Nvidia and Microsoft unveil RTX Spark AI PC chip for local agents
The chip packs 6,144 CUDA cores and 128GB memory, but is it future-proof?
At Computex 2025, Nvidia CEO Jensen Huang announced the RTX Spark, a new PC processor designed specifically for AI workloads. Built in collaboration with Microsoft, the chip (also called the N1X) features 20 Arm CPU cores, a Blackwell GPU with 6,144 CUDA cores, 128GB of LPDDR5X RAM, and up to 300 GB/s of memory bandwidth. Nvidia claims it can run 120-billion-parameter models with context lengths up to 1 million tokens, making it suitable for autonomous AI agents that operate locally on Windows PCs. The chip will debut this fall on devices from Microsoft, Dell, HP, ASUS, Lenovo, and MSI.
Despite the impressive specs, the RTX Spark faces skepticism from analysts who point to its design trade-offs. The chip allocates significant die space to GPU cores for local inference, but these are inherently limited by memory bandwidth compared to cloud GPUs. In the current agentic era, where strong CPU performance is critical for orchestrating multi-step tasks and calling cloud APIs, the Spark’s balance may be suboptimal. Critics argue it excels at 2023-style chatbots but may struggle with the demands of 2026 agent workflows. The Windows-on-ARM ecosystem also remains a compatibility hurdle. While Nvidia positions the chip as a breakthrough for edge AI, the real test will be real-world performance and software support.
- RTX Spark packs 20 Arm CPU cores, 6,144 CUDA cores, and 128GB LPDDR5X memory with 300 GB/s bandwidth
- Supports 120-billion-parameter models and 1M-token context windows for local AI agents
- Available this fall from Microsoft, Dell, HP, ASUS, Lenovo, and MSI; analysts question CPU vs GPU balance for agentic workloads
Why It Matters
Nvidia enters the PC processor market, but its AI-centric design may struggle in the agentic era where CPU and cloud matter most.