NVIDIA's new Jetson Orin Nano 2 chip aims to put frontier AI inside drones and home robots
The tiny computer uses 40% less power than its predecessor while doubling AI processing speed, and NVIDIA says developers will be able to run frontier-grade intelligence directly inside physical machines.
Key points
- NVIDIA has announced the Jetson Orin Nano 2, a compact AI computer aimed at robots, drones and smart cameras.
- The module delivers 78 trillion AI operations per second while using 40% less power than its predecessor at equivalent performance.
- It doubles the AI inference speed (the rate at which a model produces a decision) of the previous Jetson Orin Nano Super.
- Wing, the Alphabet drone-delivery subsidiary, and home-robot maker Matic Robots are among the first partners evaluating the hardware.
- The module and developer kit arrive in the first half of 2027.
A computer roughly the size of a credit-card reader could soon give a delivery drone the reasoning ability of a top-tier AI chatbot. That's the pitch behind the Jetson Orin Nano 2, NVIDIA's latest entry-level edge AI module, where "edge AI" means intelligence that runs on the device rather than on a distant server.
What exactly did NVIDIA announce?
The Jetson Orin Nano 2 is a self-contained computing module designed to sit inside drones and cameras. NVIDIA says it performs 78 trillion AI operations per second, packs 8 gigabytes of memory, runs an eight-core Arm processor, and draws 15 watts in its most efficient mode. At that power level it matches the peak performance of the previous model while consuming 40% less electricity, NVIDIA vice-president Deepu Talla told The Robot Report.
The headline number is doubled inference speed. Inference is the moment an AI model takes in information and produces a decision. Faster inference means a drone can spot an obstacle sooner, or a cleaning robot can identify a chair leg and steer around it before it makes contact.
Why does the power saving matter?
Battery life is everything for a flying drone or a cordless floor robot. Cutting consumption by 40% for the same workload could extend operating time between charges, or let engineers shrink the battery and shed weight. Both matter.
The new chip can also run large language models (the technology behind text-based assistants like ChatGPT) and vision language models, which let machines reason about what a camera sees. Talla noted that a year ago those models demanded frontier-scale server hardware running hundreds of billions of parameters. They now fit a palm-sized module.
| Specification | Jetson Orin Nano Super | Jetson Orin Nano 2 |
|---|---|---|
| AI performance (TOPS) | ~40 | 78 |
| Memory | 8 GB | 8 GB |
| Power (efficient mode) | ~15 W | 15 W |
| Inference speed vs. Predecessor | baseline | 2x faster |
| Availability | now | H1 2027 |
Who is already building with it?
Wing, the drone-delivery arm of Alphabet, plans to evaluate the chip for faster, safer package drops to residential addresses. It's worth noting Wing already runs NVIDIA's Jetson Orin Nano Super in its current fleet. AI2Day first covered Wing's delivery ambitions on 22 August 2026 when Amazon announced plans for nearly 500 US cities, a reminder of how quickly drone AI requirements are escalating.
Matic Robots wants to use Orin Nano 2 in home cleaning machines capable of conversation, gesture recognition and precise room mapping. NVIDIA also cites Cognex and Doosan Bobcat as partners evaluating the hardware.
More than 10,000 companies are currently shipping or developing products on NVIDIA's Jetson family, and over 3 million developers use its robotics software stack. The module drops into the same physical slot as previous Jetson Orin hardware, so existing designs need no new circuit boards.
What does this mean for ordinary people?
Not much immediately: hardware arrives in early 2027 at the soonest, and consumer products using it come later. But the direction's clear. Smarter drones could make same-day delivery more dependable. Home robots could become genuinely useful rather than frustrating.
The step that actually matters here is a timing one: AI capable of real conversation and visual understanding is moving off server farms and into physical devices far sooner than most robotics engineers expected even twelve months ago. Whether 2027 availability leaves room for competitors to close the gap on NVIDIA's software stack is the thing to watch.
Common questions
Will this affect the price of robots or drones for consumers?
NVIDIA hasn't released a retail price for the Orin Nano 2 module. Entry-level Jetson hardware has historically sold in the low hundreds of dollars, but the cost of a finished consumer product depends heavily on the manufacturer.
Does the robot need an internet connection to use this AI?
No. That's the whole point of edge AI: the model runs on the device. A drone using Orin Nano 2 can reason about its surroundings without a mobile signal, which matters for reliability and privacy alike.
When can developers actually get one?
NVIDIA has confirmed availability in the first half of 2027 for both the bare module and the developer kit, a board that lets engineers prototype new products.
