SUSE and NVIDIA AI Factory

SUSE said on April 21 it is packaging its software with NVIDIA’s AI tools into a ready-made stack for companies that want to run AI on their own infrastructure. (suse.com) The product, announced at SUSECON 2026 in Prague, is called SUSE AI Factory with NVIDIA. SUSE said it combines SUSE AI, NVIDIA AI Enterprise, Rancher management tools and GitOps automation so teams can move from a developer sandbox into production. (suse.com) SUSE said the stack includes NVIDIA NIM microservices for inference, open Nemotron models, NeMo tools for building agents, Run:ai for graphics processing unit orchestration, and Kubernetes operators for managing the hardware and software layers. SUSE said customers can run it from edge sites to core data centers to public cloud environments. (suse.com; suse.com) An AI factory is NVIDIA’s term for a data center tuned to turn data into AI outputs at scale, especially inference, not just to store data or run mixed workloads. NVIDIA said in March 2025 that inference and “reasoning” workloads are pushing companies toward purpose-built systems with much higher token throughput. (blogs.nvidia.com) SUSE is pitching this to companies and governments that do not want sensitive data, models or internal processes sent to a shared public service. Its announcement says the system is built for “digital sovereignty,” with private infrastructure controls, auditability and policy-based security. (suse.com; suse.com) That lands in a market where SUSE’s own March 19 survey found 59% of organizations prioritize hybrid infrastructure for AI workloads, and where the company said U.S. concerns about vendor lock-in and digital sovereignty are rising. SUSE also cited an International Data Corporation FutureScape forecast that 60% of Global 2000 companies will operate AI factories by 2028. (suse.com; suse.com) NVIDIA has been building the surrounding blueprint in parallel. In January, it expanded its Enterprise AI Factory validated design with BlueField networking and security offload, saying the hardware can isolate security, storage and networking tasks so central processors and graphics processors stay focused on AI work. (blogs.nvidia.com) For buyers, the pitch is less about a single new chip than about reducing integration work across servers, containers, drivers, models and security controls. SUSE said the goal is to standardize deployment with pre-validated blueprints and a curated software supply chain rather than forcing each customer to assemble the stack from scratch. (suse.com; suse.com) The immediate test is whether companies that have already built AI pilots now buy full private stacks to run them in production. SUSE and NVIDIA are betting that security rules, data-control demands and the cost of improvising infrastructure will push more of that work in-house. (suse.com; blogs.nvidia.com)