NVIDIA has expanded its Jetson Thor lineup with two smaller Blackwell-based modules, T3000 and T2000, designed to push physical AI hardware out of research labs and into mass-market humanoid robots, autonomous mobile robots and vision AI agents. The pair, announced on 15 July from Tokyo, will ship as production modules in Q1 2027 and can be emulated today on the existing Jetson AGX Thor developer kit.
Half the size, T5000-class performance
The Jetson T3000 delivers 865 FP4 teraflops of AI compute in a footprint roughly half the size and power of the flagship Jetson AGX T5000, pairing a Blackwell GPU with an eight-core Arm Neoverse CPU, 32GB of LPDDR5X memory, 273GB/s of memory bandwidth and 25GbE networking. An IGX T3000 variant adds integrated functional safety and NVIDIA Halos for Robotics for machines that work alongside people. The smaller T2000 offers 400 FP4 teraflops and 16GB of memory as an entry point for visual AI agents and industrial manipulators.
Cosmos 3 Edge fits on the module
NVIDIA also released Cosmos 3 Edge, a 4-billion-parameter version of its Cosmos 3 world foundation model built to run on Thor devices for on-device robot policies and vision. Developers can post-train Cosmos 3 Edge for specific embodiments in about a day, closing the sim-to-real gap and enabling closed-loop physical AI without hauling every inference off the robot.
Who's building on Thor
Announced customers include 1X, Agile Robots, Amazon Robotics, Boston Dynamics, FANUC, Hitachi and Techman Robot. UBTech and Agile Robots have already used the new Jetson agent skills to cut memory use by up to 15GB and move workloads from 64GB AGX Orin down to 32GB modules, a critical cost-side story as HBM and LPDDR5X pricing stays elevated. Combined with the JetPack 7.2.1 emulation release later this month, the T3000 and T2000 give hardware vendors like ADLINK, Advantech and Connect Tech a scalable Thor stack that reaches from 70 TOPS to 2,000 teraflops. The launch pairs with NVIDIA's Cosmos 3 Edge Coalition in Japan, its 27,500-GPU Rubin national AI factory and expanded Toyota partnership — all part of a coordinated push to standardise physical AI on the Thor stack.
Reporting based on coverage from NVIDIA and The AI Insider.