Dell unveils new agentic AI systems

Dell Technologies has expanded its Dell AI Factory with Nvidia (NVDA) portfolio with a new set of systems built for enterprise agentic AI workloads. The launch includes a Vera CPU server, a Vera CPU rack, and a rack-scale inference system based on Groq 3 Language Processing Units. Around the time of these announcements, Dell shares traded at $483.82, up approximately 4.82%, reflecting investor focus on the company’s AI infrastructure strategy and demand metrics.

Recent financial figures underscore the scale of Dell’s AI push. The company reported $60.9 billion in AI-server orders and $16.4 billion in AI-server revenue, alongside a $95 billion backlog. Its infrastructure division generated $31.8 billion in revenue, up 89%, while operating income rose 225%, highlighting rapid growth in AI-related infrastructure activity.

The Dell PowerEdge R8822 is an air-cooled, two-socket Nvidia (NVDA) Vera CPU server designed for standard 19-inch rack environments. It is targeted at agentic AI tasks such as agent orchestration, sandboxing, reinforcement learning, data processing, and analytics, with support for up to four double-wide GPUs to handle acceleration-heavy workloads.

At the core of the R8822 is the Arm-based Nvidia (NVDA) Vera CPU, built on 88 Olympus cores per socket, providing 176 threads and up to 1.2 TB/s of memory bandwidth per socket. The system is scheduled to ship in October 2026, making it the first of Dell’s new agentic AI offerings to reach customers.

The Dell PowerEdge M9882 implements Nvidia’s Vera CPU Rack reference architecture, packing 256 Nvidia Vera CPUs into a single 48U rack. Each Vera CPU carries 88 Olympus cores, bringing the rack’s total to 22,528 cores for large-scale CPU-side agentic workloads such as orchestration, tool use, code execution, and evaluation.

The design separates CPU compute from GPU clusters, allowing data centers to size each resource pool independently to match workload needs, with the rack described as purpose-built for agentic AI. The M9882 is slated for availability in the first quarter of 2027, and requires planning around power and cooling alongside existing GPU infrastructure.

The PowerEdge XE9810 is Dell’s first rack-scale inference system based on Groq 3 LPUs. It features 16 direct-liquid-cooled compute trays and is positioned as a disaggregated complement to GPU clusters, aimed at the high-throughput demands of production agentic AI workloads and consistent time to first token at scale.

Dell has not disclosed the number of LPUs per tray or aggregate throughput figures, but the XE9810, like the M9882, will require data center liquid cooling infrastructure. Availability for the XE9810 is planned for the first quarter of 2027, aligning its rollout with the Vera CPU disaggregation rack.

Alongside the data center systems, Dell is adding a 64GB GB10 deskside configuration, due October 23. It is aimed at individual developers and small teams running open-weight models locally, offering a smaller-scale entry point into agentic AI workloads. Pricing for the 64GB configuration has not been published.

GB10 systems can scale to an eight-node GB10 cluster for increased model capacity and concurrent user support.