest growth is among small and medium-sized businesses, and in 2023 about 23% of deployments came from organisations with 10 to 99 employees, the highest share since 2015.
Multi-tenancy, the feature people underrate
This is what separates OpenStack from a hypervisor cluster. Every project gets its own quotas, users, roles and private networks, which may even reuse the IP ranges of the project next door. Hosting providers, universities and platform teams serving many product teams need exactly that; on Proxmox or plain vSphere you bolt it on. It is also why public clouds run on OpenStack: OVHcloud crossed 1 million cores in 2023, STACKIT, KT Cloud and FPT Cloud appear in the foundation's recent announcements, and Deutsche Telekom describes its T Cloud Public as fully OpenStack-based.
You choose every layer
KVM through libvirt is the default hypervisor, and storage and networking are open choices. Cinder drives Ceph as well as commercial arrays (Epoxy refreshed the NetApp, Pure Storage and Hitachi drivers), and Neutron's OVN back end gained BGP support in Gazpacho and an EVPN service plugin in Hibiscus, so tenant networks can be announced straight into a data centre fabric. Already own storage arrays? OpenStack will usually drive them. If not, Ceph on plain servers is the usual answer.
It keeps up with new hardware
The Hibiscus release highlights read like a 2026 shopping list. Nova can launch instances with AMD SEV-SNP or Intel TDX memory encryption, the confidential computing features that stop even the host from reading a guest's memory, and it supports vGPUs managed by Cyborg, the accelerator service. Manila added drivers for the Weka and Lustre file systems used by AI training clusters. That matches the users: in the foundation's October 2025 figures, 30% of User Survey respondents ran AI or machine learning workloads on OpenStack.
Data sovereignty with an address
When the servers sit in a building you chose, the answer to "where is our data?" is a street address, not a region name on someone else's map. About 40% of the contributions to Gazpacho came from Europe, and 60% of the survey respondents running AI on OpenStack are European. Thierry Carrez of the OpenInfra Foundation told The Register that Europe leans heavily on US hyperscalers and wants more capacity of its own.
What are the disadvantages of OpenStack?
None of these is a secret or fatal on its own, but you should be able to name each cost before you start.
It is a distributed system you have to run
A small production cloud runs an API server for each service, schedulers for Nova and Cinder, agents on every compute node, a clustered database, RabbitMQ, a cache and load balancers in front of the APIs, usually spread over three controller nodes so that one can fail. Kolla-Ansible, OpenStack-Ansible, OpenStack-Helm and Sunbeam package all of that and help a great deal, but it stays many moving parts. When a VM refuses to boot, the cause can sit in Nova, Placement, Neutron, Glance, Cinder or the message queue, with the evidence spread over several machines' logs.
The licence bill becomes a payroll bill
Someone has to understand Linux, KVM, OVN or Open vSwitch, VLANs and overlay networks (Geneve with OVN, VXLAN with Open vSwitch), often BGP, usually Ceph, and Python tracebacks, and be on call when any of them misbehaves. Our view: if you cannot name two people who will own the control plane, do not build it yourself. Buy OpenStack as a managed service from a vendor, or choose a smaller platform. For a few dozen VMs, the engineers cost far more than the VMware licences they replace.
Upgrades never stop
OpenStack ships every six months and maintains each release for about a year and a half: Dalmatian, released on 2 October 2024, reached end of life on 29 April 2026. The Skip Level Upgrade Release Process (SLURP) eased this. The first release of each year is a SLURP release, and you can jump from one to the next, say from Epoxy (2025.1) straight to Gazpacho (2026.1), so a yearly upgrade is an official path. Still, the foundation's 2023 survey report named upgrades as a common obstacle that kept many organisations off the latest release, and releases do change behaviour. In Hibiscus, nova-compute and nova-conductor default to native Python threading as eventlet is phased out, and Horizon switched on stricter secure RBAC policy defaults, which can change what a role may do. The schedule on releases.openstack.org in early October 2026:
| Release | Shipped | Upgrade type | Status |
|---|---|---|---|
| 2027.1 Indri | Planned for about 24 March 2027 | SLURP | In development |
| 2026.2 Hibiscus | 30 September 2026 | Regular | Maintained; end of life estimated for 26 April 2028 |
| 2026.1 Gazpacho | 1 April 2026 | SLURP | Maintained |
| 2025.2 Flamingo | 1 October 2025 | Regular | Maintained |
| 2025.1 Epoxy | 2 April 2025 | SLURP | Maintained, due to leave that phase (estimated 2 October 2026) |
| 2024.2 Dalmatian | 2 October 2024 | Regular | End of life since 29 April 2026 |
Services can be retired under you
The Technical Committee prunes dead wood. In February 2024 it removed Murano, the application catalogue, from the 2024.1 release because the project was inactive and had critical bugs, and Hibiscus removed Freezer's old v1 backup API. Build on the core services and you will rarely notice; check a fringe service's activity before you depend on it.
