NVIDIA Jetson Thor T3000 & T2000 — Mainstream Blackwell Edge AI Modules

Specifications

Overview

The NVIDIA Jetson Thor T3000 and T2000 expand the Thor family with compact, power-efficient Blackwell modules designed for mainstream robotics and edge AI deployment. The T3000 delivers 865 FP4 TFLOPS at 70W — approximately half the size and power of the flagship T5000 — while achieving similar LLM/VLM inference performance. The entry-level T2000 provides 400 FP4 TFLOPS in a 40W envelope for visual AI agents, autonomous mobile robots (AMR), and industrial manipulators. Both modules share the Blackwell GPU architecture with 5th-gen Tensor Cores, enabling on-device generative AI, Cosmos 3 Edge world models, and Jetson Agent Skills for automated memory optimization.

Key Benefits

Scalable Platform: From 400 to 2,070 FP4 TFLOPS, the Thor family scales across the entire robotics and edge AI spectrum. T3000 matches T5000 inference throughput on multimodal models at half the size. Cost-Optimized: Right-size memory per deployment — T2000's 16GB suits visual AI agents and AMR perception, while T3000's 32GB handles VLMs and world foundation models. Agent Skills: Jetson Agent Skills automates memory optimization, enabling developers to move down one memory SKU without performance regression. Software-Ready: Full Isaac, Cosmos 3 Edge, and JetPack 7.x stack with emulation available on existing AGX Thor developer kits today.