NVIDIA Introduces New Jetson Thor Computers to Advance Mainstream Robotics and Edge AI
01:00 · July 16, 2026 · NVIDIA

General-purpose robots and autonomous machines are moving from research labs to real-world mass-market deployment, creating demand for compact, power-efficient AI supercomputers capable of running foundation models at the edge. To meet that need, NVIDIA today introduced the T3000 and T2000, new modules based on the NVIDIA Thor architecture that enable mass-market robotics and edge AI […]
Summary
NVIDIA has released the Jetson T3000 and T2000 modules, built on the Thor architecture, to address growing demand for compact, power-efficient hardware that can run foundation models on edge devices in robotics and autonomous systems. The T3000 delivers 865 FP4 teraflops using a Blackwell GPU, an eight-core Neoverse Arm CPU and 32 GB of LPDDR5X memory, while occupying roughly half the size and power envelope of the higher-end T5000. The smaller T2000 provides 400 FP4 teraflops and 16 GB of memory, extending the same architecture to a wider set of visual AI agents, mobile robots and industrial manipulators.
Both modules maintain inference performance comparable to larger Thor variants on multimodal workloads such as large language models, vision-language-action models and world foundation models. The T3000 also appears in an IGX variant that integrates functional safety features and supports the NVIDIA Halos safety stack for human-robot collaboration. Together with existing Jetson offerings, the new modules create a continuous performance range from 70 TOPS to 2 000 teraflops.
New Jetson agent skills automate memory optimization, system configuration and deployment steps that previously required manual tuning. Early users, including UBTech, Agile Robots and SandStar, have reported memory reductions of up to 15 GB, allowing migration to lower-capacity SKUs without loss of capability. The same skills apply across the Jetson Thor and Orin families.
NVIDIA has also introduced Cosmos 3 Edge, a 4-billion-parameter model derived from its open Cosmos 3 family. Designed for on-device inference on Thor hardware, the model supports real-time perception, reasoning and action generation for embodied systems. Developers can post-train it for specific robot embodiments and sensors within roughly one day using the open Cosmos framework.
The modules share the Thor software stack, enabling developers to begin work with the existing Jetson AGX Thor developer kit and later emulate T3000 or T2000 behavior through JetPack updates. Commercial availability is planned for the first quarter of 2027, with multiple ecosystem partners already offering compatible carrier boards and migration tools.
Why it matters
This article highlights crucial advancements in edge AI and robotics hardware, which are key growth areas for the Dutch AI market, particularly in logistics, agriculture, and smart retail. It provides a general AI audience with insights into how foundation models are transitioning from labs to real-world physical applications.





