NVIDIA RTX Spark Windows AI agents arrive October 16

What changed

NVIDIA and Microsoft announced a co-engineered platform for running AI agents natively on Windows PCs. At a Microsoft event in San Francisco on Wednesday, NVIDIA founder and CEO Jensen Huang and Microsoft CEO Satya Nadella outlined two new hardware systems and Windows infrastructure designed for the agentic era.

The first is RTX Spark, a superchip combining an NVIDIA Blackwell RTX GPU with up to 6,144 cores and an up to 20-core NVIDIA Grace CPU connected at 600 GB/s. According to the announcement, RTX Spark delivers one petaflop of FP4 AI performance and up to 128GB unified memory. The company says it can run models such as Qwen 3.8 Flash Next, a 125B model with 51B n-gram that matches the intelligence of many cloud models, without sending data to the cloud.

Laptop preorders for RTX Spark systems are open now, with availability on October 16. Compact desktop configurations will be available for sale in November. Systems are coming from Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI and Gigabyte. Microsoft also announced Surface Laptop Ultra, built around RTX Spark with up to 128 gigs of unified memory.

The second announcement is NVIDIA DGX Station for Windows, previewed at the event as the first deskside AI supercomputer to bring GB300 Grace Blackwell-class infrastructure to Windows. According to the company, DGX Station for Windows runs on the GB300 Grace Blackwell Ultra Desktop Superchip, delivering 748GB of coherent memory and up to 20 petaFLOPS of FP4 AI compute, enough to run models up to a trillion-parameter scale locally.

On the Windows side, Microsoft announced general availability of Microsoft Execution Containers (MXC), OS-level infrastructure that lets agents run safely and persistently in the background under operating system control. The company says MXC, combined with Microsoft Security and Agent 365, will unlock the full potential of agents on Windows.

Huang noted the historical connection, saying NVIDIA was founded because of Windows and now AI agents are coming to Windows. “If not for Windows there would be no GeForce,” he said in the announcement.

Why it matters

This marks a shift in where enterprise AI workloads can run. Until now, DGX Station ran only on Linux, which meant enterprise developers maintained two separate environments: Linux for heavy AI workloads and Windows for the productivity tools and applications they already use. According to the announcement, the vast majority of Fortune 500 companies are standardized on Windows, and that gap has cost developers time and resources.

RTX Spark changes the hardware equation for local AI on Windows. By placing significant AI compute directly in consumer and business laptops and compact desktops, the platform shifts computational work from cloud servers to user machines. The company says developers can now run models locally without rewriting code, running the same NVIDIA AI stack from RTX Spark to DGX Station.

For developers, the practical effect is that they can build and run AI agents that connect directly to the Windows applications and infrastructure they already use, and fine-tune and run inference on large models without leaving their primary machine. For enterprises, DGX Station for Windows brings frontier-class model capabilities to the enterprise desktop instead of requiring separate Linux infrastructure or cloud rental.

The Windows agent infrastructure also addresses security concerns. Microsoft says it built the desktop as the most secure place for agents to execute, with MXC providing observability and governance of agent behavior at the OS level.

What to test

Before adopting RTX Spark or DGX Station for Windows, organizations should verify several vendor claims against real-world performance:

Performance claims. The company claims RTX Spark can run Qwen 3.8 Flash Next (125B parameters) locally without cloud calls. Testing should verify actual inference speeds, latency, and whether claimed performance holds under production workloads with multiple concurrent agents.

Memory and model capacity. RTX Spark claims up to 128GB unified memory. Test whether models actually load, run, and complete inference within these constraints, especially at the claimed parameter scales.

Software stack compatibility. NVIDIA says RTX Spark runs the full CUDA platform. Verify that existing AI development tools, frameworks, and workflows actually port from other NVIDIA hardware without modification.

Thermal and power operation. The compact desktop is described as designed for 24/7 operation. Testing should measure actual power consumption, thermal behavior, and reliability under continuous agent workloads.

MXC security and governance. Microsoft’s claims about agent isolation, observability, and governance need validation. Verify that agents actually run persistently without interfering with Windows stability, and that security containers provide the promised isolation.

Availability and supply. Preorders have opened, but real-world availability and lead times should be monitored. The company stated laptops will be available October 16 and desktops in November. Organizations should track whether these dates hold and whether supply meets demand across manufacturers.

The conclusion

NVIDIA and Microsoft have announced hardware and infrastructure designed to bring local AI agent workloads to Windows PCs at scale. RTX Spark offers compute density in compact form factors, while DGX Station for Windows addresses a real challenge: enterprise developers forced to choose between Windows productivity tools and Linux AI infrastructure.

The announcements signal that enterprise AI development is moving away from cloud-only models toward hybrid approaches, with heavy compute available locally. Whether they deliver depends on real-world testing of performance, software compatibility, and operational stability. The fact that the vast majority of Fortune 500 companies standardize on Windows means the market opportunity is substantial, but execution matters. Developers moving to these systems should set clear benchmarks before adoption and validate that performance claims translate to their specific workloads and agent configurations.

What to watch next: actual delivery dates and customer reports on performance and reliability, Windows agent ecosystem maturity and tooling support, adoption rates among enterprises currently on Windows-only infrastructure, and competitive responses from other chip makers offering local AI solutions.