What Is NVIDIA DSX Air and Why It Matters
NVIDIA is pushing a new way to build and run large scale AI infrastructure with a platform called NVIDIA DSX Air. Announced at GTC 2026 in San Jose, DSX Air is part of the wider DSX platform, NVIDIA’s blueprint for AI factories.
Think of an AI factory as a giant, specialized data center designed to train and run AI models at scale. These environments are packed with GPUs, high speed networking, storage, security and orchestration tools. Getting everything working together can normally take weeks or months.
DSX Air changes this by offering a software as a service platform for logically simulating an entire AI factory before a single physical server shows up. It lets companies build a high fidelity digital twin of their infrastructure, including:
- GPUs, SuperNICs and DPUs
- Switches and advanced networking
- Storage platforms and routing
- Security stacks
- Orchestration and scheduling tools
By doing all the integration and troubleshooting in simulation, organizations can cut their time to first token from weeks or months to just days or even hours. For companies racing to bring new AI capacity online, this is a big shift in how infrastructure is planned and deployed.
How DSX Air Powers an AI Factory Ecosystem
NVIDIA designed DSX Air not just for end users but for the entire ecosystem that builds and supports AI factories. Server vendors, orchestration platforms, storage providers and security companies can all plug into the same simulated environment and validate how their solutions work together at scale.
Server manufacturers can model customer specific reference architectures without building massive physical labs. Because enterprise AI setups rarely follow a single fixed design, vendors can use DSX Air to create custom digital twins for each customer, tune their software stacks and deliver validated designs without touching hardware.
Orchestration vendors also benefit. At GTC, NVIDIA showed a multi tenant RTX PRO Server environment running entirely inside DSX Air. In that demo:
- Netris provided network orchestration
- Rafay handled host orchestration
- NVIDIA Run:ai optimized GPU allocation
All of this ran in simulation, letting partners test complex workflows under realistic conditions without building physical clusters.
Storage and data platforms can do something similar. At the GTC booth, NVIDIA demonstrated a video retrieval augmented generation workload running on the VAST AI Operating System inside DSX Air. The environment included a fully simulated VAST cluster, DataEngine nodes and a video search and summarization front end. DataEngine triggers and functions processed and indexed video through an end to end pipeline, showing how real AI applications can be designed and validated virtually before infrastructure is rolled out.
Security vendors, who usually face some of the most demanding validation needs, can test multi tenant policies, DPU accelerated isolation and threat detection in DSX Air as well. One example demo featured Check Point’s distributed firewall running on simulated BlueField DPUs, TrendAI Vision One for threat detection and Keysight Cyperf generating realistic traffic. That setup lets security teams discover vulnerabilities and refine policies in a customer’s digital twin long before any live deployment.
Across all of these use cases, partners highlight the same advantage. DSX Air gives them a scalable and cost effective way to validate their solutions with NVIDIA infrastructure and with each other in one shared environment.
From Months to Days: A New Operational Model for AI Factories
DSX Air is not just about speeding up the initial deployment. It introduces a new lifecycle model for how AI factories can operate and evolve over time.
First, organizations build their intended production environment entirely in simulation. They replicate their planned setup for networking, compute, storage, security, orchestration and scheduling. This virtual build lets teams verify that everything works end to end, expose issues early and confirm performance expectations.
Next, they move to physical deployment with more confidence. Since the environment has already been tested in detail, the odds of a smooth bring up are much higher. Instead of spending their first weeks fighting integration bugs, teams can focus on running real workloads. Time to first token shrinks dramatically.
Once the physical system is live, DSX Air continues to add value. It becomes a long lived sandbox for change management where teams can:
- Test upgrades to drivers, firmware and software stacks
- Rehearse maintenance windows
- Validate new policies or architecture changes
- Predict impact on performance and reliability
Only after changes pass in the simulated environment are they applied to production, helping maximize uptime and keeping critical AI services available.
This simulation first approach is already in use by real providers. Siam.AI, the largest AI cloud provider in Thailand, uses DSX Air to accelerate its deployments. By rehearsing NVIDIA best practices in simulation, they achieved day one operational readiness and validated their architecture before the hardware even arrived.
Hydra Host is another example. They are building Brokkr, an AI factory operating system for bare metal GPU provisioning that supports dozens of GPU deployments around the world. Using DSX Air, Hydra Host can simulate full stack environments and validate Brokkr’s automation and orchestration workflows across different networking and hardware setups at scale. That lets them ship tested infrastructure faster while protecting live systems as global demand for AI capacity continues to grow.
As AI factories become larger and more complex, the ability to validate full stack environments before deployment will heavily influence who can innovate fastest. NVIDIA DSX Air aims to be the backbone of that process, giving organizations a faster path to first token and a safer, more predictable way to operate high performance AI infrastructure over time.
Original article and image: https://blogs.nvidia.com/blog/dsx-air-simulation-ai-factories/
