NVIDIA and Microsoft announced co-engineered hardware and operating system updates designed to run autonomous artificial intelligence agents directly on Windows personal computers, NVIDIA reported on Wednesday.
Speaking at the Windows AI and Surface event at Dogpatch Studios in San Francisco, NVIDIA Chief Executive Jensen Huang and Microsoft Chief Executive Satya Nadella outlined their joint effort to bring local agents to enterprise and consumer devices. The session was hosted by former White House AI policy advisor Sriram Krishnan. Huang noted that Microsoft Windows was central to NVIDIA's founding and the creation of its GeForce brand.
Microsoft introduced general availability for Microsoft Execution Containers, providing operating system infrastructure that lets AI agents run persistently in the background. Pavan Davuluri, Microsoft executive vice president of Windows and devices, explained that the software coordinates with Microsoft Security and Agent 365 to govern background workloads. Huang compared the runtime framework to the historical debut of DirectX for PC graphics applications.
RTX Spark specifications
Computer makers opened preorders on Wednesday for laptops powered by NVIDIA RTX Spark, with hardware arriving on October 16. Compact desktop models designed for continuous 24-hour operation will go on sale in November. Hardware partners include Acer, ASUS, Dell, HP, Lenovo, MSI, Gigabyte, and Microsoft, which built its Surface Laptop Ultra around the chip.
The RTX Spark platform links an NVIDIA Blackwell graphics processor with up to 6,144 cores to an NVIDIA Grace processor with up to 20 cores across a 600 gigabytes-per-second bus. The system provides one petaflop of FP4 compute and up to 128 gigabytes of unified memory. According to NVIDIA, the setup can process the 125-billion-parameter Qwen 3.8 Flash Next model entirely locally without transmitting data to cloud services.
Creative workflows on RTX Spark use fifth-generation Tensor Cores with NVFP4 support, AV1 hardware acceleration, and 4:2:2 video processing. The chip also supports DLSS 5, Reflex, and G-SYNC to run games at 1440p resolutions at over 100 frames per second.
Deskside supercomputing
The event previewed NVIDIA DGX Station for Windows, marking the first time the deskside supercomputer has supported Microsoft's operating system. The system previously ran exclusively on Linux, requiring enterprise developers to manage two separate computing environments for AI research and Windows productivity software.
NVIDIA confirmed the workstation runs on the GB300 Grace Blackwell Ultra Desktop Superchip, providing 748 gigabytes of coherent memory and up to 20 petaflops of FP4 AI performance. Davuluri stated the hardware can run models scaling up to one trillion parameters locally, replacing rented cloud clusters while retaining access to Linux software toolchains through the Windows Subsystem for Linux.
