When OpenStack, the open-source Infrastructure-as-a-Service (IaaS) cloud, first appeared in October 2010, it offered two services: compute (Nova) and object storage (Swift). That was it. That was all. And that was enough. Over the years, it evolved into a full-powered IaaS for both private and public clouds. 

Now, with the release of OpenStack 2026.2, Hibiscus, it continues to be the cloud of choice for telecoms, but with this new version, which focuses on confidential computing, improved Domain Name System (DNS), faster high-performance storage, and efficiency improvements for very large deployments, it is also AI-ready. With rising interest in Digital Sovereignty, it's also an excellent choice for companies that value privacy and control over their clouds. 

Let's look closer. Starting with  Nova's expanded confidential-computing support. Hibiscus can now run confidential computing workloads for AMD SEV-SNP and Intel TDX on these popular chips. Hibiscus now enables hardware-backed memory encryption, stronger workload isolation, and better virtual machine (VM)  security.

Once upon a time, OpenStack was a great choice for private clouds. It still is, but these days it's also a top pick for organizations running regulated data, proprietary models, or other workloads where even cloud administrators and privileged infrastructure software should not be able to inspect guest memory. OpenStack frames the change as part of a broader “trusted infrastructure” push, alongside DNS and accelerator-security work.

Designate, OpenStack’s DNS service, gains TLSA/DANE support. TLSA/DANE is an Internet security protocol that links TLS certificates to domain names using secure DNS records, rather than relying solely on traditional Certificate Authorities.  Designate also offers stronger cross-tenant isolation and improved authentication. It also adds tooling to help operators identify quantum-vulnerable certificates and outdated TLS configurations as they prepare for post-quantum cryptography.

Hibiscus also targets practical AI infrastructure requirements. Nova and Cyborg can now manage mediated devices. Cyborg provides a REST API and management framework for accelerator resources, such as GPUs, vGPUs, and FPGAs.  Operators define accelerator capabilities and requirements, while Cyborg tracks, reserves, attaches, and releases the underlying hardware for workloads. 

What does all that mean in English? In an AI cloud, Cyborg can help ensure that a tenant requesting a vGPU gets an appropriate accelerator attached to the instance, rather than leaving operators to manage GPU devices manually on each compute host.

The new and improved Manila, OpenStack's File Storage as a Service, now includes drivers for the Weka and Lustre parallel file systems. These are very popular for AI and High-Performance Computing (HPC) workloads. Ironic and Nova also improve bare-metal resource tracking for when you need the best possible performance for demanding workloads.

As Thierry Carrez, the OpenInfra Foundation general manager, explained, “AI is raising the requirements on the infrastructure that organizations need to run. They need to be able to provide the specialized compute these workloads demand, while maintaining stronger control over how that infrastructure is operated, secured and trusted.”

For operators, several less glamorous changes may be just as important. Neutron, OpenStack's network connectivity as a service, reduces the memory footprint of router monitoring by more than 15 times. That's a serious saving for clouds running large numbers of highly available routers.

Glance, a service where users can upload and discover data assets that are meant to be used with other services, including images and metadata definitions, improves image distribution with parallel imports, preferred-store downloads, and expanded cache management. This works hand-in-glove with OpenStack Watcher, which is a resource optimization service for multi-tenant OpenStack-based clouds. This latest model helps improve workload consolidation by considering both live usage and reserved capacity, while adding automation options for reclaiming unused resources.

Hibiscus is a non-SLURP release. What the heck is that? Well, unless you're a hardcore OpenStack user, I can guarantee it's not any of the things you're thinking of.   OpenStack’s SLURP stands for Skip Level Upgrade Release Process. What that means is if you're running the previous SLURP release, 2026.1 Gazpacho, you can skip Hibiscus and upgrade directly to the next SLURP release, 2027.1 “Indri,” expected in March 2027.

For OpenStack users, all this means Hibiscus is less a radical architectural reset than a targeted upgrade. It brings hardware-rooted workload isolation and GPU-oriented management closer to the platform’s core, while trimming the operational overhead of running a cloud at scale or AI on top of your very own cloud. 

Most of what shapes AI next year is sitting in a research paper today. Keeping up with them is a full-time job, and the abstracts don't help.

TLDR AI is the newsletter that does the translating. Each morning you get the handful of papers worth knowing about, curated by Anthropic and ex-Google engineers, with a plain-English summary of what they found and why it matters.

You come away smarter about AI research in the time it takes to finish your coffee.

Free, and delivered to 1.1M+ inboxes every morning.

Subscribe for free.