RHEL 10 on NVIDIA DGX Spark Powers Local Enterprise AI Development
Enterprises are no longer just experimenting with generative AI. They are building it into real, governed, production-ready applications. As that shift accelerates, new challenges are emerging fast. Most businesses send all AI prompts to massive, cloud-based large language models. That approach drives up costs, increases latency, and exposes sensitive data unnecessarily.
Red Hat is tackling this problem directly. In collaboration with NVIDIA, the company has announced the development preview of Red Hat Enterprise Linux 10 (RHEL 10) on NVIDIA DGX Spark. Together, they are bringing enterprise-grade AI development directly to the developer’s workstation.
The RHEL 10 on NVIDIA DGX Spark pairing is significant. The hardware runs on the NVIDIA GB10 Grace Blackwell architecture. It delivers up to 1 Petaflop of performance, FP4 data format support, and 128GB of unified memory. That is remarkable compute power for a desktop-class device. Paired with Red Hat’s trusted OS and developer tooling, it forms a high-performance AI workstation built for local agentic workloads and inner development loop testing.
Developers can now build, test, and evaluate complex agentic workflows locally. They can use familiar tools like Red Hat Desktop before pushing anything to production. That local-first approach closes a gap that has long frustrated enterprise AI teams.
This launch addresses three key problems. First, cost control and data sovereignty. By keeping sensitive traffic local and using small language models for simpler tasks, businesses can cut cloud API spend significantly. The 128GB unified memory also supports hosting large language models locally, reducing cloud dependency during model development.
Second, evaluation-driven development. Autonomous AI agents are non-deterministic. Standard unit testing simply is not enough. Running RHEL 10 on NVIDIA DGX Spark gives teams a powerful sandbox for rigorous testing. Developers can use embedded MLflow instances for glass-box trajectory tracing. This captures tool calls and model queries, then runs LLM-as-a-Judge evaluations locally before any code reaches production.
Third, core-to-cluster consistency. Standardizing on RHEL as the underlying OS means developers can focus on building applications instead of managing infrastructure. Furthermore, moving from local development to large-scale deployments on Red Hat OpenShift becomes a far more streamlined and predictable pipeline.
The future of enterprise AI, according to Red Hat, is hybrid, and it starts at the developer’s desk. Enterprises interested in the RHEL 10 on NVIDIA DGX Spark development preview can contact their Red Hat representative to learn more.
