Red Hat, Nvidia tighten integration with AI Factory
Red Hat and Nvidia knit together a software stack, called AI Factory, meant to get AI in enterprise production faster. Red Hat also rolled out a new AI Enterprise product bundle.
Red Hat and Nvidia said this week they've begun doing more of the heavy lifting to integrate their software products before they ship, resulting in a new full-stack AI Factory platform designed to push enterprise AI into production.
Previously, Red Hat and Nvidia collaborated on open-source projects, and partners, including hardware vendors such as Dell, integrated their products to deliver AI infrastructure for enterprises such as Northrop Grumman. Now, the two companies are working together directly from an earlier stage of product development to create a turnkey system called Red Hat AI Factory with Nvidia.
"Previously … there was a little bit of a time delay for hardware enablement on new [hardware] architectures, because we had to work through this open source process," said Justin Boitano, vice president of enterprise AI at Nvidia. "We've closed that gap so we can support new hardware together at the time it comes to market, which lets enterprises buy the latest infrastructure [and] the full software stack from both companies and get support from both of us."
Red Hat AI Factory with Nvidia comes with pre-configured AI models, including IBM Granite, Nvidia Nemotron and Cosmos open models, delivered as Nvidia NIM microservices, along with Nvidia NeMo AI agent lifecycle software.
It also builds in a set of Red Hat AI products newly collected under the name Red Hat AI Enterprise: Red Hat AI Inference Server, Red Hat OpenShift AI and Red Hat Enterprise Linux AI.
Other pre-integrated components include:
AI observability and security tools that Red Hat acquired with Chatterbox Labs in December, integrated with the Nvidia DOCA framework that runs applications on BlueField DPUs and SuperNICs
AI inference utilities such as Red Hat vLLM, Nvidia TensorRT-LLM and Nvidia Dynamo.
This combined tool set produces AI Factory features such as automatic GPU orchestration using pooled infrastructure and checkpointing to protect long-running jobs, the companies said.
The upside for organizations is ease of use and automation … plus pricing pressure at the hardware level.
Rob Strechay,Analyst, TheCube Research
The Red Hat AI Factory with Nvidia faces competition from other large IT vendors with their own AI factory products, including Broadcom, HPE and all the major cloud providers. But it adds another strong choice for enterprises, said Rob Strechay, an analyst with TheCube Research and Smuget Consulting.
"The upside for organizations is ease of use and automation from a skill set perspective, if you use Kubernetes already and Red Hat Kubernetes in particular, plus pricing pressure at the hardware level if you get all the automation and secret sauce from Red Hat," he said. "It is a real challenger and solves a number of things Nvidia doesn't have that will put it on par with Azure Stack from Microsoft."
Enterprise AI production gap persists
All AI Factory platforms are up against the challenge of a stubborn gap between enterprise AI aspirations and production use: a Feb. 4 Process Optimization Report by data processing firm Celonis SE found that while only 19% of 1,649 surveyed businesses use multi-agent systems today, 85% aspire to be "agentic enterprises" within two to three years. Operational readiness was cited as the biggest blocker to enterprise AI adoption in the Celonis report.
"We're seeing a lot of good ideas come to the proof-of-concept stage, but as far as, 'How do we get this into an operational state? And how do we actually maintain, build, debug and go through the standard software lifecycle? There are just way more questions than there are answers,’" said an AI product developer at a healthcare company who spoke on condition of anonymity about sensitive internal issues. ."
The AI product developer said his company already has some AI infrastructure in place, such as Snowflake and Databricks for data management, and Azure Kubernetes Service. The Red Hat AI Factory with Nvidia can integrate with Snowflake and Databricks, and customers can choose to run inference workloads on Azure, according to a Red Hat spokesperson. But the AI product developer said he wasn't sure whether adding Red Hat's platform would be feasible at this point.
"If Red Hat is able to solve the question of how to give these concept applications a path to production, and it's clear enough that a non-data scientist could take it and experiment with it, that would be wonderful," he said. "But it still becomes a very difficult conversation to have with leadership about investing more money in platforms and tools.
"I would need to understand what the value of running this platform is on top of all our existing tech, and if it becomes a 'one-stop shop' for all our AI needs, or if we would still have people logging into all our different tools and platforms in order to get their work done. I know Red Hat has a great ecosystem where its products work really well together, but my company has never been able to embrace a single platform."
Integrating Red Hat AI Factory with Nvidia into existing AI infrastructure, particularly in the public cloud, could create a complicated architecture, Strechay said, but that disruption could also be worthwhile to get AI up and running using corporate data on-premises and to manage it as part of a hybrid cloud.
"Most companies still have at least 30% of their data on-premises alone, and another 30% that is considered hybrid, leaving only about 30%+ in the cloud alone," he said.
Beth Pariseau, a senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism. Have a tip? Email her.
