At a Microsoft event in San Francisco, NVIDIA and Microsoft announced a co-engineering effort to bring AI agents to Windows PCs. Microsoft announced general availability of Microsoft Execution Containers (MXC), OS-level infrastructure for running agents securely and persistently in the background. The collaboration aims to make Windows a secure platform for AI agents.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
On October 7, 2026, Jensen Huang and Satya Nadella shared a stage in San Francisco to announce RTX Spark preorders, MXC general availability, and a preview of DGX Station for Windows. The core question is not whether a PC can run AI, but whether local compute, operating system governance, and developer workflows can be unified under a single stack once agents become Windows' new core scenario.
On October 7, 2026, at the Windows AI and Surface event at Dogpatch Studios in San Francisco, NVIDIA founder and CEO Jensen Huang and Microsoft CEO Satya Nadella jointly announced an AI agent solution for Windows PCs. Huang placed the launch within the company's historical narrative, calling Windows central to NVIDIA's journey; Nadella emphasized the need to make the desktop the safest place for agents to execute.
The core question of this launch can be summarized as: when agents become a new PC use case, will compute, operating system governance, and developer workflows be unified under a single stack. The answer from NVIDIA and Microsoft is a three-layer combination—RTX Spark provides local compute, MXC provides operating system-level containers, and DGX Station for Windows provides frontier compute at the enterprise deskside.
RTX Spark combines a Blackwell RTX GPU (up to 6144 cores) with up to a 20-core NVIDIA Grace CPU over a 600 GB/s interconnect, delivering 1 petaflop of FP4 AI performance and up to 128GB of unified memory. Laptops opened for preorder that day and go on sale October 16, with compact desktops arriving in November.
The Microsoft Surface Laptop Ultra is built around RTX Spark, with up to 128GB of unified memory and up to 1 petaflop of AI compute. Pavan Davuluri, Microsoft's Corporate Vice President of Windows and Devices, said this configuration can run models that traditional machines cannot fit. It should be noted that the Surface Laptop Ultra specifications are the source's statements rather than independently verified results.
Systems will be released by Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI, and Gigabyte, spanning from thin-and-light laptops to compact desktops designed for always-on agents. The compact desktops are described as dedicated local AI systems built for 24/7 operation, intended to let agents run continuously.
Microsoft announced general availability of Microsoft Execution Containers (MXC). According to the official description, this is OS-level infrastructure that lets agents run securely and persistently in the background under operating system control. Davuluri said Microsoft built these primitives directly into Windows so that agents can be protected, observed, and governed, and listed MXC, Microsoft Security, and Agent 365 together as components that unlock the potential of Windows agents.
Huang compared MXC to how Windows and DirectX changed the way applications are developed, saying it will change how agents are built and deployed. This analogy is a corporate position statement, and there is no independent security assessment of MXC's actual effect on agent security and governance.
NVIDIA previewed DGX Station for Windows, calling it the first deskside AI supercomputer for Windows enterprise desktops. It is based on the GB300 Grace Blackwell Ultra Desktop Superchip, offers 748GB of coherent memory and up to 20 petaFLOPS of FP4 AI compute, and can run models up to the trillion-parameter class locally.
Previously, DGX Station ran on Linux, so enterprise developers had to maintain two environments: Linux for heavy AI workloads and Windows for productivity tools and applications. NVIDIA says the vast majority of Fortune 500 companies use Windows as their standard environment, and this gap has cost developers time and resources—either migrate to Linux to get AI compute, or stay on Windows but with limited hardware options for heavy model development and multi-agent workloads.
DGX Station for Windows is currently only in preview, and the blog did not give a specific release date or price. When needed, the Linux AI toolchain remains accessible through WSL.
RTX Spark runs the full NVIDIA CUDA platform, so developers can migrate models, tools, and workflows between RTX Spark and DGX Station without rewriting. NVIDIA describes this continuity as the same AI stack extending from laptops to the deskside supercomputer.
NVIDIA also says RTX Spark can run Qwen 3.8 Flash Next (a 125B model, 51B n-gram), with intelligence comparable to many cloud models, without metering and with data never leaving the local device. This model capability claim comes from NVIDIA's own statements, lacks verifiable evaluation details, and should be treated as a source claim rather than a confirmed fact.
NVIDIA says gamers can run AAA titles at 1440p above 100 frames per second with DLSS 5, Reflex, and G-SYNC. Creators get fifth-generation Tensor Cores with NVFP4 support, hardware-accelerated AV1 and 4:2:2 video encoding and decoding, and DLSS and RTX ray tracing throughout the production pipeline.
These gaming and creation capabilities likewise come from NVIDIA's own statements and have not been verified by third-party benchmarks. Their role is to position RTX Spark as a general-purpose PC chip rather than dedicated hardware aimed only at AI workloads.
Almost all information from this launch comes from NVIDIA's official blog, lacking cross-validation from Microsoft, OEMs, or independent third-party sources. The performance, memory, and model capability figures for RTX Spark and DGX Station for Windows all come from NVIDIA's own statements and have not been verified by third-party benchmarks.
The actual performance, power consumption, and thermal behavior of RTX Spark and DGX Station for Windows remain unclear; the pricing, official release date, and target customer range for DGX Station for Windows have not been disclosed; and verifiable data on the actual speed, context length, and experience of running 125B-parameter-class models locally has not been provided. Configuration differences, prices, and launch timing across OEM models are likewise unclear.
The October 16 laptop launch and November desktop sale for RTX Spark are the manufacturer's announced planned dates, and actual supply has not been independently confirmed. The actual adoption rate and ecosystem impact of this solution among developers and enterprises remain to be seen.
All facts in this article come from a single NVIDIA official blog report; the event was a vendor-hosted launch, and the statements by Huang and Nadella are corporate positions, not a neutral assessment. The performance, memory, and model capability figures for RTX Spark and DGX Station for Windows are all NVIDIA's own statements and have not been verified by third-party benchmarks; the claim that Qwen 3.8 Flash Next is 'comparable to cloud models' lacks verifiable evaluation details. DGX Station for Windows is only in preview, with no release date or price announced.
This launch redefines Windows from 'a place where AI applications run' to 'a platform where agents reside,' and supports that narrative with three layers of hardware and system primitives: RTX Spark, MXC, and DGX Station for Windows. But all key performance numbers come from NVIDIA's own statements, DGX Station is still in preview with no price or date set, and the actual experience and adoption rate can only be judged once independent verification and supply information are available.
Secondary coverage
This article also covers this event; its overall headline is not the event headline.
Also covered by 1 sources