DSX Platform Gives Infrastructure Builders a Playbook for AI Factories: what changed

At GTC Taipei, NVIDIA introduced DSX, a codesigned platform intended to standardize how large — scale AI infrastructure is planned, built and run. The announcement frames DSX as an integrated playbook — combining open-source software, APIs, reference designs and validated hardware architectures — to shorten lead times, lower deployment risk and drive down the cost of AI inference and training at scale. For data-center operators and hyperscalers, the platform aims to make capacity planning and production rollouts more predictable.

DSX presents a unified framework that spans compute, networking, storage, facilities and partner technologies. Rather than treating those layers independently, the stack provides coordinated reference architectures and integration guidance so design, procurement and operations teams can validate choices up front. The company says that alignment across layers is intended to improve operational reliability and resiliency as installations scale.

A central performance element is DSX MaxLPS, a suite engineered to maximize token performance per megawatt within a fixed power budget. MaxLPS uses 45‑degree-C liquid cooling and in-rack optimizations to let operators run up to 40% more GPUs at their most energy — efficient operating point while keeping workload impact minimal. The package is presented as a lever to reach the platform’s stated goal of the lowest token cost, by squeezing higher GPU density and efficiency from a constrained power envelope.

On the software and operations side, DSX OS is an open-source, modular layer built for AI factory lifecycles. It provides lifecycle management, intelligent scheduling, runtime consistency, health automation, resiliency, multi — tenant controls and platform services tailored to production AI workloads. Complementary pieces include DSX Sim, a high-fidelity simulation layer for modeling and validating infrastructure decisions before spending capital, and DSX Reference Design, which supplies generation — specific validated architectures covering compute, networking, storage, cluster layout and facilities.

DSX also includes operational integration tooling aimed at tying AI deployments into broader energy and plant systems. DSX Flex links AI factories to power — grid signals — supporting load shedding, demand response and pricing events — and orchestrates utility supply, onsite renewables and storage. DSX Exchange enables secure, scalable exchange of compute, network, energy, power and cooling telemetry and control between IT, operational — technology and operations agents to coordinate plant — level responses.

The company says industry — leading manufacturers are building DSX-ready systems through extreme codesign, and that expanding partnerships across every layer of the stack will accelerate global rollouts. Vendor claims emphasize end-to-end alignment of silicon, systems, software and facilities as a way to speed time to first production and improve operational predictability at scale.

For infrastructure builders, the platform promises concrete benefits: simulate an entire AI factory in DSX Sim before committing capital, validate expected performance prior to rack installation, and operate production AI with automated resilience and multi — tenant controls. “We’re not just shipping chips — we’re giving every infrastructure builder a complete playbook to build AI factories,” said Jensen Huang, founder and CEO, describing DSX’s intent to reduce deployment risk and cost.