Aetina expands its NVIDIA Jetson Thor portfolio with support for new Jetson modules
Aetina has announced its plans to support the newly introduced NVIDIA Jetson T3000 and T2000 modules, the latest additions to the NVIDIA Jetson Thor family. Building on its established portfolio of NVIDIA Jetson Thor- powered robotics computing systems, Aetina will extend its DeviceEdge AIE-KT series of fan-based edge computing systems, alongside new AIE-PT series fanless system designs, to support the T3000 and T2000 system-on-modules, bringing scalable, power-efficient physical AI compute to compact, cost effective and memory-optimised platforms for robotics and industrial applications.
NVIDIA Jetson T3000 delivers up to 865 FP4 TFLOPS through a NVIDIA Blackwell GPU with 1,536 CUDA cores, an 8-core Arm Neoverse CPU, and 32GB of LPDDR5X memory at 273 GB/s, within 70W. It achieves similar inference performance as the flagship T5000 at roughly half the size and power — ideal for scaled robotics deployments. NVIDIA Jetson T2000 delivers up to 400 FP4 TFLOPS via a NVIDIA Blackwell GPU with 1,024 CUDA cores and 16GB of LPDDR5X memory at 137 GB/s, enabling visual AI agents, autonomous mobile robots, and manipulators.
To bring both modules to market, Aetina plans to expand two dedicated product lines: the DeviceEdge AIE- KT series, a fan-based platform engineered for deployments requiring additional thermal headroom and sustained performance, and the new DeviceEdge AIE-PT series, a fanless design built for enclosed, rugged, and space-constrained environments where silent, maintenance-free operation is required. Together, AIE-KT and AIE-PT give customers a clear path to scale mainstream robotics deployments from prototype to mass production.
NVIDIA Jetson T3000 and T2000 modules are expected to be available in Q1 2027, with full support for the complete NVIDIA software stack, including open models as NVIDIA Nemotron, NVIDIA Cosmos 3, and NVIDIA GR00T, along with new Jetson agent skills. Among these, NVIDIA Cosmos 3 stands out as the first fully open omnimodal world model for physical AI, delivering real-time, multi-view vision understanding for smart infrastructure, robots, and autonomous systems. As a lightweight, cost-efficient 4B-parameter model, it can be readily customised into world-action models tailored to specific robot embodiments.
