HeadsUpAI

NVIDIA launches RTX Spark to run autonomous AI agents locally on Windows

The NVIDIA RTX Spark superchip integrates a Blackwell GPU and a 20-core Grace CPU to power a new class of AI-native Windows PCs. With 128GB of unified memory—a shared pool of high-speed RAM accessible by both processors—the chip runs 120-billion-parameter models and autonomous agents entirely on-device. This hardware shift moves AI from a reactive tool to a persistent teammate.
AI Performance
1 petaflop
Unified Memory
Up to 128GB
GPU Specs
6,144 CUDA cores
CPU Specs
20-core NVIDIA Grace
Availability
Fall 2026

The platform addresses privacy and latency barriers that previously kept autonomous agents confined to the cloud. By partnering with Microsoft on native security primitives and the NVIDIA OpenShell runtime, the system enables agents to navigate local files and cross-app workflows securely. This mirrors the local-first trajectory seen in the Perplexity Personal Computer but brings it to the broader Windows ecosystem.

RTX Spark laptops from ASUS, Dell, and Lenovo arrive this fall. The hardware is optimized for the Nemotron 3 Nano Omni model to handle multimodal tasks like private coding and research. Developers can also scale these agentic workflows to enterprise desksides using the NVIDIA DGX Station for Windows, which uses the same underlying architecture.

View the full update on nvidianews.nvidia.com
NVIDIA Newsroom
NVIDIA Newsroom
@nvidianewsroom
X

News at #NVIDIAGTC Taipei: @NVIDIARTXSpark debuts at GTC Taipei as a new beginning for personal computers. The 1-petaflop superchip brings the full CUDA and RTX ecosystem, Windows-native agents and NVIDIA AI and graphics technology to slim Windows laptops and compact desktops built for creators, AI developers and gamers.

32retweets222likes
View on X

Still wondering? A few quick answers below.

NVIDIA RTX Spark is a new superchip designed to power Windows PCs built for personal AI agents. It combines the Blackwell GPU architecture with a Grace CPU and up to 128GB of unified memory to deliver 1 petaflop of AI performance for local model execution and autonomous tasks.

The platform uses a collaboration between NVIDIA and Microsoft to provide a secure environment for agents. This includes new Windows security primitives for identity and containment, alongside the NVIDIA OpenShell runtime, which allows users to set policies for what local agents can access and execute.

RTX Spark-powered laptops and compact desktops will be available starting in fall 2026. Initial manufacturing partners include ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI. Additional models from Acer and GIGABYTE are expected to follow the initial launch window.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update