Small clouds pay the full overhead
Three controllers, a database cluster and a message queue cost the same whether they serve four compute nodes or four hundred. Below about ten hosts, the control plane is a large share of the hardware and nearly all of the operating effort. That is why our Proxmox vs OpenStack comparison sends small, single-team setups to Proxmox.
Migration is a project in its own right
Moving VMs off vSphere means converting disks and drivers, rebuilding networks and retraining people. When Epoxy came out, The Register quoted Gartner calling VMware migrations "long, expensive and risky", and Gartner also doubted how many VMware customers really want to leave. Tools such as virt-v2v, the OS-Migrate Ansible collection, Cloudbase's Coriolis and VEXXHOST's MigrateKit help, but run a pilot long before any renewal deadline.
Is OpenStack still relevant in 2026?
By the numbers, yes. In the foundation's 2023 survey report, production cores had grown 350% in five years, from 10 million in 2018 to 45 million. The early analysis of the 2025 survey raised that to more than 55 million. No results from the 2026 survey had been published when we checked in early October 2026, so 55 million is still the latest official figure.
Development is speeding up. The foundation's Hibiscus announcement counts about 600 contributors, 21% more than for Gazpacho, and 11,500 merged changes, 28% more. Since 2010, roughly 10,000 people have contributed nearly a million changes. There have also been 40 security advisories and 13 security notes so far in 2026: proof of a maintained project, and of the patching it asks for.
What has shrunk is attention. At the Paris summit in October 2025, Carrez told The Register the crowd was about 1,200 people, where earlier summits had been "like 10,000" in his words, and that those who still come have concrete reasons: VMware pricing, fear of lock-in and European sovereignty. Its users today are mostly telecoms, governments, research centres, hosting providers and large enterprises, a narrower crowd than in OpenStack's hype years but a busy one.
Why are VMware customers looking at an OpenStack cloud?
Because the bill changed. After the Broadcom deal, perpetual licences gave way to subscriptions and many customers faced much larger renewals. The foundation's VMware migration page says 80% of OpenInfra member companies have had requests to move users from VMware to OpenStack and 60% have completed a migration, though it gives no date or sample size. Gazpacho, the spring 2026 release, was aimed squarely at these migrations, with parallel live migrations in Nova and, per its release highlights, live migration for VMs with a virtual TPM, such as Windows 11 guests, so they can move between hosts while running.
Our take: a VMware renewal is a good reason to look and a poor reason to choose. If the estate is a few hosts run by one team, Proxmox VE is the lighter move in time and people. At hundreds of hosts, several teams and a self-service portal, OpenStack is the stronger landing place. More on the drivers in why teams migrate from VMware to OpenStack.
Who runs OpenStack today?
These names come from the foundation's own announcements and survey reports from 2023 to 2025, plus Deutsche Telekom's own description of its cloud:
- Retail and enterprise IT: Walmart and Workday, both past 1 million cores, plus GEICO and Hyundai.
- Public and sovereign clouds: OVHcloud, STACKIT, Deutsche Telekom's T Cloud Public, KT Cloud, FPT Cloud and UNIQCloud.
- Telecoms: BT Group, with Ericsson among the companies writing code for recent releases.
- Science and research: CERN, past 1 million cores, and ECMWF, the European Centre for Medium-Range Weather Forecasts.
- Internet services: LINE, at 6.9 million cores in the 2023 report.
Two stories from an earlier version of this article need an update. AT&T was for years the showcase telecom user, but on 30 June 2021 it announced that its 5G mobility core would move to Microsoft's Azure for Operators; Microsoft took over AT&T's Network Cloud technology and offered jobs to the engineers behind it. The other is the often-repeated figure that over 85% of OpenStack deployments run Kubernetes. It comes from the foundation's 2022 survey report, which added vanilla Kubernetes (73%) and OpenShift (12%) together. The 2023 report measured something different, Linux, OpenStack and Kubernetes running together, and found that combination in over 70% of deployments.
Should you run OpenStack? A short decision guide
Start from the shape of your organisation; the feature list can wait.
| Your situation | Best fit | Why |
|---|---|---|
| One team, a handful of VMs, nobody whose job is the platform | A few VPS or VDS servers | You need servers; a cloud would be one more system to look after |
| One IT team, 3 to 20 hosts, leaving vSphere | A Proxmox VE cluster | Clustering, live migration, Ceph and backups with far fewer moving parts |
| Several teams or customers who each need quotas, users and their own networks | OpenStack | Multi-tenancy and self-service are built in |
| Hosting provider, telecom operator, university or research centre with hundreds of hosts | OpenStack | At this scale its APIs and economics repay the effort |
| Mostly containers, few or no VMs | Kubernetes on bare metal or VMs | Add OpenStack underneath only if you also need tenants and VMs |
| Spiky demand, worldwide users, no wish to own hardware | A public cloud | You get elasticity and managed services you would otherwise build |
| Data must stay in one country, at scale | OpenStack in-house, or an OpenStack-based European cloud | Local control with standard APIs, so you can still change provider |
| Large VMware estate with a renewal coming | A pilot on OpenStack, and on Proxmox if the estate is small | Test the migration tools on real VMs before you sign anything |
Four questions settle most cases. Do you need tenants, meaning teams or customers who each get their own quotas and networks? Will you have more than about ten hosts within two years? Can you name two people who will own the platform and get time to upgrade it every year? Is your main driver something OpenStack actually fixes, such as licence cost, lock-in or data location? Four yeses: build or buy OpenStack. Three: run a small pilot before you commit. Two or fewer: look at Proxmox, Kubernetes or plain virtual servers first.
