NVIDIA and Microsoft launch RTX Spark and bring 24h AI agents to Windows PC

NVIDIA and Microsoft announced RTX Spark, a superchip designed for PCs Compact Windows that keep AI agents running locally 24 hours a day, without relying on the cloud. At the same event, the two companies previewed the DGX Station for Windows, described as the first supercomputer with GB300 Grace Blackwell class infrastructure focused on the Windows ecosystem. The announcement was detailed in the official NVIDIA blog.

What does RTX Spark change in practice for AI agents?

So far, AI agents — programs that perform tasks alone, such as organizing files, answering emails or consulting internal systems — have depended almost always on cloud servers to think about. The RTX Spark reverses this logic: it is a superchip designed for small machines that start hosting the agent at the table itself, keeping it active continuously.

The practical difference appears on three fronts. The first is latency, since the answer does not need to cross the internet. The second is privacy, because sensitive data stops coming out of the machine. The third is the cost, since the work is no longer charged by token in cloud services. NVIDIA was already exploring this path with self-improvable AI agents for RTX and DGX Spark PCs.

DGX Station for Windows: data center with RTX Spark in the desk

The second announcement is the DGX Station for Windows, previewed as the first table supercomputer with GB300 Grace Blackwell class infrastructure for Windows. It is worth measuring the scale: the same technology family, when organized in racks, is used to train and serve large models — as in the case of NVIDIA Vera Rubin NVL72, which our coverage described as 10x more energy efficient.

Bringing this hardware class to an individual workstation means that tasks currently restricted to clusters — fine-tuning of models, execution of complex agents, heavy prototyping — become the developer’s side. It is the same logic that NVIDIA has already applied in reduced scale with the Jetson line, for local IA without cloud, but at another level of performance.

Why does RTX Spark matter to Brazilian Devs and Companies?

Two factors weigh. The first is the exchange rate: local inference exchanges a monthly dollar account for a unique hardware investment, which changes the mathematics of those who run intense AI loads. The second is compliance: keeping personal data inside the machine simplifies conversation with LGPD, especially in sectors such as health, legal and financial.

For the Brazilian market, the most likely effect in the short term is the popularization of internal assistants who do not need cloud approval to go into production — something relevant for small teams that today discard AI projects at infrastructure cost.

What's the RTX announcement? Spark still won't fix it

NVIDIA and Microsoft did not disclose price, launch date, power consumption or minimum system requirements for RTX Spark or DGX Station for Windows. It is also unclear which AI models will be supported from factory and how it will support older Windows versions.

There are obvious technical limits: local agents compete with giant models that continue to depend on data centers, and the quality of the experience will depend on the size of the model that fits the available hardware. The announcement is a preview — the real assessment depends on availability and price, which still do not exist.

Frequently Asked Questions about NVIDIA, Microsoft and RTX Spark

What is RTX Spark?

It is an NVIDIA superchip geared to compact Windows PCs, designed to keep AI agents running locally 24 hours a day, without relying on cloud servers.

What is DGX Station for Windows?

It is the first supercomputer with GB300 class infrastructure Grace Blackwell focused on the Windows ecosystem, according to the preview presented by NVIDIA and Microsoft.

When do these products reach the market?

There is still no price or availability date disclosed. The event presented a preview of the two products, without a confirmed commercial schedule.