NVIDIA has unveiled the Jetson Orin Nano 2, a compact edge AI computer that delivers twice the inference performance of its predecessor while consuming 40% less power in 15-watt mode, making local AI processing practical for battery-powered robots and autonomous machines. The system delivers 78 trillion operations per second (TOPS) of AI compute with 8 gigabytes of memory and an 8-core Arm CPU, targeting developers building robots, delivery drones, smart cameras, and other physical AI systems that need to process information locally rather than sending data to cloud servers.

Why Does Local AI Processing Matter for Robots and Drones?

Robots and autonomous machines often cannot afford the delay of sending sensor data to a remote server and waiting for a response. When a delivery drone needs to avoid an obstacle or a warehouse robot must navigate around workers, millisecond-level decisions matter. Edge AI, which runs artificial intelligence models directly on the device itself, eliminates that cloud round-trip latency. The Jetson Orin Nano 2 enables machines to see, understand, and react to their surroundings in real time without relying on constant internet connectivity.

For applications in remote locations or areas with unreliable connectivity, this shift is transformative. Mining inspections, agricultural monitoring, warehouse automation, security systems, and infrastructure monitoring all benefit from machines that can process camera and sensor data locally. The device doesn't need to upload every frame or sensor reading to a cloud server, reducing bandwidth demands and enabling faster responses when connectivity is inconsistent or unavailable.

What AI Models Can Run on This Compact Hardware?

The real breakthrough is what developers can now deploy on hardware this small. The Jetson Orin Nano 2 supports large language models (LLMs) and vision-language models, which are AI systems trained to understand both images and text, designed specifically for memory-efficient local inference. NVIDIA lists several models developers can use, including Cosmos, Nemotron, Gemma 4, and Qwen 3. These are not toy models; they represent the same class of generative AI that powers modern chatbots and image-understanding systems, but optimized to fit within the device's 8 gigabyte memory constraint.

NVIDIA's Cosmos 3 Edge model exemplifies this approach. It is designed to help robots and intelligent cameras reason about visual information directly on edge hardware, enabling capabilities such as object detection, scene understanding, and spatial reasoning without cloud dependency. This opens possibilities for machines to perform tasks like mapping, conversational interaction, gesture detection, and autonomous navigation entirely on-device.

How to Deploy Edge AI on Robotics Hardware

• Choose a Supported Model: Select from NVIDIA-optimized models like Cosmos, Nemotron, Gemma 4, or Qwen 3 that are designed to run efficiently on edge devices with limited memory and power budgets.
• Leverage NVIDIA's Robotics Stack: Use NVIDIA's robotics tools and software ecosystem, which already supports more than 3 million developers building on the platform, reducing development time and complexity.
• Optimize for Power Efficiency: Configure your application to run in low-power modes; the Jetson Orin Nano 2 can deliver the same performance as its predecessor while consuming 40% less power, extending battery life for mobile robots and drones.
• Test with Real-World Scenarios: Deploy your model on actual hardware in target environments, such as warehouses or remote sites, to validate latency, accuracy, and power consumption before full-scale rollout.

Who Is Already Building With This Technology?

Several major companies are already adopting or exploring the Jetson Orin Nano 2. Cognex, a leader in machine vision, Doosan Bobcat, known for construction equipment, and Matic, which builds home robots, are among the early adopters. Alphabet-owned Wing, which operates a delivery drone service, plans to evaluate the computer for its autonomous aircraft. Matic specifically intends to use the platform for home robots that require mapping, conversational abilities, gesture recognition, and autonomous navigation capabilities.

This adoption signals confidence in the hardware's capabilities and suggests a broader industry shift toward embedding AI directly into physical machines rather than treating them as remote-controlled cloud clients. NVIDIA is also working with LG on a humanoid robot powered by NVIDIA technology, while its higher-end Jetson Thor systems target more demanding physical AI workloads, creating a product ladder that serves different performance and cost requirements.

What Is NVIDIA's Broader Strategy Beyond Selling Chips?

NVIDIA's approach extends far beyond simply selling processors. The company is building an integrated ecosystem that connects AI training, model optimization, robotics software tools, and hardware deployment. By offering models, development frameworks, and hardware together, NVIDIA aims to reduce friction for developers entering the robotics and edge AI space. This ecosystem approach mirrors successful strategies in other technology markets, where the company that controls the full stack often captures the most developer mindshare and long-term revenue.

The performance improvements in the Jetson Orin Nano 2 reflect this strategy. The doubled inference performance comes from improved Tensor Cores, which are specialized processors optimized for AI math, and higher memory bandwidth, which allows data to flow faster between the processor and memory. NVIDIA maintained the same compact form factor as the previous model, meaning developers can upgrade without redesigning their hardware enclosures.

When Will Developers Be Able to Buy This Hardware?

NVIDIA expects the Jetson Orin Nano 2 module and developer kit to become available during the first half of 2027. The company has not yet announced official pricing, which means the true entry-level value proposition will become clearer once pricing and regional availability details emerge. For developers in markets such as South Africa and other regions outside major tech hubs, availability and local support will be critical factors determining adoption.

The timing matters because it gives developers several months to plan projects and prepare their teams. Universities, startups, and smaller companies can begin experimenting with the previous generation of Jetson hardware while waiting for the Orin Nano 2 to launch, building expertise that will accelerate deployment once the new hardware arrives. The question now is whether more affordable, efficient edge AI hardware will be enough to trigger a wave of locally built robots and intelligent machines from smaller teams and emerging markets.