Computing power in the right place, not just less power
As artificial intelligence matures from laboratory curiosity to industrial necessity, a quiet but significant shift is underway: the intelligence is moving closer to the world it observes. Aetina's Mini Series edge AI systems — palm-sized, fanless, and capable of 157 trillion operations per second — embody this migration, placing real-time vision processing directly inside factories, security posts, and inspection sites where latency, space, and power have long been the limiting factors. Powered by NVIDIA's Jetson Orin processors, these devices signal that the era of centralized AI is giving way to something more distributed, more immediate, and more embedded in the physical fabric of work.
- The core tension is practical: companies have proven AI works in labs but now need it to survive freezing warehouses, sun-baked rooftops, and cramped factory floors where cloud connectivity is a liability, not an asset.
- Aetina's Mini Series compresses up to 157 TOPS of computing power into a device smaller than a paperback book, running without fans across temperatures from -25°C to +55°C — engineering designed for environments that punish fragility.
- Flexible camera inputs, modular wireless support for LTE, 5G, and Wi-Fi, and compatibility with standard industrial power sources mean these systems can slot into existing infrastructure rather than demanding it be rebuilt around them.
- At COMPUTEX 2026, Aetina will demonstrate vision-language models running entirely on-device — a technician asking a question in plain language, a camera analyzing the scene, an answer returned locally, with no cloud in the loop.
- All four Mini Series models are already available for mass production, and Aetina is pairing the hardware with software support, customization services, and technical consulting to accelerate the journey from prototype to deployed system.
Aetina has released the Mini Series, a line of compact edge AI systems built on NVIDIA's Jetson Orin processors, designed to bring artificial intelligence processing directly into industrial environments — factories, security checkpoints, inspection sites — where space is limited and real-time decisions cannot wait for a distant server.
The hardware is strikingly small. The AIE-CO23 and AIE-CN33 models are roughly the size of a paperback book, while the slimmer AIB-MO23 and AIB-MN33 variants compress further to just 127×85.75×28.45 millimeters. All four run without fans, reducing maintenance and failure risk in harsh conditions. Despite their size, they deliver up to 157 trillion operations per second — enough to run vision models capable of recognizing objects, detecting anomalies, and understanding scenes in real time.
Practical flexibility is central to the design. The systems accept standard industrial DC power, operate reliably between -25°C and +55°C, connect to cameras via MIPI or Power over Ethernet, and support LTE, 5G, Wi-Fi, and Bluetooth through modular expansion. They run NVIDIA's JetPack 6.2 software environment, and Aetina offers customization and consulting services to help customers move from development into production.
At COMPUTEX 2026, Aetina will partner with Japanese software firm ISP to demonstrate the AIE-CN33 running vision-language models — AI that understands both images and natural language simultaneously — for smart inspection and security monitoring, all processed on the edge device itself without cloud connectivity. Fitting that kind of multimodal capability into a palm-sized system represents a meaningful compression of what was recently considered demanding infrastructure.
Troy Lin, who leads Aetina's edge AI division, framed the release as a response to a fundamental shift: computing power is no longer the scarce resource. Space, power consumption, and the ability to process data where it is generated — those are the constraints now. The Mini Series is Aetina's answer, built for companies ready to move AI from the lab into the world.
Aetina, a company focused on edge AI infrastructure, has released a new line of compact systems designed to bring artificial intelligence processing directly to where the work happens—factories, security checkpoints, inspection sites, places where space is tight and power is scarce. The Mini Series, powered by NVIDIA's Jetson Orin processors, represents a shift in how companies think about deploying AI: not as a centralized cloud operation, but as distributed intelligence living at the edge of their networks, processing video and sensor data in real time without sending everything back to a distant server.
The hardware itself is small enough to hold in your hand. The AIE-CO23 and AIE-CN33 models measure just 132 by 91.5 by 68.5 millimeters—roughly the size of a paperback book standing on end—and they run without fans, which means less maintenance, less noise, less to break in a harsh environment. The even slimmer AIB-MO23 and AIB-MN33 variants compress the footprint further to 127 by 85.75 by 28.45 millimeters, thin enough to fit into spaces where traditional computing equipment simply won't go. Despite their size, these systems deliver up to 157 trillion operations per second, enough computational muscle to run vision models that can recognize objects, detect anomalies, and understand scenes in real time.
