NVIDIA launched the T3000 and T2000 edge AI modules based on its Thor architecture to enable mass-market deployment of r
NVIDIA introduced two new Jetson and IGX modules designed to power the growing wave of humanoid robots and edge AI systems moving from research into real-world deployment. The Jetson T3000 delivers 865 FP4 teraflops of AI compute in a form factor roughly half the size and power consumption of the T5000. It combines an NVIDIA Blackwell GPU with an eight-core Neoverse Arm CPU, 32GB of LPDDR5X memory, and 273GB/s of memory bandwidth, along with 25 GbE connectivity. The IGX T3000 offers the same performance with integrated functional safety for robots operating alongside humans. 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 during periods of high memory pricing.
The Jetson T2000 brings the Thor architecture to a broader range of applications, offering 400 FP4 teraflops of compute and 16GB of memory as an entry point for developers building visual AI agents, autonomous mobile robots, industrial manipulators and other intelligent machines. Together, these modules extend NVIDIA's edge AI platform from 70 TOPS to 2000 teraflops, covering virtually any edge AI workload.
NVIDIA also released new Jetson agent skills that automate memory optimization and system configuration. Developers can now optimize their entire software stack and achieve significant memory savings in days instead of weeks. The skills support the full Jetson portfolio including Jetson Thor and Jetson Orin. Companies have already seen substantial results: UBTech and Agile Robots reduced memory usage by up to 15GB, enabling downgrades from the 64GB Jetson AGX Orin to the 32GB module. SandStar reduced memory by 4GB to deploy on the Jetson Orin NX 8GB instead of 16GB. GROOVE X optimized memory usage for the LOVOT robot, and NoTraffic achieved a 30 percent memory reduction on Jetson TX2 NX for its smart traffic platform.
NVIDIA expanded its Cosmos 3 frontier foundation model family with Cosmos 3 Edge, a lightweight 4-billion-parameter model designed for NVIDIA Thor platforms. This robot foundation model enables embodied systems to see the world, reason about 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 to close the sim-to-real gap, 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 support to follow. The Jetson T3000 and T2000 modules are scheduled to become available in Q1 2027. ADLINK, Advantech, AAEON, Aetina, Auvidea, AVerMedia, Connect Tech, ForeCR, JWIPC, NEXCOM Robotic Solutions, Realtimes, Seeed Studio, Twowin, TZTEK and YUAN are among ecosystem partners already providing Thor-based solutions, while software partners including Antmicro, Neurealm, REBOTNIX and RidgeRun will provide emulation and migration solutions for customers transitioning to the new modules.