NVIDIA unveiled the Jetson Orin Nano 2 on 25 August — a compute module for robots, drones and machine-vision systems. It packs 78 TOPS, 8 GB of memory and an 8-core Arm CPU in the same form factor as its predecessor. The company promises twice the inference performance and 40 percent lower power draw in 15 W mode. The module ships in the first half of 2027.
Key takeaways
- 78 TOPS, 8 GB of memory, 8-core Arm CPU
- Twice the inference performance of the Jetson Orin Nano Super
- 15 W mode: 40% lower power draw at the same performance
- Unchanged form factor — the module drops into existing carrier boards
- Module and developer kit launch in H1 2027, price not disclosed
What Changed
The predecessor, the Jetson Orin Nano Super, offers 67 TOPS in sparse INT8, 8 GB of LPDDR5 at 102 GB/s and six Cortex-A78AE cores. The new NVIDIA module adds two Arm cores and eleven TOPS while keeping the 69.6 × 45 mm footprint and the SO-DIMM connector.
| Spec | Jetson Orin Nano Super | Jetson Orin Nano 2 |
|---|---|---|
| AI performance | 67 TOPS (sparse INT8) | 78 TOPS |
| CPU | 6× Cortex-A78AE, 1.7 GHz | 8× Arm |
| Memory | 8 GB LPDDR5, 102 GB/s | 8 GB, bandwidth undisclosed |
| Form factor | 69.6 × 45 mm, 260-pin SO-DIMM | unchanged |
| Power | 7–25 W | 15 W mode: −40% draw at equal performance |
| Availability | shipping | H1 2027 |
Why 78 TOPS Is Not Twice as Much
This is where it gets interesting. 78 TOPS against 67 TOPS is a sixteen percent increase, yet the vendor claims twice the inference performance. That is not a contradiction. TOPS measures a chip’s raw throughput, not how many frames it squeezes out of a particular model. The doubling therefore has to come from architecture and the software stack, not from the spec sheet.
Energy Rather Than Raw Power
In 15 W mode the new module is meant to deliver the same performance as its predecessor while drawing 40 percent less power. For a battery-powered robot that translates directly into runtime.
Who Is Already Taking It
NVIDIA names Cognex, Doosan Bobcat and Matic among early adopters — the last plans to put the module into home cleaning robots to handle conversation with the user and autonomous navigation. Wing, the Alphabet subsidiary, will evaluate the platform for its delivery drone fleet. More than twenty companies, including AAEON, ADLINK and Advantech, have announced carrier boards and reference designs.
The Jetson Orin Nano 2 computer puts that breakthrough within reach of millions of developers, delivering the performance and energy efficiency needed for real-time reasoning at the edge.
Deepu Talla, vice president of robotics and edge AI at NVIDIA.
Why It Matters
The entry-level Edge AI tier decides how many robots get built at all. A module that fits a model on board and needs no carrier-board redesign lowers the barrier for small teams. Keeping the footprint matters more here than the TOPS number — it lets you swap the compute inside an existing product instead of rebuilding it. The price, which the vendor did not disclose, will decide whether the promise lands.
What’s Next
- Module and developer kit are slated for H1 2027 — until then there is no hardware for independent testing
- NVIDIA disclosed neither the price nor the memory bandwidth, the two numbers that fix its position against the predecessor
- The 2× inference claim cannot be verified without a published measurement methodology
