NVIDIA introduced Jetson T3000 and T2000 modules based on Thor architecture for mass-market robotics and edge AI. Leading companies including 1X, Agile Robots, Amazon Robotics, Boston Dynamics, FANUC, Hitachi, and Techman Robot are building on the platform.

The T3000 delivers 865 FP4 Tera FLOPS with Blackwell GPU, eight-core Neoverse Arm CPU, 32 GB LPDDR5X memory, and 273 GB/s bandwidth in half the size and power of T5000. The T2000 offers 400 FP4 Tera FLOPS and 16 GB memory for broader edge AI. Together they span 70 TOPS to 2,000 Tera FLOPS.

New Jetson agent skills automate memory optimization across all Jetson devices. UBTech, Agile Robots, and Connect Tech reduced memory by up to 15 GB, moving from 64 GB to 32 GB modules. Sand Star saved 4 GB, GROOVE X optimized workload distribution, and No Traffic cut usage 30% on TX2 NX.

NVIDIA expanded Cosmos 3 with Cosmos 3 Edge, a 4-billion-parameter model for on-device inference on Thor. Developers can start today using Jetson AGX Thor dev kit with T3000 emulation in JetPack 7.2.1 this month; T2000 emulation follows. Modules arrive Q1 2027 with broad ecosystem support.