Image: blogs.nvidia.com
Compact, more power-efficient Jetson Thor modules extending Blackwell-class compute to mainstream robotics.
Key specs
| Architecture | Blackwell-class |
|---|
Technical details
Jetson T3000 and T2000 are mainstream Thor-architecture modules announced July 15, 2026. T3000 pairs a Blackwell GPU with an 8-core Neoverse Arm CPU, 32 GB LPDDR5X at 273 GB/s and 25 GbE for 865 FP4 TFLOPS, at roughly half the size and power of T5000 while NVIDIA claims similar multimodal inference performance; IGX T3000 adds functional safety. T2000 offers 400 FP4 TFLOPS with 16 GB memory as an entry point. Developers can emulate T3000 on Jetson AGX Thor with JetPack 7.2.1; both modules ship in Q1 2027.
| Architecture | NVIDIA Thor-architecture modules with Blackwell GPU and Arm Neoverse CPU (T3000: 8-core Neoverse); IGX T3000 variant adds integrated functional safety. |
|---|---|
| Params | T3000: 865 FP4 TFLOPS, 32 GB LPDDR5X, 273 GB/s; T2000: 400 FP4 TFLOPS, 16 GB |
| Inputs outputs | Edge robotics computer for LLM, VLM, VLA and world-model inference; T3000 has 25 GbE. |
| Embodiments supported | Humanoids, AMRs, industrial manipulators, visual AI agents. |
| Hardware requirements | T3000 roughly half the size and power of T5000; development via Jetson AGX Thor dev kit in emulation mode (T3000 emulation with JetPack 7.2.1); modules scheduled for Q1 2027. |
| How to try | developer.nvidia.com/…/jetson |
| Sources |
