Perplexity AI unveils hybrid local-cloud inference system at Computex 2026

Perplexity AI unveiled a hybrid local-cloud inference system at Computex 2026 that decides in real time which parts of an AI task run on a user’s device and which go to cloud models. The company demonstrated the feature with Intel, showing its Personal Computer agent keeping confidential materials local while sending less sensitive, harder reasoning work to frontier models. The feature is expected to launch in the coming weeks.

The announcement reflects a broader shift toward hybrid AI architectures that balance privacy, latency, cost, and model capability instead of relying entirely on the cloud. If it works as described, Perplexity could help define how AI agents operate on next-generation PCs, while giving chipmakers and software companies a clearer path to monetize local inference.