Nvidia has rolled out a new Jetson Thor kit aimed at humanoids, mainstream robotics and edge AI systems.
Nvidia chief executive Jensen Huang turned up in Japan and announced the Nvidia Jetson Thor T3000 and T2000, two machines built from the Jetson AGX Thor platform.
Kawasaki Heavy Industries is already using Nvidia technology to improve hospital operations with its FORRO, Nyokkey and NURABOT robots.
The company plans to use Nvidia Holoscan IGX, Isaac for Healthcare, Isaac GR00T and Cosmos for surgical support, nursing assistants and hospital transport robots.
Direava is developing a surgical vision language model for real-time surgical video understanding and natural language interaction with operating room scenes.
The outfit wants to turn that technology into an intelligence layer for future surgical AI and physical AI in the operating room.
Nvidia says the T3000 and T2000 computers are aimed at mass-market robotics and edge AI at scale.
X, Agility, Agile Robots, Amazon Robotics, Boston Dynamics, FANUC, Hitachi, Medtronic and Techman Robot are among those using the kit.
The Jetson Thor T3000 is a trimmed version of the flagship Thor T5000. It offers 865 TFLOPs of FP4 AI compute in a form factor roughly half the size and power of the T5000.
Like the larger model, the T3000 has an Nvidia Blackwell GPU, up to eight Arm Neoverse CPU cores and 32GB of LPDDR5X memory.
It comes with 273GB/s of memory bandwidth, 25GbE connectivity and the same safety features when running the Nvidia Halos Robotics stack.
Nvidia claims the T3000 delivers roughly the same inference performance as the T5000 in multimodal workloads. That includes LLMs, VLMs and world foundation models, because apparently robots now need to understand the universe before carrying a tray.
The T3000 is rated at 70 watts and is pitched as a balanced platform that dodges rising memory costs. The Jetson Thor T2000 is aimed at edge AI and offers 400 TFLOPs of FP4 compute with 16GB of memory.
It is designed for visual AI agents, autonomous mobile robots and other intelligence-based operations. The T2000 is rated at 40 watts, which should make it easier to squeeze into smaller, less thirsty machines.
The new Thor additions fill out the Jetson AGX Thor roadmap from entry-level Orin computers to higher-end Thor platforms.
Nvidia has rolled out new Jetson agent skills to help developers optimise the full software stack. The company says those tools can cut memory use enough for customers to drop down a module, saving power and cash.
Nvidia showed Jetson AGX Orin handling industrial and humanoid workloads while cutting memory use by half.
Humanoid robotics outfits UBTech and Agile Robots, along with industrial provider Connect Tech, cut memory use by up to 15GB. That allowed them to move from Nvidia Jetson AGX Orin 64GB modules to 32GB versions.
Smart retail firm SandStar cut memory use by up to 4GB, letting it use the Jetson Orin NX 8GB module instead of the 16GB model.
Companion robotics company GROOVE X, maker of the LOVOT robot, uses Jetson’s mixed AI accelerators to spread workloads more efficiently. That cut memory requirements and helped the company deploy on lower-memory configurations.
In intelligent transport, NoTraffic reduced memory usage by 30 per cent on the Jetson TX2 NX. That gives its smart traffic platform room for more AI tricks without needing extra hardware.
Jetson Thor T3000 will be available in emulation mode this month with JetPack 7.2.1.
The T2000 will arrive in a future release, with both modules scheduled for launch in the first quarter of 2027.
Partners offering Thor-based systems include ADLINK, Advantech, AAEON, Aetina, Auvidea, AVerMedia, Connect Tech, ForeCR, JWIPC, NEXCOM Robotic Solutions, Realtimes, Seeed Studio, Twowin, TZTEK and YUAN.
