Nvidia is betting that the era of cloud-dependent AI is ending for physical machines. The company’s new Jetson Orin Nano 2 packs 78 trillion operations per second into a board that costs $249 and draws as little as 15 watts — hardware aimed squarely at developers building drones, home robots, and industrial vision systems that need to think on their own.
The pitch hinges on a shift Nvidia says happened quietly over the past year: small and medium AI models now match the accuracy that only frontier-scale models could deliver twelve months ago. Deepu Talla, Nvidia’s VP of Robotics and Edge AI, framed the launch as putting ‘that breakthrough within reach of millions of developers, delivering the performance and energy efficiency needed for real-time reasoning at the edge.’ The board doubles the inference performance of the previous Jetson Orin Nano Super while drawing 40 percent less power in its 15-watt mode — a meaningful efficiency jump for battery-constrained devices like delivery drones.
Early adopters give a sense of where this is heading. Wing, Alphabet’s drone delivery subsidiary, already flies the Orin Nano Super across its fleet and plans to evaluate the new board for faster real-time perception. Matic Robots is adopting it for home cleaning bots that will add conversational AI, gesture detection, and semantic mapping of living spaces. Dinuka Abeywardena, Wing’s Head of Perception, said the company is exploring the hardware for ‘more responsive, energy-efficient drones’ — a signal that drone delivery economics depend heavily on squeezing more intelligence per watt.
The broader ecosystem play matters as much as the silicon. Nvidia claims over three million developers already build on its Jetson stack, and a long list of hardware partners — AAEON, Advantech, Seeed Studio, and roughly two dozen others — are shipping carrier boards and reference designs. That’s the same playbook Nvidia used to dominate data center AI: own the software stack, let partners handle the physical integration, and make the entry point cheap enough that developers default to your platform. The open question is whether edge deployments actually scale this time or remain a niche compared to the cloud workloads that made Nvidia a trillion-dollar company. The hardware is finally capable. Whether the market follows is less certain.
