OpenStack Hibiscus Targets AI Workloads with Confidential Computing

Hibiscus arrives as organizations grapple with the infrastructure requirements of the AI era, specifically the need to manage heterogeneous hardware while maintaining strict workload isolation. The release integrates advanced confidential computing features, including AMD SEV-SNP and Intel TDX support within the Nova compute service, to provide hardware-level memory encryption. These protections extend to accelerator management, where the Cyborg service now enforces tighter access controls for multi-tenant environments.

Performance for data-intensive tasks sees significant upgrades. The Manila project now includes support for Weka and Lustre parallel filesystems, directly addressing the storage bottlenecks common in machine learning pipelines. Furthermore, networking efficiency has been overhauled; the Neutron service features a 15x reduction in memory footprint for HA router monitoring, allowing operators to scale operations without proportional increases in overhead.

Development velocity for this cycle reached 11,500 code changes, supported by 1.6 million automated jobs run via the Zuul CI system. While Hibiscus introduces these technical advancements, it remains a non-SLURP release, meaning operators currently running the Gazpacho version can bypass this update and wait for the upcoming Indri release in March 2027.