Nvidia Unveils RTX Spark ARM Chip and Vera Rubin AI Infrastructure at Computex 2026

​Nvidia has introduced its new ARM-based chip, the RTX Spark, marking a significant shift for AI PCs. During the GTC Keynote at Computex in Taipei, CEO Jensen Huang highlighted the role of agentic AI in transforming Windows PCs. He stated that AI is now a profit generator, and agentic AI will automate routine tasks for users.

Key Highlights

• Nvidia introduced the ARM-based RTX Spark chip for Windows PCs at Computex 2026.
• RTX Spark supports up to 128GB RAM, 6,144 GPU cores, and 20 CPU cores.
• Microsoft Surface Laptop Ultra will feature RTX Spark and launch by end of 2026.
• Nvidia unveiled Vera Rubin, a multi-rack supercomputer for agentic AI infrastructure.
• Nemotron 3 Ultra AI model is five times faster and 30 percent cheaper than other models.

RTX Spark Chip and Features

The Nvidia RTX Spark is the foundation of Nvidia's agentic platform for Windows. The chip supports up to 128GB of LPDDR5X RAM, 6,144 GPU cores, and 20 CPU cores. Less powerful versions will offer as little as 16GB of RAM. The RTX Spark is similar to the GB10 chip used in the DGX Spark system launched last year.

Huang claimed the RTX Spark can run all existing software and AI agents locally through Nvidia's OpenShell runtime on Windows PCs. This enables users to automate tasks directly on their devices. For creators, the chip can render 90GB-plus 3D scenes, edit 12K video, and generate AI videos in 4K resolution.

Nvidia announced that Adobe is optimizing Photoshop and Premiere for RTX Spark. Features like Firefly-powered Generative Fill in Photoshop and Generative Extend in Premiere will be accelerated on this platform. Nvidia states the RTX Spark can deliver up to two times faster AI, editing, coloring, and effects across creative workflows.

The RTX Spark integrates technologies such as CUDA, RTX, DLSS, TensorRT, OptiX, Reflex, and G-SYNC. It will appear in slim Windows laptops and compact desktop PCs, promising all-day battery life.

Device Availability and Partnerships

RTX Spark-powered laptops will range from 14-inch to 16-inch sizes, with thickness as slim as 14 millimeters and weight as low as 1.3 kilograms. Devices will feature aluminum chassis and tandem OLED displays with G-SYNC support. Nvidia expects these laptops and PCs to be available this fall from Asus, Dell, HP, Lenovo, Microsoft, and MSI.

Microsoft has announced the Surface Laptop Ultra, powered by the RTX Spark. This model will offer up to 128GB of LPDDR5X RAM and a 15-inch mini-LED display with 2,000 nits peak brightness. The Surface Laptop Ultra will launch before the end of 2026.

The RTX Spark launch signals Nvidia's renewed focus on PC chips, following its earlier attempt with the Microsoft Surface RT in 2012. The timing aligns with the likely end of Qualcomm's exclusive Windows 11 ARM chip license.

Vera Rubin AI Infrastructure

Nvidia also introduced Vera Rubin, the first multi-rack podscale supercomputer for agentic AI. Huang explained that the economics of CPUs for AI agents differ from those for humans. To address this, Nvidia developed new Vera CPUs with a unique architecture, designed specifically for agents.

Early adopters of Vera CPUs include OpenAI, Anthropic, and SpaceX. Vera Rubin is described as Nvidia's most ambitious project, providing full-stack AI infrastructure for enterprise workflows. The system uses Nvidia's DSX platform, which enables more GPU deployment in data centers while reducing energy and water consumption.

Vera Rubin is now in full production and is twice as large as the previous Grace Blackwell computing platform. Assembly time for a Grace Blackwell stack has been reduced from two hours to five minutes. Microsoft and Dell already have operational Vera Rubin stacks.

Additional AI Announcements

Nvidia unveiled a new agent toolkit for enterprise AI, including models and agents like Claude Code and Codex, the OpenShell runtime, CUDA-X tools, and harnesses for privacy and security. The company also introduced Nemotron 3 Ultra, an open AI model that is five times faster and 30 percent cheaper than other efficient models. Partners in open model development include Sarvam and Perplexity.

Other announcements included Cosmos 3, an open multimodal platform for physical AI, and Isaac Groot for humanoid platforms. Nvidia also presented Alpamayo, the first reasoning autonomous vehicle model that can communicate verbally while driving.