NVIDIA DSX OS: An Open Operating System for AI Factories of Any Size

NVIDIA has announced DSX OS, an open operating system designed to manage "AI factories" of any size. According to a post on the NVIDIA Developer Blog, the company describes artificial intelligence as a new, critical infrastructure, and the data centers that run it as factories that produce intelligence in the form of tokens. As demand for this computing grows, these factories have to scale, and the new operating system is meant to address exactly that.

What an "AI factory" means in NVIDIA's terminology

The "AI factory" metaphor has been a key part of NVIDIA's public rhetoric in recent years. The company describes modern data centers not merely as data storage or hosting for web services, but as a new type of industrial plant: the raw materials are electricity and compute, and the finished product is tokens, generated by AI models in response to requests from users and applications. By this logic, a factory's efficiency is measured not by the number of servers but by the number of tokens produced per unit of energy and time, a metric much closer to industrial production than to traditional IT indicators.

Key facts about the announcement:

• The product is called NVIDIA DSX OS.
• The company positions it as an open operating system.
• It is intended for "AI factories" of any size, from small data centers to the largest.
• The announcement was made through NVIDIA's corporate blog, the NVIDIA Developer Blog.

Why AI infrastructure needs its own OS

As models grow and the clusters used to train them and run inference link tens of thousands of GPUs, managing such infrastructure can no longer be handled by standard operating systems and orchestration tools originally designed for web applications or classic enterprise workloads. Scheduling jobs across thousands of accelerators, using the network between nodes efficiently, and tolerating failures of individual GPUs during long training runs all call for a specialized software layer that understands the specifics of AI computing, rather than simply virtualizing general-purpose resources.

Why openness matters

Positioning DSX OS as an open system fits NVIDIA's broader strategy of building a whole software ecosystem around its chips, one available to a wide range of developers, cloud providers and enterprise customers, not just a select few partners. An open operating system for "AI factories" could potentially let both giant cloud platforms and small teams deploying their own AI infrastructure use a single software foundation, instead of assembling GPU cluster orchestration from disparate tools each time. For the industry as a whole, it is a step toward standardizing how the compute behind modern large language models is managed, regardless of who builds the data center around NVIDIA chips.

Historically, the software layer, and not hardware superiority alone, is considered one of the main factors that cemented NVIDIA's dominant position in the AI accelerator market: for many years, its ecosystem of developer tools and libraries acted as a barrier that made it harder for competing companies to move to alternative hardware. A separate operating system at the level of an entire data center, rather than only libraries for individual GPUs, extends this logic to a new level. It turns the management of an "AI factory" as a single whole, rather than a collection of separate servers, into one more layer where NVIDIA sets standards for the whole industry.

What this means for data center operators

The phrase "of any size" in the DSX OS description indicates that the system is designed not only for giant cloud providers with tens of thousands of GPUs, but also for far more compact deployments: corporate data centers of individual companies or regional operators that build their own AI infrastructure without the resources to develop orchestration from scratch. If an open operating system really does lower the barrier to building a full "AI factory", it could speed the spread of local, rather than only cloud, AI infrastructure among companies that care about data control and long-term cost predictability.