The EC-ThorT5000 is an industrial-grade edge-AI PC built by Firefly, designed to handle really heavy AI workloads right at the edge (rather than relying purely on cloud servers). It’s packed with high-end hardware and geared for things like robotics, vision analytics, smart-city sensors, and generative AI models.
Under the hood, the machine uses the NVIDIA Jetson Thor T5000 module (SoM), which integrates a 14-core 64-bit ARM Neoverse-V3AE CPU (clocking up to ~2.6 GHz) and a 2560-core “Blackwell” GPU with 96 tensor cores. It delivers up to 2070 TFLOPS (FP4, sparse) of AI compute power. It supports huge memory and storage too: up to 128 GB LPDDR5X memory, dual M.2 NVMe SSD slots, etc. Additionally, it offers extremely strong video I/O up to 92 channels of 1080p30 video decoding, and heavy encoding workloads (e.g., 50 channels of 1080p30 or multiple 4K channels).
On the connectivity front, you’ll find four 10-Gigabit Ethernet ports and four standard Gigabit ports (with PoE support) built in. It is also capable of expansion with WiFi 6/BT 5.2, 4G/5G cellular variations, GPS, as well as camera interfaces such as GMSL2 (compatible with multi-camera vision systems).
The box is made of an all-aluminium industrial design and has two fans to cool the box, and this is the reason why this box may be used around the clock, even in severe conditions. These are over 278 x 136 x 88 mm in size, and the weight measures between 2.83 to 3 kg. It can maintain a wide operating temperature (-20 °C to 60 °C) in severe industry. Power input is flexible: it runs off 24 V DC (though it supports a 9V to 36V wide-voltage input).
If you’re developing AI solutions at the edge, say, autonomous robots, smart cameras in factories, or distributed generative-AI inference nodes, the EC-ThorT5000 brings server-level compute closer to the data source. That means less lag, better privacy (models stay local), and lower reliance on cloud bandwidth. Due to its compatibility with numerous frameworks (PyTorch, TensorFlow, PaddlePaddle, et cetera) and workflows in deep-learning and generative-AI (LLMs, vision models, image-generation tools, etc), it is not limited in how it fits modern AI work.
Software-wise, the EC-ThorT5000 is compatible with other popular AI toolsets such as PyTorch, TensorFlow, PaddlePaddle, and generative-AI engine, but is also powered by NVIDIA JetPack, CUDA, TensorRT and cuDNN. It also supports a Linux environment with Docker support so that it can be easily deployed with AI workloads, multi-camera processing, video analytics, and robotics applications in the box.
NVIDIA has previously showcased the Jetson Thor Developer Kit with T5000 Blackwell GPU and other similar products, such as the Vecow EAC-7000, the Connect Tech Gauntlet Carrier Board, and the GIGABYTE AI TOP ATOM. Feel free to check those out for your reference.
As of now, the EC-ThorT5000 is already listed as available for purchase on Firefly’s website, but the company has not published an official retail price for the unit. Since it’s built around the NVIDIA Jetson T5000 module—which itself sells for around $3,199 from distributors-the complete system is expected to fall into the high-end industrial price range, likely several thousand dollars depending on configuration and storage options. Availability is immediate through Firefly’s official store and authorised distributors, with global shipping offered for industrial and AI deployment projects.
Debashis Das