On-device AI, as the name suggests, is AI implemented within the device without going through cloud servers. Fast operation is possible in low-latency environments because information is processed within the terminal device, and most security issues can be resolved as it does not go through a central server. It is expected that new AI devices will also support on-device services.

The key element in implementing on-device AI is the proprietary AI chip. Many tech companies are now developing their own AI chips for △fast hardware for real-time AI services △securing stable GPU supply for mass production △optimizing device design. In the past month, major companies have emphasized their own designed chips along with on-device AI.

Firstly, Apple unveiled iOS 17 in September, introducing on-device AI features (auto-correction & dictation) based on their own 'Apple Silicon' chip. Similarly, Google's recently released smartphone Pixel 8 features a custom silicon chip 'Tensor G3', and Samsung Electronics also highlighted 'on-device AI' as their AI strategic keyword at the annual Developer Conference (SDC) on 5th.

Meta plans to offer 'Lama2' as on-device AI with Qualcomm, targeting application in 2024. They plan to support Lama2 on the Qualcomm Snapdragon platform to provide AI-based services without internet connectivity on smartphones, PCs, tablets, cars, etc.

Therefore, recent speculative articles suggesting 'OpenAI is exploring making its own artificial intelligence chips and has gone as far as evaluating a potential acquisition target' are gaining traction. Reuters reported this on 5th, with multiple insiders stating, “since at least last year it discussed various options to solve the shortage of expensive AI chips”