NVIDIA ships smaller Jetson Thor modules

NVIDIA introduced two new modules built on its Thor architecture, aimed at pushing humanoid and edge robotics from research labs into mass-market deployment. The Jetson and IGX T3000 delivers 865 FP4 teraflops of AI compute, an NVIDIA Blackwell GPU, an eight-core Arm CPU, 32GB of memory and 25 GbE connectivity, in roughly half the size and power draw of the existing T5000 module, while matching its inference performance on multimodal workloads like vision-language and world-foundation models. The IGX variant adds integrated functional safety for robots that share space with people.

The Jetson T2000 is the lower-tier option: 400 FP4 teraflops and 16GB of memory, meant as an entry point for visual AI agents, autonomous mobile robots, and industrial manipulators. Together with the existing Jetson AGX Thor and Jetson Orin lines, NVIDIA says its edge AI platform now spans 70 TOPS to 2,000 teraflops. Early adopters building on the Thor platform span both humanoid specialists like 1X and Agile Robots and larger industrial names, among them Amazon Robotics, Boston Dynamics, FANUC, Hitachi, and Techman Robot. NVIDIA also released Jetson agent skills, tools that use AI agents to automate memory optimization and system configuration across the whole Jetson lineup, cutting a task the company says used to take weeks down to days.

What it means for operators

Two things worth tracking here if you touch physical or edge AI: the T3000 makes near-flagship Jetson performance viable in smaller, cheaper, lower-power products, which is the kind of move that turns a research demo into a shippable device; and NVIDIA using agent skills to automate its own hardware-optimization workflow is a preview of where deployment tooling is heading across the edge stack, not just in the cloud. It also lands the same week Grok 4.3 arrived on Amazon Bedrock, another sign that both the model layer and the hardware underneath it are being tuned specifically for agentic, not just chat-style, workloads.