NVIDIA has unveiled the RTX Spark, a flagship Windows “superchip” built hand-in-hand with Microsoft, in a clear move to put powerful, local AI computing directly into the hands of Microsoft’s vast user base.
Announced in a fireside chat between NVIDIA’s Jensen Huang and Microsoft’s Satya Nadella, NVIDIA cast the RTX Spark as “a new beginning for Windows PCs,” fusing AI development, content creation and gaming onto a single machine.
The specifications are formidable. The chip pairs a Blackwell RTX GPU of up to 6,144 CUDA cores with a 20-core NVIDIA Grace CPU, a 600GB/s NVLink-C2C interconnect, up to 128GB of unified memory and up to one petaflop of FP4 AI compute.
Crucially, the pitch is local AI. Rather than leaning on the cloud, the RTX Spark runs models directly on the device, cutting latency, keeping data on the machine and enabling offline use.
The Microsoft tie is the heart of the story. Microsoft’s launch device is the new Surface Laptop Ultra, billed as its most powerful Surface yet, a Copilot+ PC with Windows 11 AI experiences baked in. Windows chief Pavan Davuluri said the hardware lets users run models that “don’t fit on a traditional machine.”
Microsoft is also embedding OS-level infrastructure, Microsoft Execution Containers (MXC), which, with Agent 365 and Microsoft Security, give IT teams visibility and governance over the background AI agents increasingly running on Windows.
The ecosystem is widening, with HP integrating the RTX Spark into upcoming Copilot+ PCs and MediaTek co-designing the custom CPU. NVIDIA separately previewed a DGX Station for Windows, bringing GB300-class supercomputing to the platform.
Ultimately, this is a strategic alliance to make Windows the home of always-on personal AI, positioning the PC itself, not the cloud, as the next battleground for AI assistants.