What makes this practical is the flexibility built into the design. The systems accept power from standard 12 to 24 volt DC sources—the kind of power available in industrial settings—and they operate reliably across a temperature range from minus 25 degrees Celsius to plus 55 degrees Celsius, meaning they work in freezing warehouses and sun-baked outdoor installations alike. They connect to cameras through either MIPI interfaces or Power over Ethernet, so they can pull video directly from existing surveillance infrastructure or new sensors. They also support LTE, 5G, Wi-Fi, and Bluetooth through modular expansion, giving integrators options for how data moves in and out of the system.
The timing of this release reflects a broader maturation in AI deployment. Companies have moved past the proof-of-concept phase—the stage where they test whether an idea works in a lab. Now they need systems that actually work in the field, in production, under real constraints. A factory floor doesn't have room for a rack of servers. A security checkpoint can't afford the latency of sending every frame to a cloud service hundreds of miles away. A bridge inspection needs to process sensor data locally, in the moment, to catch structural problems before they become dangerous. Aetina's Mini Series addresses these practical realities.
The company is demonstrating the potential at COMPUTEX 2026, where it will partner with a Japanese software firm called ISP to show how the AIE-CN33 can run vision-language models—AI systems that understand both images and natural language—for tasks like smart inspection and security monitoring. A technician could point a camera at something and ask a question in plain language, and the system would analyze the image and respond, all happening on the edge device itself, without cloud connectivity. This kind of multimodal processing, fusing video and text understanding, has traditionally required significant computing power. Fitting it into a palm-sized device represents a genuine compression of capability.
Aetina is backing the hardware with software support. The systems run NVIDIA's JetPack 6.2 and later, a standardized software environment that makes it easier for developers to build applications. The company also offers customization services and technical consulting to help customers move from development to production faster. All four models in the Mini Series are already available for mass production, meaning companies can order them today and begin integrating them into their systems.
Troy Lin, who directs Aetina's edge AI computing division, framed the release as a response to a fundamental shift in how AI gets deployed. The constraint is no longer computing power—that's become abundant and cheap. The constraint is now space, power consumption, and the ability to process data where it's generated, without the latency and cost of sending everything to a distant data center. The Mini Series is Aetina's answer to that constraint, a platform designed to let companies scale their AI deployments from the lab into the real world, one compact system at a time.
Notable Quotes
AI deployment is no longer restricted to large-scale infrastructure. In actual field scenarios, high-performance and reliable intelligent computing are required within a limited footprint.— Troy Lin, Director of Edge AI Computing Product Division at Aetina
The Hearth Conversation Another angle on the story
Why does size matter so much for edge AI? Couldn't companies just use smaller versions of existing servers?
Because edge AI isn't about having less computing power—it's about having computing power in the right place. A factory floor, a security checkpoint, a remote inspection site—these aren't places where you can install a server rack. You need something that fits in a confined space, draws minimal power, and can sit in extreme temperatures without failing. The Mini Series is engineered for those constraints.
So these are essentially stripped-down versions of larger systems?
Not really. They're purpose-built. A fanless design isn't just smaller—it's fundamentally different. No moving parts means lower maintenance, longer lifespan in harsh environments, and no noise. The power input is 12 to 24 volts, which is what industrial equipment already uses. It's not about doing less; it's about doing the right things in the right way.
What changes when you process video locally instead of sending it to the cloud?
Everything. Latency disappears. A security system doesn't have to wait for data to travel to a distant server and back—it can detect a threat in milliseconds. You also reduce bandwidth costs and privacy concerns. The data stays local. And if your internet connection drops, the system keeps working. That's critical in industrial settings.
The specs mention up to 157 TOPS. Is that enough for real AI work?
It depends on the task, but yes. Vision models have become much more efficient. You can run object detection, anomaly detection, even lightweight vision-language models on that much compute. The demo at COMPUTEX will show a system understanding both images and natural language questions—that's not trivial work, and it's happening on a device the size of a paperback book.
Who actually needs this? What's the real use case?
Anyone doing visual inspection or monitoring in the field. A factory checking for defects on a production line. A utility company inspecting power lines or bridges. A security operation monitoring multiple locations. A healthcare facility doing equipment maintenance checks. Anywhere you need real-time visual intelligence without the infrastructure of a data center.