On June 1, NVIDIA announced the DGX Station for Windows — a deskside AI supercomputer built to run frontier AI models and hundreds of concurrent AI agents entirely on-premises, with no cloud dependency required.
The machine is powered by the GB300 Grace Blackwell Ultra Desktop Superchip, delivering 20 petaflops of FP4 compute and up to 748GB of unified memory. That’s enough headroom to run models at the 1-trillion-parameter scale locally — the kind of capability that until recently existed only inside hyperscaler data centers.
The announcement was made at NVIDIA GTC Taipei. Hardware partners Dell Technologies, ASUS, GIGABYTE, HP, MSI, and Supermicro will bring commercial versions to market in Q4 2026.
What Makes This Different
Most enterprise AI deployments today are cloud-dependent. You send data to an API, get a response, and pay per token. That model works, but it comes with constraints: latency, data privacy exposure, ongoing API costs that scale with usage, and no guarantee of consistent performance.
DGX Station for Windows flips the model. You own the hardware, the models run locally, and the data never leaves your building. It was built in collaboration with Microsoft and integrates with the full Windows enterprise management stack, so IT teams can govern it the same way they manage any other enterprise infrastructure.
For regulated industries — healthcare, finance, legal, defence — that local residency isn’t a nice-to-have. It’s a compliance requirement that has previously blocked full AI adoption. This machine removes that blocker.
Hundreds of Agents, Running Simultaneously
NVIDIA’s announcement specifically called out agentic workloads. DGX Station can run hundreds of AI agents executing tasks simultaneously, not just serve individual inference requests.
That’s a meaningful distinction. An AI agent isn’t a chatbot — it’s a system that takes autonomous actions across time, calling tools, making decisions, and completing multi-step workflows. Running many of them in parallel on a single on-premises machine opens up enterprise automation possibilities that cloud-only infrastructure makes economically impractical.
Think of it as dedicated agent infrastructure. A law firm running document review agents, a hospital running clinical workflow agents, a manufacturer running quality inspection agents — all processing internally, at scale, without per-call cloud costs eating into the ROI calculation.
What This Means for Business
For business leaders evaluating AI, this announcement matters in a few ways.
The “we can’t put data in the cloud” objection goes away. Many enterprises have stalled AI adoption because of data residency rules or internal security policies. Local frontier-model compute at this scale removes that barrier without sacrificing capability.
The economics of agentic AI change. Cloud inference costs are real and they compound at scale. A fixed hardware investment with predictable running costs can make the business case for large-scale agent deployment much more straightforward.
The capability gap between enterprise and hyperscaler narrows. When a company can run the same class of model locally that OpenAI uses in its products, the competitive dynamics around AI capability shift. You no longer need to be a cloud-native tech company to run serious AI.
The timeline is Q4 2026. This isn’t available today — it’s a product announcement with a purchase horizon later this year. That gives decision-makers six months to plan, budget, and define the use cases before the hardware lands.
The Bigger Trend
DGX Station is part of a broader infrastructure shift. As AI moves from experimental to operational, enterprises want more control over cost, latency, and data governance. The GPU cloud was the only option when models were too large to run locally. That constraint is lifting.
NVIDIA’s bet is that serious enterprises will want dedicated AI infrastructure — not as a replacement for cloud, but as a tier for their most sensitive, most intensive, most business-critical workloads.
For organisations building internal AI capabilities — whether through in-house teams, AI advisory partners, or managed agent deployments — understanding this hardware layer is increasingly part of the strategy conversation, not just IT procurement.
Enterprise DNA helps business leaders navigate decisions like these. If you’re thinking through your AI infrastructure strategy, talk to an Omni advisor.