Dell Expands Enterprise Agentic AI with NVIDIA

• • Always-on AI agents make hundreds of model calls per workflow and not every step needs a frontier model.
  • NVIDIA Nemotron 3.5 Lightning delivers up to four times higher throughput for high-volume specialized tasks and can be post-trained for specific enterprise domains.
  • NeMo Switchyard automatically routes each agent step to the best available model with no configuration required.
  • Dell Deskside Agentic AI gives enterprises local, secure access on purpose-built AI workstations.

The default approach to building AI into enterprise workflows has been straightforward: pick a frontier model, send it your prompts, get your answers. But as organizations move from AI assistants to always-on AI agents, systems plan, act, verify and repeat across complex multi-step workflows, and they need more optimized AI solutions.

Agents make dozens, sometimes hundreds, of model calls across a single workflow. Not every call needs the same level of intelligence. Routing each step through a frontier model regardless of complexity wastes compute, along with the tokens, slows agents down and drives up costs unnecessarily.

Meet the models built for how agents actually work

NVIDIA is addressing this shift to more efficient agentic AI with two new enhancements to the NVIDIA AI software stack: NVIDIA Nemotron 3.5 Lightning and NeMo Switchyard.

NVIDIA Nemotron 3.5 Lightning is a fully customizable open model designed for the high-volume execution layer of agentic AI. Built on a hybrid Mixture of Experts (MoE) architecture with up to 30 billion parameters and 3 billion active, it delivers up to four times higher throughput compared to similarly sized models. It’s distilled from NVIDIA’s frontier Nemotron 3 Ultra model, bringing strong out-of-the-box accuracy for tool calling, instruction following and multi-turn workflows.

Organizations can post-train Nemotron 3.5 Lightning on their own domain data, policies and workflows to push accuracy higher for specialized tasks and use cases, such as:

• • Cybersecurity agents triaging alerts and classifying incidents
  • Financial services agents extracting data and monitoring risk signals
  • Telecom agents handling network alarms and customer care
  • Retail agents resolving inventory exceptions and answering order questions

• Smarter routing, without the complexity

  NVIDIA NeMo Switchyard is an open source model routing library that automatically paths each step of an agent workflow to the best available model from your pool, open or closed. No training or configuration is required to get started.

It gets smarter as your agents run, improving routing decisions over time based on actual usage. The result is frontier-level accuracy where you need it and specialized efficiency everywhere else.

Running it locally with Dell Deskside Agentic AI

A system of models is only as good as the infrastructure running it. Cloud-only approaches introduce latency, unpredictable costs and data sovereignty risks that compound quickly as agentic token usage scales. The most efficient agentic architectures run workhorse models locally for high-volume specialized tasks and reserve frontier models only when complexity demands it.

Dell Deskside Agentic AI, part of the Dell AI Factory with NVIDIA, gives enterprises a local, secure and cost-predictable foundation to deploy and run agentic workflows at the desk. It supports the NVIDIA NemoClaw reference stack, part of the NVIDIA Agent Toolkit, giving teams direct access to Nemotron open models including Nemotron 3.5 Lightning, NeMo Switchyard routing and the NVIDIA OpenShell secure runtime on local hardware. To help organizations move from deployment to production, Dell Services provides end-to-end guidance from initial strategy through ongoing optimization.

Dell Deskside Agentic AI addresses different workload requirements:

• • Dell Pro Max with GB10 handles individual agent prototyping and smaller-scale workflows, supporting models from 30 billion to 200 billion parameters.
  • Dell Pro Max T2 and Dell Pro Precision 9 T2/T4/T6 with NVIDIA RTX PRO™ Blackwell GPUs deliver scalable performance for workhorse-class GPU workloads, supporting models from 30 billion to 500 billion parameters.
  • Dell Pro Max with GB300 is purpose-built for frontier-level inference, supporting models from 120 billion to 1 trillion parameters, and on-premise deployment of an agent fleet at scale across an enterprise.

The infrastructure advantage starts now

The shift from single-model AI to systems of models is already underway. The organizations that get ahead of it, with the right models, the right routing and the right local infrastructure, will build agents that are faster, more accurate and far more cost-effective than ever before.

Availability

• • NVIDIA Nemotron 3.5 Lightning and NeMo Switchyard are available now as part of the NVIDIA Agent Toolkit on Dell Deskside Agentic AI.
  • NeMo Switchyard is also available on GitHub.
  • Dell Pro Max with GB10, Dell Pro Max T2, Dell Pro Precision T2/T4/T6 and Dell Pro Max with GB300 are available now.