NVIDIA introduced its Jetson T3000 and T2000 modules alongside Cosmos 3 Edge, a 4-billion-parameter model that lets machines perceive and act through on-device inference.
On 15 July 2026 NVIDIA announced two edge computing modules built on its Thor architecture. The T3000 pairs a Blackwell GPU with 32GB of memory and delivers 865 FP4 teraflops. The entry-level T2000 delivers 400 teraflops.
The aim is high-performance perception inside robots, industrial machines and visual edge systems, without relying on the cloud. Cosmos 3 Edge, a 4-billion-parameter foundation model, lets these systems see their surroundings, reason about them in real time and generate actions through on-device inference.
The platform runs vision-language and vision-language-action models. Developers can post-train Cosmos 3 Edge for a specific machine in roughly a day. Emulation support begins in late July 2026, and the modules are expected in the first quarter of 2027.
Perception, reasoning and control are moving onto small, low-power devices. For Ergonet, which works on machine vision and robot-agnostic warehouse automation, cheaper on-device vision means faster progress in Physical AI.
Source: NVIDIA