NVIDIA DSX Air accélère le temps de génération des tokens grâce à la simulation pour les usines d'IA

Setting up AI factories in simulation — decreasing deployment time from months to days — is accelerating the next industrial revolution. Nowhere was that more apparent than at GTC 2026, in San Jose, where NVIDIA founder and CEO Jensen Huang introduced NVIDIA DSX Air. Part of NVIDIA DSX Sim in the DSX platform, NVIDIA’s blueprint for AI factories , DSX Air is a software-as-a-service platform for logically simulating AI factories. It delivers high‑fidelity digital simulations of NVIDIA hardware infrastructure, including GPUs, SuperNICs, DPUs and switches, and it integrates with leading partner solutions for storage and routing, security, orchestration and more via open, API-based connectivity. NVIDIA DSX Air enables a complete AI factory ecosystem, uniting NVIDIA infrastructure with partner technologies to deliver full‑stack simulation and accelerate complex AI deployments. Companies building some of the world’s most advanced AI infrastructure, including CoreWeave , are already using DSX Air to simulate and validate their environments long before hardware reaches the loading dock. The development underscores a new reality: simulation is now essential to accelerating AI deployment at scale. DSX Air allows organizations to construct a full digital twin of their AI factory — compute, networking, storage, orchestration and security — before a single server is unboxed. By shifting integration and troubleshooting into simulation, customers are reducing the time to first token from weeks or months to mere days or hours, saving enormous amounts of time and costs. An industry analogy for this AI factory simulation phenomenon explains it well: It’s like IT mirroring your laptop to set up a new one, except the “laptop” is a hyperscale AI factory and the “mirroring” is a complete, high‑fidelity replica of the production environment. For operators racing to bring new AI capacity online, this change is transformative. Building a Platform for an Entire Ecosystem The NVIDIA DSX Air simulation platform is designed to support the entire AI factory ecosystem. Server manufacturers, orchestration vendors, storage providers and security partners can all validate their offerings alongside NVIDIA infrastructure — together, in one environment, at scale. This ecosystem‑wide capability is already reshaping partner workflows. Server manufacturers, which serve as the primary channel for enterprise inference, can now model and validate their reference architectures without building expensive physical labs. Enterprise AI environments rarely fit rigid designs, and customers often require bespoke configurations. With DSX Air, manufacturers can create digital twins tailored to specific customer needs, test their software stacks and deliver validated solutions without touching hardware. Orchestration vendors — critical for enterprises and tier‑2 clouds that need turnkey AI services — gain the ability to test at scale. At GTC, NVIDIA showcased a multi‑tenant RTX PRO Server environment running entirely in simulation, with Netris providing network orchestration, Rafay handling host orchestration and NVIDIA Run:ai optimizing GPU allocation. These partners can now validate complex workflows under realistic conditions without deploying physical clusters. The simulation environment is also valuable for validating the data platforms that power AI factories. Instead of requiring large physical clusters, DSX Air allows ecosystem partners to model complete AI workflows alongside NVIDIA compute, networking and software infrastructure. At GTC, the booth demonstration features a video retrieval-augmented generation workload running on the VAST AI Operating System, including a fully operational VAST cluster with DataEngine nodes and the video search and summarization front end. DataEngine triggers and functions process and index video content through an end-to-end pipeline, illustrating how AI applications can be designed, tested and validated inside the DGX Air simulation before deploying physical infrastructure. Security vendors — facing some of the most demanding validation requirements — can now test multi‑tenant policies, DPU‑accelerated isolation and threat detection in a realistic environment. The GTC demo includes Check Point ’s distributed firewall running on simulated BlueField DPUs, TrendAI Vision One for threat detection and Keysight AI Inference Builder, an emulation and analytics platform designed to validate inference-optimized AI infrastructure at scale. Security partners can identify vulnerabilities and validate policies in a customer’s digital twin long before production goes live. Across the ecosystem, partners emphasized the same point: DSX Air gives them a complete, scalable and cost‑effective way to validate their solutions with NVIDIA infrastructure and with each other. Operating With a New Model to Accelerate Time to Token NVIDIA DSX Air isn’t just a deployment accelerator — it introduces a new operational model for AI factories. On