On November 3, edge-AI vendors such as InexAI gathered ecosystem partners intensively, racing to lock down the on-device market from large-model chips to real-world scenarios. According to IIM, the global market for on-device large-model software, hardware and services will reach US$41.27 billion in 2026, up 53.8% year on year, and is expected to surpass US$128 billion by 2030.
On-device AI is moving from concept to mass deployment: AI PCs let agents operate computers, robots adopt vision-language-action models, and smart-home and automotive-cockpit players race to embed large models into devices. The crux of running a large model on-device is balancing compute and memory — mainstream edge AI chips already run 1.5B–32B parameter models locally, using NPU heterogeneous compute, quantization and knowledge distillation to keep privacy-sensitive, low-latency tasks on the device and leave complex tasks to the cloud, forming edge-cloud collaboration.
From a competitive-structure perspective, the on-device surge reflects an across-the-board acceleration of “terminal intelligence”. Chip vendors compete on parameters, ecosystems and scenario fit; whoever ties models tightly to real scenarios (office, cockpit, home, security) wins the edge entrance. Market forecasts and funding rounds move up in parallel, and edge AI is moving from demonstration to scaled commercialization.
Edge computing is precisely Junyi Microelectronics’ home field. Our K-series edge motherboards, core boards J-3588, J-3576 and J-3568, and edge computing boxes AI-BOX-3588 and 5G-EDGE-3399Pro provide tens to over one hundred TOPS of on-device compute, capable of running 1B–8B-class models locally. Paired with touch all-in-ones, they form an integrated “model + screen + device” edge solution and serve as a compute carrier for AI PCs, interactive terminals and security scenarios.
Source: https://www.sohu.com/a/1096524382_122055437 (2026-11-03)