
Firefly AIBOX-9075 Brings 200 TOPS and Local LLMs to the Edge
Firefly's AIBOX-9075 edge AI box pairs a Qualcomm Dragonwing IQ-9075 with up to 200 TOPS and 36GB RAM to run private, on-device LLMs — detailed June 26, 2026.
Data-Center AI Muscle in a Rugged Little Box
The dream of running serious AI locally and privately keeps getting more achievable, and the Firefly AIBOX-9075, detailed on June 26, 2026, is a striking example. This is a compact edge AI box built to run vision models and even local language models entirely on-site — no cloud round-trip required. For anyone who cares about privacy, latency, or operating where connectivity is unreliable, that's a meaningful capability in a small, tough package.
The Qualcomm Dragonwing IQ-9075 Inside
The engine here is Qualcomm's Dragonwing IQ-9075 system-on-chip — an octa-core Kryo Gen 6 CPU clocking up to 2.36 GHz, paired with four Cortex-R52 real-time cores and an Adreno 663 GPU. The standout figure for edge-AI workloads is the NPU rated up to 200 TOPS (INT8, sparse). To put that in perspective, that's the kind of neural throughput we associated with rack hardware not long ago, now sitting in a fanless box. Firefly cites roughly 12 tokens per second for local LLM inference, which is enough to make on-device assistants and document processing genuinely practical.
Memory, Storage, and a Wall of I/O
This box is clearly aimed at builders of robotics and computer-vision systems. It carries a generous 36 GB of LPDDR5 with ECC for reliability, 128 GB of UFS 2.2 storage, and an M.2 2280 PCIe 4.0 NVMe slot for expansion. The I/O is where it really flexes: dual 2.5GbE with TSN, eight GMSL2 camera inputs, HDMI 2.0, USB 3.0, CAN-FD, RS485, and opto-isolated digital I/O. Eight camera inputs is a clear signal this is meant for serious multi-sensor vision deployments.
Wireless options include Wi-Fi 6, Bluetooth 5.2, and optional 4G/5G, and the rugged aluminum enclosure is rated for a wide −40°C to 85°C operating range. Software-wise it runs Ubuntu or Yocto on Qualcomm's Linux stack, with support for the usual frameworks — QNN, plus TensorFlow, PyTorch, ONNX, and Caffe.
Where It Fits
At around $1,239 direct from Firefly, this isn't an impulse SBC purchase — it's a purpose-built tool. But for the jobs it targets, the value is clear: industrial inspection, autonomous machines, smart-city cameras, and any application where you want private, on-device AI running reliably in a harsh environment. The combination of 200 TOPS, eight camera inputs, ECC memory, and a rugged temperature range is a tidy fit for that brief.
The Takeaway
The AIBOX-9075 is a great marker of how far edge AI has come. Running capable vision models — and even local LLMs — on a fanless, ruggedized box keeps data on-site, cuts cloud dependence, and puts real intelligence exactly where the sensors are. For the robotics and computer-vision crowd, that's an exciting direction.
Sources: CNX Software — "Firefly AIBOX-9075 edge AI box features Qualcomm IQ-9075 SoC with 200 TOPS NPU" — June 26, 2026; Firefly official product page (en.t-firefly.com) — June 2026; Let's Data Science — "Firefly launches AIBOX-9075 industrial edge AI box" — June 2026.
More Mini Computers Stories

ESP32-P4 Desktop OS: Sylphide OS and BMDS Turn MCUs Into PCs
Two open-source desktop OS projects, Sylphide OS and BMDS, give the $25 Olimex ESP32-P4-PC windows, apps, SSH and even Doom at 1280x720 and 60 FPS.

Tektronix 1 Series Oscilloscope: $450 Scope With AI Control
Tektronix's 1 Series DPO starts at $450 with 12-bit resolution, up to 200 MHz, I2C/SPI/UART decoding and an AI connector for natural-language control.

Eight Used V100 GPUs: How a Lawyer Built a Private AI Server
A lawyer's private AI server runs DeepSeek V4.1 Flash on eight used V100 GPUs at up to 215 tokens per second, keeping case files off the cloud.
