NVIDIA introduced the T3000 and T2000 modules based on the Thor architecture to enable compact, power-efficient edge AI

The emergence of general-purpose robots and autonomous machines moving into mass-market deployment has created demand for compact, power-efficient AI supercomputers capable of running foundation models at the edge. To address this need, NVIDIA today announced the T3000 and T2000 modules based on the NVIDIA Thor architecture, designed to enable mass-market robotics and edge AI applications at scale.

The Jetson AGX Thor platform is powering the next generation of humanoid and robotic systems with growing adoption across industries. Leading companies including 1X, Agile Robots, Amazon Robotics, Boston Dynamics, FANUC, Hitachi and Techman Robot are already building on the platform.

The Jetson and IGX T3000 modules deliver 865 FP4 teraflops of AI compute in a compact form factor roughly half the size and power of the T5000. The Jetson T3000 combines an NVIDIA Blackwell GPU, an eight-core Neoverse Arm CPU, 32GB of LPDDR5X memory and 273GB/s of memory bandwidth, along with 25 GbE connectivity. The IGX T3000 delivers the same performance with integrated functional safety while running the NVIDIA Halos for Robotics full-stack safety system. Despite its smaller footprint, the T3000 achieves similar inference performance to the T5000 for multimodal workloads including large language models, vision language models, vision language action models and world foundation models, helping reduce costs amid high memory prices.

The Jetson T2000 brings Thor architecture to a broader range of edge AI systems with 400 FP4 teraflops of compute and 16GB of memory, providing an entry point for developers building visual AI agents, autonomous mobile robots, industrial manipulators and other intelligent machines. With these new modules, NVIDIA now offers a scalable edge AI platform spanning performance from 70 TOPS to 2,000 teraflops.

Newly released Jetson agent skills enable developers to optimize the entire software stack and achieve significant memory savings in days instead of weeks, supporting the entire Jetson portfolio including Jetson Thor and Jetson Orin. This allows developers to run more capable workloads on lower-memory configurations, resulting in lower system cost and faster deployment.

Companies across industries have demonstrated substantial memory savings through software optimization. Humanoid robotics leaders including UBTech and Agile Robots, along with industrial solutions provider Connect Tech, reduced memory usage by up to 15GB, enabling migration from NVIDIA Jetson AGX Orin 64GB to the 32GB module. In smart retail, SandStar reduced memory usage by 4GB, enabling deployment on the NVIDIA Jetson Orin NX 8GB instead of 16GB. In companion robotics, GROOVE X, creator of the LOVOT robot, used Jetson's heterogeneous AI accelerators to optimize workload distribution and reduce memory usage. In intelligent transportation, NoTraffic achieved a 30% memory reduction on Jetson TX2 NX, creating headroom to add more AI capabilities without increasing hardware requirements.

NVIDIA also expanded its NVIDIA Cosmos 3 frontier open world foundation model family with Cosmos 3 Edge, a lightweight 4-billion-parameter model compatible with NVIDIA Thor platforms designed as a robot foundation model for embodied systems. Cosmos 3 Edge helps embodied systems see the world, reason over it in real time, and predict and generate actions through on-device inference. Using the open Cosmos framework, developers can post-train Cosmos 3 Edge for specific embodiments and sensors in about a day, then deploy on Jetson Thor for real-time vision analysis and on-device robot policy.

Developers can begin using T3000 emulation mode later this month with JetPack 7.2.1, with T2000 emulation mode support following in a future release. The Jetson T3000 and T2000 modules are scheduled to become available in Q1 2027. Partners in the Jetson ecosystem including ADLINK, Advantech, AAEON, Aetina, Auvidea, AVerMedia, Connect Tech, ForeCR, JWIPC, NEXCOM Robotic Solutions, Realtimes, Seeed Studio, Twowin, TZTEK and YUAN are already providing Thor-based solutions, while software partners such as Antmicro, Neurealm, REBOTNIX and RidgeRun will provide emulation and migration solutions.