Red Hat Expands Hybrid AI Platform for Agents, Inference and Cost Control
- Cargo
- Machine Learning
- Shipped
- Shipper
- Devony Hof
- Readers
- 866
Red Hat AI 3.4, Ansible 2.7 and new hardware and cloud collaborations target agent deployment, controlled automation and lower inference costs.

Red Hat’s latest enterprise AI push is aimed at a practical problem: getting agentic systems and inference into production across mixed infrastructure while keeping spending and operations manageable. At Red Hat Summit 2026 in May, the company outlined platform updates, automation tools and collaborations intended to help organizations run AI workloads where their data and systems already reside.
The product centerpiece is Red Hat AI 3.4, an updated enterprise platform designed to support agent deployments and large-scale inference across hybrid environments. Red Hat’s argument is that organizations need one way to deploy and manage agents across that infrastructure as demand for inference grows. “What’s really going to drive inference demand exponentially is AI agents,” said Joe Fernandes, Red Hat’s vice president and general manager of AI.
Two managed services extend that approach onto IBM Cloud: Red Hat AI Inference and Red Hat OpenShift Virtualization Service. The first is designed to put real-time inference into production workflows. The second supports migrating and running virtual machines at scale. Together, they offer an option for organizations seeking to add AI and modernize virtualized workloads within a hybrid cloud setup.
Connecting AI insights to controlled actions
Agents become operationally consequential when they can change infrastructure. Red Hat’s Ansible Automation Platform 2.7 is intended to connect AI-generated insights with infrastructure actions under established controls. Its self-service automation lets authorized staff at places such as branch offices and factory floors manage servers and networks without bypassing those controls.
Sathish Balakrishnan, vice president and general manager of Red Hat’s Ansible business unit, said customers can use their own data to calculate savings and efficiencies. He described a shared control path for actions initiated by people, events or agents, including access controls, approval gates, audit trails, content signing and credential management. The design matters because it makes an agent’s action subject to the same operational checks as other automation, rather than giving it a separate route into infrastructure.
Red Hat also framed the platform work as a response to the gap between AI use and production deployment. TheCUBE Research projects AI cloud spending could approach $2.5 trillion by 2026, growing at roughly 44% year over year. The same research reports that 50.7% of organizations rely on public AI tools such as ChatGPT and Copilot, while 20.2% report enterprise-wide AI deployments built on a governed framework.
Rob Strechay, principal analyst at theCUBE, described many organizations as “token consumers” still trying to move beyond pilots. Red Hat’s pitch is that customers can use OpenShift infrastructure and tools such as vLLM to produce tokens more efficiently, helping lower costs and improve the return from AI applications. The practical question is whether those infrastructure choices fit each workload; platform capacity alone won’t turn experimentation into measurable value.
Matching compute to inference workloads
Red Hat’s work with AMD and Intel addresses that cost question through hardware selection. AMD and Red Hat are encouraging customers to match compute resources to workload needs instead of defaulting to the newest or largest systems. AMD executive John Hampton said some enterprises bought large GPU clusters to address AI demand, then found the expense difficult to sustain.
Red Hat and Intel are likewise emphasizing a mix of CPUs and GPUs for inference. Intel’s Bill Pearson said customers can get better results at a lower price point by assessing the CPUs already in their data centers and balancing them with GPUs to meet the outcome they need. The approach turns inference economics into a workload-matching exercise: expensive accelerators may be appropriate for some tasks, while existing CPU capacity may handle others.
Red Hat also announced Red Hat AI Factory with Nvidia. The collaboration includes support for Nvidia’s Blackwell architecture and upcoming Vera Rubin platform, as well as participation in Nvidia’s OpenShell project for AI agent sandboxing and secure execution. Red Hat CTO Chris Wright described the goal as providing a consistent hybrid cloud foundation and more architectural control for scaling agentic AI.
Extending the ecosystem to edge and sovereign deployments
Other collaborations target production deployment and edge computing. Boomi and Red Hat are working on an integrated stack for deploying agentic AI at scale. Work with Panasonic Connect covers edge computing, while the collaboration with Voyager focuses on space edge micro data centers. These efforts point to uses beyond centralized cloud deployments, where systems may need to run close to operational equipment or in remote environments.
Red Hat is also expanding support for sovereign cloud while allowing customers to keep infrastructure they have already invested in. Its work with Core42, which specializes in sovereign-by-design infrastructure, and with European collaborator Sopra Steria addresses organizations seeking greater control over data and infrastructure. Gartner forecasts that the sovereign cloud infrastructure-as-a-service market will grow at a compound annual rate of 36% through 2028.
Security is another part of the deployment story. Fedora Hummingbird Linux is a container-native operating system designed for agent-first development, with an automated software delivery pipeline. Red Hat Hardened Images offers minimal container images intended to reduce vulnerabilities and strengthen software supply-chain security. These tools address parts of the software environment around agents; organizations still need to decide what permissions agents receive and how their actions are reviewed.
Red Hat president and CEO Matt Hicks said organizations need to connect AI capabilities to specific business needs, with domain knowledge helping teams identify where to begin. That’s a useful frame for a broad product strategy: the value of agent infrastructure depends on choosing tasks where automation, data access and control requirements can be made concrete. Red Hat’s announcements provide more pieces for that effort, while the harder test remains whether customers can deploy them at a sustainable cost and demonstrate results in production.