Cutting cloud costs and locking down sensitive AI workloads don’t usually show up in the same product release. Red Hat just did both anyway with OpenShift 4.22, which became generally available on July 14.
The update leans hard into security first. A new minimal Red Hat Universal Base Image strips out non-essential packages, shrinking the attack surface teams have to defend. Alongside that, OpenShift sandboxed containers 1.12 brings confidential containers on bare metal to general availability, giving organizations a way to run workloads with stronger isolation guarantees.
The more interesting addition is confidential AI, currently shipping as a technology preview. It lets organizations run sensitive workloads and proprietary algorithms inside a cryptographically isolated slice of memory and CPU, so the data stays protected even while the workload is actively running. For companies handling proprietary models or regulated data, that’s the kind of feature that closes a real gap, not just a checkbox one.
Virtualization gets attention too. A new Ethernet VPN integration connects containerized and virtualized workloads to outside infrastructure, while volume groups now support multi-volume snapshots for VMs, giving teams a crash-consistent backup option that should simplify disaster recovery. Red Hat also introduced two-node OpenShift with fencing, aimed squarely at edge environments where resources are tight and reliability still matters.
On the cost side, the Red Hat build of Karpenter, an open source auto-scaler, is now generally available for OpenShift Service on AWS with hosted control planes. Teams running that setup can also tap AWS EC2 Spot Instances for fault-tolerant workloads, a straightforward way to cut compute spend for jobs that can tolerate interruption.
The release closes with the JobSet operator, built to coordinate large-scale distributed training runs and LLM fine-tuning jobs as a single unit. For teams burning through expensive GPU time, that kind of coordination matters more than it sounds like on paper, since poorly orchestrated jobs waste compute fast.
Taken together, the release reads less like a single headline feature and more like Red Hat chipping away at several pain points at once: security hardening, cost control, and the operational mess that comes with running both VMs and AI workloads on the same platform.
