Nvidia introduced two new Jetson Thor modules – the T3000 and T2000 – expanding its robotics and edge AI portfolio beyond the flagship AGX Thor platform. The compact modules are designed for mainstream robotics, humanoid machines, and edge AI applications where size, power, and cost constraints rule out the full-scale Thor system.
The T2000 delivers up to 400 TFLOPS of AI performance with a 1024-core Blackwell GPU, 16 GB of LPDDR5 memory at 137 GB/s bandwidth, and a power envelope optimized for embedded deployments. The T3000 targets higher-end applications with additional compute cores and memory bandwidth for complex perception workloads.
Both modules support Nvidia’s full Isaac robotics software stack, including cuOpt for path planning and Replicator for synthetic data generation. Availability is expected in the first quarter of 2027, with partners including Seeed Studio and ADLINK already announcing support.
The expansion comes as Nvidia positions its Thor architecture – first announced in 2023 – to capture the growing market for physical AI systems, from autonomous mobile robots to humanoid platforms that require real-time sensor fusion and decision-making at the edge.
