A 2U rack server in a data-center cabinet — illustration for the Dell PowerEdge R8822 with NVIDIA Vera news

News ·

Dell PowerEdge R8822 — a new NVIDIA Vera server for AI agents

Dell introduced the PowerEdge R8822, a two-socket 2U server with NVIDIA Vera CPUs and support for up to four GPUs, designed for AI agents, analytics and reinforcement learning.

Source: Dell Technologies

Dell introduced the PowerEdge R8822, a two-socket 2U server with NVIDIA Vera CPUs and support for up to four GPUs, designed for AI agents, analytics and reinforcement learning.

Dell announced the server on 21 October 2026 at NVIDIA GTC in Berlin. The R8822 matters in the Dell portfolio because it brings NVIDIA Vera CPUs into a chassis that fits a typical enterprise rack.

NVIDIA Vera in a standard rack

The PowerEdge R8822 is a two-socket 2U server for standard 19-inch racks, cooled by air. Dell states support for up to four double-wide GPUs, so CPU work can sit alongside accelerated workloads.

It is an expansion path next to large, liquid-cooled rack installations. That does not mean every R8822 configuration is suited to training large models — the fit depends on the GPUs, memory, software and the workload itself.

What is the Dell PowerEdge R8822 for?

According to Dell, the server is optimised for several classes of work:

  • AI agent orchestration — planning tasks, calling tools and coordinating many processes.
  • Sandboxing and code execution — isolated environments for work started by agents.
  • Reinforcement learning — compute that supports training loops and evaluation of model behaviour.
  • Analytics and data processing — preparing data for AI use and running analytical jobs.
  • Mixed CPU/GPU tasks — keeping the agent and data layer separate from GPU-accelerated model work.

Why new AI servers need capable CPUs

In many AI environments the spotlight stays on GPU throughput, for example during language-model inference. Agentic systems raise the importance of another layer: tool use, program execution, data preparation, many concurrent services and process control.

In that model the CPU and GPU do not have to do the same job. Processors can handle orchestration and execution, while accelerators run the heaviest model computation. The R8822 follows that direction, although the performance benefit still has to be measured for a specific scenario.

Part of the wider Dell AI Factory with NVIDIA

In the October announcement Dell also described other platforms:

Model

Role

Timing stated by Dell

PowerEdge R8822

Agentic AI and analytics in a standard rack

October 2026

PowerEdge M9882

Large-scale CPU compute — 256 NVIDIA Vera CPUs in a 48U rack

Q1 2027

PowerEdge XE9810

High-throughput inference with Groq 3 LPU accelerators

Q1 2027

The split shows that AI infrastructure is increasingly made of specialised resources for orchestration, data processing and inference, rather than one server type for every job.

A broader look at the same announcement is in Dell and NVIDIA: new PowerEdge servers and Dell AI Factory.

What this means for enterprises

The PowerEdge R8822 is relevant for organisations building their own AI-agent environments, automation platforms and systems that use company data. A standard rack chassis and air cooling make it easier to consider inside an existing data centre.

Before a deployment it is worth checking operating-system, library and AI-tool compatibility with NVIDIA Vera CPUs, choosing the GPUs, estimating memory and power, and confirming which configurations can actually be ordered. Dell said the R8822 would be available in October 2026. Official materials do not separately confirm stock availability in Poland.

Planning a local AI environment is covered in Local AI in the enterprise — how to deploy your own intelligence.

Sources

1. Dell Technologies — press release, 21 October 2026

2. Dell — Compute That Matches the Moment: Latest PowerEdge Platforms Take on AI at Scale

Planning secure AI agents in your organisation?

Talk to Alterkom about local LLMs, access control and infrastructure for enterprise AI agents.