NVIDIA targets entry-level physical AI, Jetson Orin Nano 2 doubles inference performance and reduces power consumption by 40%.
```
Nvidia is further extending the capabilities of artificial intelligence (AI) from data centers to real-world devices such as robots.
On Tuesday the 25th Eastern Time, Nvidia announced the release of its new generation entry-level robotics computing platform, Jetson Orin Nano 2, targeting edge AI applications such as robots and drones that are sensitive to cost, power consumption, and real-time computing capabilities.
Nvidia stated that while maintaining the same compact size, Jetson Orin Nano 2’s inference performance is double that of the previous generation Jetson Orin Nano Super; in 15-watt power mode, it can achieve the same performance as the previous generation with 40% lower power consumption.
In terms of hardware, Jetson Orin Nano 2 is equipped with an 8-core processor, 8GB of memory, and delivers AI computing power up to 780 trillion operations per second. Nvidia said the new platform enhances AI inference ability by upgrading tensor cores and increasing memory bandwidth, while continuing to support Nvidia’s software ecosystem for robots and edge AI. It can run memory-efficient large language models and vision-language models, including Cosmos, Nemotron, Gemma 4, and Qwen 3.
The significance of this product launch is not just the release of a more powerful compact AI computer. As AI models become more compact and inference efficiency continues to improve, tasks such as visual understanding, language interaction, and real-time decision-making, which previously relied mainly on the cloud, are gradually migrating to terminal devices like robots and drones. Nvidia hopes Jetson Orin Nano 2 will further lower the threshold for physical AI to enter entry-level devices.
Miniaturized AI Models Accelerate Adoption, Demand for On-device Robot Intelligence Heats Up
The launch of Jetson Orin Nano 2 comes as the AI industry extends from the cloud to endpoints.
Unlike data center AI, robots and drones have much stricter requirements for computing platforms: the device size is limited, usually cannot carry high-power processors, and they must perform continuous visual perception, environmental understanding, and decision-making during movement. Therefore, how much AI computing power can be delivered per unit of power is often more important than simply pursuing peak performance.
Nvidia said Jetson Orin Nano 2 can deliver the same performance as the previous generation with 40% lower power consumption in 15-watt mode. For battery-powered robots and drones, this means manufacturers can reduce energy consumption for the same performance, or use the saved power consumption for enhanced battery life, heat dissipation, or other sensor configurations.
At the same time, the rapidly improving capabilities of small AI models are also changing the boundaries of on-device AI applications.
Tara, head of Nvidia’s robotics and edge AI business, said that nowadays the accuracy of some small and medium-sized frontier models has reached last year's level of large frontier models, creating conditions for real-time intelligence on edge devices.
In the past, edge AI mainly handled tasks such as object detection and image classification; as model capabilities increase, end devices are gaining more complex language understanding, visual reasoning, and multimodal interaction capabilities.
This also allows robots to evolve from “able to perceive the environment” to “able to understand the environment and take action.”
From Visual Recognition to Multimodal Understanding, Real-Time Decision-Making Becomes Key
Nvidia is positioning the Jetson platform as the crucial computing foundation for robots to acquire “physical intelligence.”
Currently, the information processing needs of robots are no longer limited to images captured by cameras, but also include natural language commands, speech, spatial information, and data from various sensors. For robots, quickly integrating this information locally and responding with low latency is essential for autonomous operation.
Jetson Orin Nano 2 thus targets several typical scenarios, including home robots, vision AI systems, delivery drones, and inspection drones.
For example, Google’s drone delivery company Wing has already adopted Jetson Orin Nano Super and Nvidia's software stack in its delivery drones, and plans to evaluate Jetson Orin Nano 2. Nvidia said higher performance and efficiency could help Wing further improve real-time AI perception and inference capabilities of its drones.
Home robotics company Matic plans to use Jetson Orin Nano 2 to enhance robots’ natural language interaction, gesture recognition, spatial mapping, and semantic understanding capabilities, enabling robots to autonomously complete tasks such as cleaning.
In addition, Nvidia also demonstrated the case of Jetson Orin Nano 2 running the Reachy Mini robot. Nvidia said the platform can simultaneously process language models, speech models, and real-time vision AI tasks, showing the ability of entry-level edge platforms to handle multiple AI models.
For robotics manufacturers, this capability means robots no longer just follow preset programs to perform fixed actions, but can understand the environment through visual information and language commands, and then adjust actions dynamically.
This is also why Nvidia has continuously emphasized “physical AI” in recent years: artificial intelligence is moving from software services on the screen into the real world, and will ultimately drive physical devices such as robots, cars, and drones.
Expanding Developers and Partners, Nvidia Continues to Strengthen Robotics Ecosystem
In addition to chip performance, software ecosystem has always been an important competitive advantage of the Nvidia Jetson platform.
Nvidia said more than three million developers are developing based on its robotics technology stack. With the launch of Jetson Orin Nano 2, Nvidia is further expanding the software, hardware, and developer ecosystem for robotic AI. Cognex, and Matic
On the software side, Jetson Orin Nano 2 continues to support Nvidia’s robotics development platform, and can run memory-optimized large language models and vision-language models for edge devices, including Nvidia’s Cosmos, Nemotron, as well as Gemma 4, Qwen 3, and other models.
This means developers can deploy language understanding, visual understanding, and reasoning capabilities directly to robots with relatively limited hardware resources, without relying entirely on cloud servers.
The hardware ecosystem is also expanding in sync.
Nvidia revealed that partners including Aaeon, ADLINK, Advantech, Aetina, Antmicro, Aptiv, Connect Tech, and Seeed Studio are developing carrier boards, hardware systems, AI software, and reference designs based on Jetson Orin Nano 2, helping robotics manufacturers shorten product development and market launch cycles.
Nvidia also said that the first batch of companies to adopt or explore Jetson Orin Nano 2 include Cognex, Doosan Bobcat, Matic, etc., and Wing also plans to evaluate the platform.
In the longer term, Nvidia is aiming not just to establish a robotics chip business, but to build a complete ecosystem covering AI models, software tools, computing platforms, and robotics terminals.
This model is somewhat similar to the competitive advantages Nvidia has built in the data center AI sector: synergy through hardware performance and software ecosystem, further boosting platform attractiveness through a large developer base.
New Product Expected to Hit the Market in First Half Next Year, Commercial Contribution Yet to Materialize
However, it is worth noting that Jetson Orin Nano 2 is still in the announcement stage, and it will take some time before it actually hits the market.
Nvidia stated that the Jetson Orin Nano 2 module and development kit are expected to launch in the first half of 2027. Therefore, the new product’s direct contribution to Nvidia’s near-term performance remains to be seen.
In terms of product planning, Nvidia is gradually improving its product system for robotics and physical AI: on one hand, meeting complex application needs with high-performance robotics computing platforms; on the other, with entry-level platforms like Jetson Orin Nano 2, bringing AI inference deeper into cost- and power-sensitive terminal devices.
If in the future, compact language models and vision-language models continue to advance rapidly while inference costs continue to fall, the demand for on-device AI computing in robots, drones, and similar devices may further expand.
For Nvidia, this means the market it is targeting is gradually expanding from traditional data center AI computing to become the “brains” behind real-world devices such as robots and drones. Jetson Orin Nano 2 is thus an important step in further expanding into a broader end-device market.
Risk Warning and DisclaimerThe market has risks, so invest cautiously. This article does not constitute personal investment advice, nor does it take into account the specific investment objectives, financial situation, or needs of any individual user. Users should evaluate whether the opinions, views, or conclusions in this article are suitable for their particular circumstances. Investing based on this is at your own risk. ```