Nvidia N1X superchip laptops run 74B-parameter Nemotron models locally with 128GB memory

Nvidia launched RTX Spark laptops on Friday, designed to execute AI agents directly on the device. According to The Next Web, these machines feature the N1X superchip, which pairs a Blackwell RTX graphics chip with a Grace processor. This architecture provides up to 128GB of shared memory and one petaflop of AI performance, enabling large language models to run locally.

Hardware architecture

The design stems from a collaboration between Nvidia and MediaTek on the processor, as reported by The Next Web. Jensen Huang, Nvidia’s founder and chief executive, stated that the platform combines NVIDIA AI, RTX graphics and CUDA with Microsoft Windows. This integration aims to allow personal AI agents to function on the PC itself for creating and gaming.

Control mechanisms

Security controls for these local agents include Nvidia’s OpenShell runtime and Microsoft Execution Containers, according to The Next Web. These tools let users dictate what actions agents are permitted to perform. GitHub Copilot and OpenClaw already support Microsoft’s containers, while Hermes will add OpenShell controls by the end of October, Nvidia said in a blog post.

Model specifications

The Nemotron 3.5 Bolt model, released on Thursday, has 74 billion parameters with nine billion active at any time, fitting into 24GB of memory. The Next Web reports it runs on supported RTX Spark configurations. A smaller 4GB version for voice assistants follows on 23 October. The laptops themselves use the N1X chip, available in 20-core or 18-core variants, according to The Verge.

Operational guidance

• Use Microsoft Execution Containers to restrict agent permissions.
• Employ Nvidia’s OpenShell runtime to control agent behaviour.
• Utilise the Switchyard model router to split work between local and cloud models.
• Ensure agents are installed from the first start-up interface, avoiding command line installations, as Nvidia advised.

What it means for organisations

IT leadership should assess whether local AI processing aligns with data sovereignty policies. Teams usually evaluate if sensitive data remains on-device rather than traversing external networks. Procurement must verify that vendor-supported agents comply with internal security standards. Staff need guidance on managing local model weights and understanding which tasks run locally versus in the cloud. Risk managers should review access controls for agent permissions.

Background: AI agents

AI agents matter for security because they move beyond generating text to executing code and accessing systems. This autonomy means a single compromised prompt can trigger cascading failures across your infrastructure, requiring new controls for intent verification and action boundaries rather than just content filtering.

Read the full guide: Why AI Agents Matter for Security: Autonomy Risks

What to do and how to stay safe: Nvidia

• Verify which AI agents are permitted to run locally on your network.
• Check if your devices support the required memory configurations for local models.
• Monitor agent permissions using available containerisation tools.
• Review data handling policies for local AI processing workflows.

General security guidance from the TechSec Observer newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Can RTX Spark laptops run AI without an internet connection?

Yes, they are built to run AI agents and large language models locally on the device itself.

Which AI agents are available for RTX Spark laptops?

Users can install agents such as GitHub Copilot, Hermes, OpenClaw and Perplexity from the first start-up.

What is the memory capacity of the RTX Spark N1X chip?

The N1X superchip has up to 128GB of shared memory, according to The Next Web.