The move and the stated figures
Dell Technologies announced on October 22 the expansion of its offering for enterprise agentic AI in collaboration with Nvidia, extending the Dell AI Factory solution and its partner ecosystem from desktop to data center. The stated goal is to enable companies to develop and manage AI in environments they own and control internally. According to the source, more than 6,500 companies have adopted Dell AI Factory with Nvidia, up from 5,000 in the previous quarter.
Dell frames the context with its own forecast: by 2030, inference costs will fall by more than 90%, while token consumption for agentic AI will increase 24-fold. Hence the argument that integration is needed among hardware, software, models and operational tools. The company also claims that its infrastructure can offer up to 90% greater efficiency than the public cloud over two years, but specifies that this is an analysis based on specific workloads and cost assumptions, with results varying by environment.
Desktop, models and security
On the software side, Reflection's open-weights Beam model has been validated on Dell AI Factory with Nvidia and can be used on owned infrastructure for software engineering and long-running agentic tasks; it is distributed under the Apache 2.0 license. Dell has also added Dynatrace AI observability, Mistral's knowledge assistant and agentic platform, and Foghjtspear computer vision to the ecosystem, while SUSE Linux expands the software options.
For security, Dell combines Red Hat AI Factory software, which manages platform orchestration, with the NVIDIA OpenShell reference architecture for runtime governance. The company says the multilayer approach allows AI agents to be controlled across runtime, network, data and infrastructure even at scale. On the desktop side, the Dell Deskside Agentic AI solution with Nvidia offers verified reference templates for use cases such as agentic coding, NemoClaw agents and AI labs for higher education, designed to scale from individual experiments to more than 1,000 systems.
Hardware and availability timelines
The desktop hardware includes Dell Pro Precision with GB10, with 64 GB of unified memory to run leading open-weights models locally and connectable in clusters of two or four units. The 64 GB model will be on sale from October 23. Also planned is Dell Pro Precision with GB300, intended for Windows environments, with up to 748 GB of coherent memory and 20 petaflops in FP4.
For the data center, Dell is developing a modular data center based on the NVIDIA DSX POD reference design: a containerized AI Factory that integrates compute, storage, networking, rack power and liquid cooling, with a factory pre-integrated configuration. Announced for early next year, it joins servers such as PowerEdge M9882, with 256 Nvidia Vera CPUs in a 48U rack, expected in the first quarter of 2027, and the air-cooled PowerEdge R8822 with two Vera sockets and up to four double-wide GPUs. For large-scale inference, PowerEdge XE9810 is planned with 16 Groq 3 LPU trays with direct liquid cooling, also in the first quarter of 2027. Finally, the Nvidia Aerial RAN Computer platform on PowerEdge servers aims to combine RAN and AI workloads for telecom operators.
What remains to be verified
The executives' statements, as reported by the source, include CEO Michael Dell arguing that enterprises treat AI infrastructure as a factor for growth and productivity, while Nvidia's Jensen Huang speaks of the need for a new computing environment for agents that reason, use tools and learn. The data on customers, savings and forecasts to 2030 come from the company and cannot be verified with the available material. Availability timelines vary: Beam and the Mistral solutions are already available, Dynatrace and Foghjtspear in the fourth quarter of 2026, SUSE Linux at the beginning of the first quarter of 2027.
