Nvidia and Microsoft unveil RTX Spark PCs built for local AI agents
Arm-based superchip with 1 petaflop AI performance runs 120B-parameter models locally.
At Computex 2026, Nvidia and Microsoft unveiled RTX Spark, a new category of Windows PCs purpose-built for running AI agents locally. The hardware combines Nvidia's Arm-based Grace CPU (20 cores, co-developed with MediaTek) with a Blackwell RTX GPU, supporting up to 128GB of unified memory and 1 petaflop of AI compute. This architecture lets users run large models (up to 120 billion parameters), edit 12K video, render complex 3D scenes, and power high-end gaming—all without sending data to the cloud. Microsoft CEO Satya Nadella called it 'unmetered intelligence for every desk,' while Nvidia's OpenShell tool gives users granular control over what agents can access and share.
First RTX Spark systems are expected this fall from Microsoft, Dell, HP, Lenovo, ASUS, and MSI, with Acer and GIGABYTE following later. Pricing hasn't been disclosed but Forbes notes these will be premium devices initially aimed at creators, AI developers, and gamers. Adobe is already optimizing Photoshop and Premiere for RTX Spark, claiming up to 2x faster AI and graphics performance in select workflows. This move marks Nvidia's deeper push into the PC market long dominated by Intel, AMD, Qualcomm, and Apple, betting that users want more AI horsepower on local machines rather than relying on cloud services.
- RTX Spark uses an Arm-based Nvidia Grace CPU (20 cores) with a Blackwell RTX GPU, delivering 1 petaflop AI performance and up to 128GB unified memory.
- Systems can run 120-billion-parameter AI models locally, edit 12K video, and support high-end gaming, targeting creators and developers.
- First devices from Microsoft, Dell, HP, Lenovo, ASUS, and MSI launch fall 2026; pricing is premium but undisclosed.
Why It Matters
Local AI agents on RTX Spark PCs reduce cloud dependency, enabling faster, private task automation for professionals.