Innodisk unveiled a five-layer edge artificial intelligence ecosystem at Computex 2026 spanning compute, memory, storage, sensing and communication, and software.
At the centre of the display is a set of on-premises artificial intelligence systems for enterprise users that want to keep data within their own infrastructure. They include AccelBrain, which runs on the APEX-X200 and is designed for on-premises deployment of open-source large language models, and AccelTune, a no-code fine-tuning product shown on the APEX-S100 with Intel Xeon 6700-series processors and support for dual Nvidia RTX Pro graphics processors.
With those products, Innodisk is targeting businesses that want application-specific large language model workflows without relying on public cloud services. The emphasis on local deployment reflects broader demand from corporate customers for tighter control over data handling as generative artificial intelligence moves into operational use.
Industrial focus
Alongside its enterprise systems, Innodisk outlined several industrial edge artificial intelligence applications built for demanding operating conditions. One demonstration focuses on heavy machinery safety, using eight ruggedised GMSL2 camera modules rated IP67 and IP69K.
The setup combines surround-view image stitching, a Driver Monitoring System, and Blind Spot Detection to improve operator visibility in field environments. Innodisk is also presenting an OCR container identification demonstration running on an integrated edge system, alongside rugged camera products using GMSL2, MIPI, and USB interfaces.
Another system on show is the APEX-E400 edge artificial intelligence platform, which uses Intel Core Ultra Series 3 processors. The machine is built around a CPU, GPU, and NPU architecture and can handle up to 16 simultaneous streams with parallel model inference.
A live demonstration based on Intel OpenVINO is designed to show how workloads can be distributed across those processing engines. The approach has become common in edge computing as suppliers try to balance performance, power use, and response times in systems deployed outside central data centres.
Qualcomm systems
Innodisk is also expanding its work with Qualcomm Dragonwing processors for compact edge inference systems. It is showing Dragonwing processors across its IQ9, IQ10, and IQ-X series products.
Using a COM-HPC Mini platform built on the Dragonwing IQ-9075 processor, Innodisk is demonstrating artificial intelligence screw thread inspection and multi-stream vision processing. It is also showing an autonomous mobile robot application that combines GMSL and depth cameras for spatial awareness, ROI safety zone detection, and PWM-triggered automatic emergency braking.
Software layer
Software forms the fifth layer of the ecosystem and is being presented through a link between Edge Impulse's machine learning operations platform and iCAP, Innodisk's cloud management software. The combination is intended to support automated remote model updates and deployment through what Innodisk describes as Agentic AI.
This software layer sits alongside the hardware stack that has long formed the core of Innodisk's business. Founded in Taiwan in 2005, the company built its position in industrial memory and storage before expanding into edge artificial intelligence systems and integrated applications.
Component range
Supporting technologies on display include a next-generation DDR5 8000 RDIMM, CUDIMM and CSODIMM products, as well as 12800 MRDIMM and a new CXL Add-in Card. In storage, Innodisk is presenting data centre solid-state drives in EDSFF and U.2 form factors, together with 218-layer 3D TLC solid-state drives.
In networking, Innodisk introduced the ELPL-82F1 SFP28 25GbE LAN card. It is also showing the EGPL-T2F1, recognised at Embedded World 2026 as the first SFP+ LAN card in an M.2 form factor, according to the company.
The breadth of the line-up reflects how edge artificial intelligence deployments are increasingly assembled from a mix of computing platforms, memory, storage, cameras, connectivity, and management software rather than from a single device. Innodisk's approach groups those elements into one stack aimed at enterprise and industrial users with different operating requirements.
The exhibits are intended to show "how enterprises and industrial customers can accelerate AI deployment for production-ready applications", Innodisk said.