NVIDIA unveils Jetson Thor-based modules T3000 and T2000 to push mainstream robotics and edge AI, aiming at mass-market deployment of autonomous systems.

A broad ecosystem of channel partners, hardware and software collaborators, and development kits are already offering emulation and migration solutions to support Thor-based systems.

Industry players such as UBTech, Agile Robots, Connect Tech, SandStar, GROOVE X, NoTraffic, and others are adopting Jetson to boost memory and performance while enabling lower-memory configurations.

NVIDIA emphasizes an end-to-end platform strategy with development tools, emulation mode, and a complete software stack (including Isaac, NemoClaw, Cosmos 3, and open models) to accelerate deployment from development to production.

Emulation mode and JetPack 7.2.1 let developers begin building now, with T3000 emulation available this month and T2000 emulation later; physical modules are expected in Q1 2027.

Cosmos 3 Edge debuts a 4-billion-parameter robot foundation model compatible with Thor platforms to enable on-device real-time vision, reasoning, and action generation.

Jetson T3000 delivers 865 FP4 teraflops, 32GB LPDDR5X, an eight-core Neoverse Arm CPU, a NVIDIA Blackwell GPU, 25 GbE, and integrated safety via IGX T3000, supporting multimodal on-site processing for robotics while targeting cost reduction.

Jetson T2000 expands Thor into broader edge AI with 400 FP4 teraflops and 16GB memory, extending Thor to visual AI agents, autonomous mobile robots, and industrial manipulators.