NVIDIA RTX Spark Is the Company's First All-in-One AI Superchip for Windows PCs

NVIDIA's Computex 2026 keynote opened with something the PC industry has not seen before: a single chip integrating a full Blackwell-class GPU and a 20-core custom CPU, designed from the ground up to run personal AI agents locally on a Windows laptop. Jensen Huang called it RTX Spark, and it represents NVIDIA's most direct move yet into the consumer PC market -- not just as a GPU supplier, but as the architect of the whole silicon platform.

What RTX Spark Actually Is

RTX Spark is a system-on-a-chip that combines a Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores supporting FP4 precision, a 20-core NVIDIA Grace CPU, and 128GB of unified LPDDR5X memory. The two compute units are connected via NVIDIA NVLink-C2C rather than a standard PCIe bus, giving the GPU direct access to the full memory pool at high bandwidth. The chip is built on TSMC 3-nanometer process and delivers up to 1 petaflop of AI compute performance.

Running a 70-billion-parameter model locally -- something that currently requires a dedicated workstation -- will be straightforward on RTX Spark hardware.

What It Enables

NVIDIA is positioning RTX Spark not primarily as a gaming chip but as the foundation for always-on personal AI agents. The company demonstrated 2x inference performance on leading agentic models, using the new NVIDIA OpenShell runtime for Windows. Adobe and Blender are both rebuilding their applications specifically for RTX Spark. DLSS 4.5 Ray Reconstruction is also coming to RTX Spark devices.

Hardware and Availability

RTX Spark will power slim Windows laptops engineered to be as thin as 14 millimeters and as light as 3 pounds, in 14- to 16-inch sizes, alongside compact ultra-efficient desktop PCs. ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI will have devices at launch this fall, with Acer and GIGABYTE to follow.

NVIDIA also announced the DGX Station for Windows -- a deskside workstation designed to run frontier AI models with up to 1 trillion parameters locally, targeting enterprise and research teams.

The Bigger Picture

RTX Spark marks a significant strategic shift. NVIDIA has spent thirty years supplying GPUs to system builders. This is the first time it has designed and branded the entire compute platform -- CPU, GPU, memory, and interconnect -- for a consumer product category. The move mirrors what Apple did with the M-series chips: consolidate control of the full silicon stack to optimize for a specific use case, in this case AI inference on the edge.

NVIDIA has also committed to at least two future generations: Rubin with LPDDR6 memory, and Rosa Feynman to follow. Devices are expected this fall at premium price points.