Where a VPS fits: OpenStack labs, tooling hosts and the cloud you may not need
RS Computers rents KVM virtual servers, VPS and VDS, with NVMe storage and their own IPv4 and IPv6 addresses, in Amsterdam (Netherlands), Prishtina (Kosovo) and Dublin (Ireland). The plans page has the same plans and prices in all three cities. The vCPUs are shared, VPS plans have a 1 Gb/s port and VDS plans a 10 Gb/s port, and every plan comes with free weekly backups. Around OpenStack, that helps in three ways.
A lab for learning the APIs
DevStack, the community's developer installer, builds a complete single-host OpenStack from source. Its documentation says to run it only on a server or VM dedicated to it, names Ubuntu 24.04 as the most tested system, and by default keeps its networks inside that one host, so a single network interface is enough. Kolla-Ansible's all-in-one quick start asks for two network interfaces, 8 GB of RAM and 40 GB of disk, so on a server with a single interface DevStack is the simpler start. Either way, we would treat 8 GB of RAM as the floor, which means VDS Small (4 vCPU, 8 GB RAM, 240 GB NVMe); we would pick VDS Medium (8 vCPU, 16 GB RAM, 480 GB NVMe) so that a few test instances fit too. Inside a virtual server, those may have to run under plain QEMU emulation (virt_type = qemu in Nova's libvirt settings) instead of KVM: slow, but fine for a tiny CirrOS image. If Linux and networking are the gap, start with our 30-day plan for learning Linux on a VPS.
Tooling and monitoring around a real cloud
Teams running OpenStack on their own hardware still need machines outside it: a bastion and deployment host for Kolla-Ansible or OpenStack-Ansible, a home for the openstack client, Terraform and CI jobs, and a monitor that checks the cloud's APIs and dashboard from another network, so it still raises the alarm when the whole cloud is down. VPS Micro (2 vCPU, 2 GB RAM) is enough for a bastion with the client tools, and VPS Mini (4 vCPU, 4 GB RAM) adds room for monitoring and a CI runner. Put it in a different city from the cloud it watches.
The simpler answer
If the decision guide pointed at "a few servers", that is not a consolation prize. A few KVM virtual servers with backups cover plenty of businesses that once assumed they needed a private cloud. Compare the sizes on the VPS and VDS plans page, or message us on Telegram or email info@rscomputers-ks.com with what you run today, and we will suggest a plan and quote the setup.
Frequently asked questions
Is OpenStack free?
The software is free and open source under the Apache 2.0 licence, with no fee per core, socket or VM. Running it is not free: you pay for hardware, for the engineers who operate it and, if you want one, for a support contract or managed service.
Is OpenStack the same as AWS?
No. AWS is a public cloud run by Amazon, while OpenStack is software for building your own cloud, private or public. The concepts line up, with Nova roughly matching EC2, but OpenStack offers far fewer managed services, and the hardware belongs to you or a provider you choose.
What are the three main components of OpenStack?
There is no official list of three, but the usual answer is Nova (compute), Neutron (networking) and Keystone (identity). A working cloud also needs Glance for the images VMs boot from and Placement, which Nova uses to track resources, while Cinder for block storage and Swift for object storage are optional but common.
Is OpenStack better than VMware?
For large, multi-team or service-provider environments, often yes: no licence fees, built-in multi-tenancy and open APIs. VMware is easier for a small team to run but, since Broadcom's takeover in November 2023, is sold only as a subscription. Small estates leaving VMware usually land more easily on Proxmox VE.
Why do people say OpenStack failed?
In the mid-2010s many vendors pitched OpenStack as a public-cloud rival to AWS; the hyperscalers won that race and many of those vendors left. As private-cloud and provider infrastructure, OpenStack kept growing, from 10 million production cores in 2018 to more than 55 million in the early analysis of the 2025 User Survey.
Does NASA still use OpenStack?
NASA co-founded OpenStack with Rackspace in 2010, but stopped contributing in 2012 and moved much of its cloud work to Amazon Web Services. The latest public record we found of NASA using OpenStack is from November 2021, supporting the environment around the Discover supercomputer at Goddard.
What is the next OpenStack release?
2027.1 Indri, a SLURP release planned for around 24 March 2027, so clouds on Gazpacho (2026.1) can upgrade straight to it. The current release is 2026.2 Hibiscus, which shipped on 30 September 2026.