- Red Hat CEO Matt Hicks argued enterprise AI success depends on building on an open foundation across hybrid environments
- The company is trying to repeat its successes with Linux and Kubernetes with a new platform for the AI era
- Analyst Steven Dickens said there are good reasons for enterprises to seek out open AI solutions
RED HAT SUMMIT, ATLANTA – Red Hat secured its place as the open-source foundational layer for two major tech movements – Linux in the web era and Kubernetes for the cloud. Now, it’s trying to do it again for AI.
“The organizations that will have the most success in this environment, they are not necessarily going to be the ones that spend the most, or the ones that move the fastest. They will be the ones that build on the right foundation,” Red Hat CEO Matt Hicks said in a keynote address. “Red Hat will bring AI to the enterprise to make that an open standard.”
The company this morning unveiled Red Hat AI 3.4, a major upgrade to its two-year-old platform that aims to make it the place to scale and manage AI models and agents across hybrid cloud environments. Among other things, the 3.4 release includes a Model-as-a-Service (MaaS) offering to allow enterprises to tap into curated models, track performance and enforce policy. It also includes prompt management, agent tracing and automated safety testing.
The company also updated its Ansible Automation Platform with a new automation orchestrator and Model Context Protocol (MCP) server.
Take those two things together and the message is clear.
“Red Hat is making a deliberate play to own the operating layer for enterprise AI,” HyperFrame Research CEO Steven Dickens told Fierce. “The governed, open substrate that sits beneath every model, every agent, every inference pipeline. If they execute, they become the default fabric for operationalizing AI across hybrid estates.”
The merits of open source
Hicks indicated that what Red Hat is bringing to market is largely based on its own experience implementing generative AI across the company. He noted that while Red Hat initially deployed frontier models, it quickly assess whether and where those could be replaced by open weight models built on the open vLLM inference library.
Its initial count of 10 agents has now ballooned to 200 and 85% of the calls running through its Deep Research agent system go to open-source, open-weight models running on Red Hat infrastructure.
“Of course, this was more efficient but our results actually got better,” Hicks said. “When you own the model and you own the infrastructure, you can optimize in a way that a frontier model that is a general purpose could never anticipate.”
Hicks spent a lot of time talking up the merits of open source, particularly models like Nvidia’s Nemotron and IBM Granite. And CTO Chris Wright argued that only open source can offer the right level of flexibility needed to keep up with and adapt to a rapidly moving environment in the AI era.
Asked for his take on the argument for an open-source foundation, Dickens told Fierce that there is a case to be made, though for reasons beyond those Red Hat gave.
“The proprietary model vendors are building brilliantly capable black boxes that no enterprise compliance team can fully audit. Open source does not just solve for cost or flexibility; it solves for trust,” he explained. “When you are deploying autonomous agents that can execute transactions, modify infrastructure, and make decisions without human approval, you had better be able to inspect every layer of that stack.”
The takeaway? While frontier models may continue to generate headlines, Dickens predicted it will be open-source models generate the most purchase orders.