Red Hat launches unified AI Enterprise hybrid cloud stack

Red Hat has launched Red Hat AI Enterprise, a unified platform for deploying and managing AI models, agents and applications across hybrid cloud environments, alongside updates branded as Red Hat AI 3.3.

The new platform spans Red Hat's existing AI products, including Red Hat AI Inference Server, Red Hat OpenShift AI and Red Hat Enterprise Linux AI. Red Hat describes the combined offering as a "metal-to-agent" stack that links infrastructure, model operations and agent deployment under a single approach for organisations running AI workloads across datacentres and public cloud services.

Many organisations have struggled to move beyond limited AI pilots because of inconsistent tooling and fragmented infrastructure. Red Hat is positioning AI Enterprise as an effort to standardise AI delivery and governance in ways that resemble established enterprise software operations.

What it includes

Red Hat AI Enterprise brings together several parts of the AI lifecycle, including inference, model tuning, and agent deployment and management. It is built around Red Hat OpenShift, the company's Kubernetes-based platform for running and managing applications across hybrid cloud estates.

The platform supports a range of models and hardware choices across different environments, and includes observability and lifecycle management focused on governance and operational control of AI workloads.

For inference, Red Hat cited use of the vLLM inference engine and the llm-d distributed inference framework, which it says improve deployment of generative AI models across mixed hardware setups.

Nvidia partnership

Alongside the platform launch, Red Hat said it has co-engineered a new Red Hat AI Factory with Nvidia. The package combines Red Hat AI Enterprise with Nvidia AI Enterprise for organisations using Nvidia's AI infrastructure.

The announcement comes amid rising interest in agent-based AI workflows, which can involve multiple model-driven tasks running with less direct human interaction than chat-style systems. Red Hat aims to make that shift more manageable for IT teams already using OpenShift and Red Hat Enterprise Linux in production.

Red Hat AI 3.3

Red Hat AI 3.3 introduces updates across the portfolio, including changes to model access, hardware support and operational tooling. Red Hat said the release expands the set of validated and compressed models available via the OpenShift AI Catalog, including versions of Mistral-Large-3, Nemotron-Nano and Apertus-8B-Instruct.

Red Hat also said the release enables deployment of models such as Ministral 3 and DeepSeek-V3.2 with sparse attention. Other updates include multimodal changes, a "3x Whisper speedup", geospatial support, improved EAGLE speculative decoding and enhanced tool calling aimed at agent workflows.

A technology preview in the release covers a Models-as-a-Service approach. Red Hat said this would let IT teams provide self-service access to privately hosted models through an API gateway, with centralised management of access and consumption.

Hardware options

Red Hat is broadening hardware support in Red Hat AI 3.3, including a technology preview for running generative AI inference on CPUs, starting with Intel processors. It has also expanded hardware certification for Nvidia's Blackwell Ultra and added support for AMD MI325X accelerators.

Red Hat has added a Red Hat AI Python Index, described as a trusted repository for "hardened" versions of tools including Docling, SDG Hub and Training Hub, aimed at improving repeatability and security for AI development pipelines.

For observability and safety, Red Hat pointed to telemetry across AI workloads, llm-d deployments and Models-as-a-Service clusters. It also referenced a technology preview of integrated NeMo Guardrails for operational safety controls.

Red Hat AI 3.3 also introduces features it describes as internal GPU-as-a-Service. Red Hat said organisations can pool GPU resources through orchestration and add automatic checkpointing for long-running training jobs to preserve state when jobs are interrupted.

"For AI to deliver true business value, it must be operationalised as a core component of the enterprise software stack, not as a standalone silo. Red Hat AI Enterprise is designed to bridge the gap between infrastructure and innovation by providing a unified metal to agent platform. By integrating advanced tuning and agentic capabilities with the industry-leading foundation of Red Hat Enterprise Linux and Red Hat OpenShift, we are providing the complete stack - from the GPU-accelerated hardware to the models and agents that drive business logic. Additionally, with Red Hat AI 3.3 organisations can move beyond fragmented pilots to governed, repeatable and high-performance AI operations across the hybrid cloud.", said Joe Fernandes, Vice President and General Manager, AI Business Unit, Red Hat.

Red Hat said Red Hat AI Enterprise and Red Hat AI 3.3 are aimed at organisations that want a consistent way to deploy and operate AI across hybrid cloud environments, while managing model access, infrastructure choices and governance requirements.