Nvidia cuts AI cluster build times with new containerised data centre designs

Nvidia has introduced new reference architectures designed to simplify the construction of artificial intelligence data centres and robotic systems for its customers. Chief Executive Officer Jensen Huang presented these technologies at the company’s annual GTC Berlin event, highlighting tools that streamline hardware assembly and software deployment.

The primary offering, DSX POD, serves as a blueprint for manufacturers to build AI clusters that fit within standard shipping containers. Each unit contains Rubin graphics processing units, Vera central processing units, networking equipment, cooling systems and power distribution modules. Nvidia states that these components are integrated and tested at the factory, eliminating the need for customers to assemble the systems themselves.

What we know

According to SiliconANGLE, Nvidia estimates that DSX POD can reduce hardware deployment times from six months to just 14 days. The company expects the first systems to ship early next year. More than a dozen hardware makers, including Dell Technologies and Hewlett Packard Enterprise, have confirmed plans to adopt the DSX POD architecture.

Huang emphasised that value is created when an AI factory begins producing tokens, rather than when construction is merely complete. This perspective drives the push for pre-assembled, factory-tested units that accelerate time-to-market for enterprises seeking to deploy large-scale AI infrastructure quickly and efficiently.

Alongside DSX POD, Nvidia debuted a second reference architecture tailored for OpenShell, an AI agent toolkit released last month. This blueprint facilitates the deployment of OpenShell on Nvidia silicon by combining Vera CPUs, Rubin GPUs and BlueField-4 accelerators. The BlueField chips offload auxiliary computing tasks from the main processors, improving overall system efficiency.

The new architecture also allows users to run OpenShell environments at the edge. It supports Nvidia’s DGX Station workstations, the DGX Spark miniature desktop and laptops featuring the RTX Spark system-on-chip. This flexibility enables organisations to deploy AI agents across diverse hardware form factors, from large data centres to portable devices.

Defender's checklist

• Verify that AI agents run in isolated sandboxes to prevent access to sensitive infrastructure.
• Ensure a supervisor module monitors all traffic exiting agent sandboxes for potential risks.
• Confirm that hardware components are pre-integrated and tested to reduce deployment errors.
• Check that edge devices support the necessary security boundaries for agent containment.

The background in plain language

Reference architectures act as detailed blueprints that guide how hardware and software components should be assembled. They reduce complexity by providing tested, standardised designs that partners can replicate. For AI deployments, this means less time spent on custom integration and more focus on operational readiness. Sandboxing isolates processes so that if one fails or behaves unexpectedly, it does not compromise the wider system. This containment strategy is crucial for managing autonomous agents that interact with external data or systems, ensuring they operate within defined safety limits without exposing core infrastructure to unnecessary risk.

Background: AI agents

AI agents are the software entities that perform specific tasks using instructions from large language models. Agentic AI is the architectural approach that grants these agents autonomy to plan, reason and act. You use agents as tools within agentic AI systems to achieve complex, multi-step outcomes without manual oversight at every stage.

Read the full guide: AI Agents vs Agentic AI: The Distinction That Matters

What to do and how to stay safe: Nvidia

• Review your current AI deployment plans to see if pre-integrated architectures could reduce build times.
• Assess whether your existing infrastructure supports sandboxed environments for AI agent isolation.
• Monitor vendor announcements for updates on reference architecture compatibility with your hardware.
• Evaluate the need for supervisor modules to monitor agent traffic in your security protocols.

General security guidance from the TechWire Beat newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What is DSX POD?

DSX POD is a reference architecture that enables manufacturers to build AI clusters fitting into containers, with components pre-tested at the factory.

How much faster does DSX POD claim to be?

Nvidia estimates DSX POD can cut hardware deployment times from six months to 14 days.

What is OpenShell?

OpenShell is an AI agent toolkit that runs agents in sandboxes, isolating them from sensitive infrastructure while a supervisor monitors traffic for risks.