NVIDIA launches Jetson Thor T2000 and T3000 modules for mainstream robotics and edge AI

NVIDIA announced two new mid‑range Jetson Thor modules, the T2000 and T3000, as lower‑cost alternatives to its high‑end T4000/T5000 boards. The T3000 delivers up to 865 FP4 TFLOPS of AI compute, uses a 1536‑core Blackwell GPU, eight Arm Neoverse‑V3AE CPU cores, 32 GB LPDDR5X memory and 273 GB/s bandwidth, while the T2000 offers 400 FP4 TFLOPS with a 1024‑core GPU, six CPU cores, 16 GB memory and 137 GB/s bandwidth. Both modules are roughly half the size of the T5000, consume about 65 W (T3000) and 40 W (T2000), and include 25 GbE (T3000) or 10 GbE (T2000) networking. Availability is slated for the first quarter of 2027.

The new boards target humanoid and industrial robots, autonomous mobile platforms, smart‑city edge AI, and visual AI agents. NVIDIA cited partners such as Boston Dynamics, Agile Robots, Amazon Robotics, FANUC, Hitachi and others that are already developing solutions on the Thor platform. An IGX‑type T3000 with functional‑safety features and the Cosmos 3 Edge 4‑billion‑parameter world model were also unveiled, enabling on‑device perception and action planning for embodied AI. NVIDIA released “Jetson agent skills” to automate memory‑optimisation, allowing developers to shrink memory footprints by up to 15 GB in certain robot platforms. AAEON announced fanless edge AI systems built around the new modules, underscoring their applicability to smart‑city infrastructure.