Small AI models have gotten good enough to match last year's biggest ones. Nvidia's newest chip is built to run them on cheap, low-power hardware inside actual robots.
Nvidia announced the Jetson Orin Nano 2 this week, a new entry-level computer built specifically for edge AI. The company says it targets robots, delivery drones, inspection drones and vision AI systems.
What's Actually New Here
Nvidia says the chip delivers twice the inference performance of its predecessor. That's the Jetson Orin Nano Super, and it comes in the same compact form factor. It claims 40% lower power consumption at matched performance in 15-watt mode.
The specs behind that claim: 78 trillion operations per second of AI compute, 8GB of memory, and an 8-core Arm CPU.
Deepu Talla is Nvidia's VP of robotics and edge AI. "Today's small and medium frontier models have reached the accuracy of last year's largest frontier models," he said. "The Jetson Orin Nano 2 computer puts that breakthrough within reach of millions of developers."
The chip supports open models including Nvidia's own Cosmos and Nemotron, alongside Google's Gemma 4 and Alibaba's Qwen 3. That cross-compatibility matters. It means developers aren't locked into a single company's model ecosystem to use the hardware.
Who's Already Building With It
Nvidia says more than 3 million developers already build on its broader robotics stack. Wing, the drone delivery subsidiary of Alphabet, currently runs the previous-generation chip in its delivery fleet. It plans to evaluate the new one.
"Wing is exploring Jetson Orin Nano 2 to give us a path to more responsive, energy-efficient drones," said Dinuka Abeywardena. He leads perception at Wing.
Home robotics startup Matic is adopting the chip for its cleaning robots. The goals include conversational AI, gesture detection and precise home mapping.
Navneet Dalal, Matic's cofounder and CEO, says the new chip changes what's possible. "Matic can run state-of-the-art AI models at the edge in a compact home robotics platform," he said.
Not Available Until 2027
One detail is easy to miss in an announcement this detailed. Nvidia isn't shipping the chip yet.
The module and developer kit are expected in the first half of 2027. Everything described here, the performance figures, the partner adoption, is based on pre-release hardware and Nvidia's own internal testing.
That gap between announcement and shipping product is standard practice in the chip industry. It's still worth keeping in mind before treating any of these specific multipliers as settled fact.
What This Means for Miami
This is the hardware layer underneath the broader physical AI race MAIN has tracked this month. That race spans Chinese world models startups and General Intuition's gaming-footage-trained robotics models. Cheaper, more efficient edge chips are what eventually let that research reach an actual product.
For Miami's robotics and drone startups specifically, this matters directly. Entry-level edge AI hardware getting twice as capable at the same price point lowers the barrier to building a real product. That's true whether or not Nvidia's specific performance claims survive independent testing once the chip actually ships.
China Cracks Down on AI Companions Over Emotional Dependence