Perplexity unveils hybrid AI system splitting tasks between PCs and cloud
Perplexity has introduced a hybrid inference system that distributes artificial intelligence workloads between personal computers and cloud servers. The company presented the technology at Computex 2026 in Taipei, describing it as an agentic orchestration layer that decides in real time where each AI task should run.
The system evaluates incoming requests and assigns them based on complexity and resource needs. Simple operations such as summarization, formatting, or light classification are processed locally on a user’s device. More demanding tasks that require multi-step reasoning or retrieval-augmented generation are routed to cloud infrastructure. The routing process operates automatically and remains invisible to users during execution.
The design aims to balance privacy, performance, and cost. Local execution reduces data exposure and lowers cloud dependency, while remote processing handles heavier computational loads. Industry participants at the event linked this approach to rising concerns about the cost and energy demands of large-scale inference workloads across global data centers.
Perplexity also emphasized that the system is hardware agnostic. It is designed to work across different chip architectures, including Intel and Nvidia-based systems. The company positioned the platform as a software layer capable of running on a wide installed base of personal computers, expanding available compute capacity beyond centralized cloud infrastructure. This approach aligns with broader industry efforts to distribute AI workloads as inference demand accelerates and strains existing data center capacity.
-
09:45
-
09:44
-
09:30
-
09:15
-
09:15
-
09:09
-
09:00
-
08:54
-
08:45
-
08:38
-
08:30
-
08:19
-
08:15
-
08:00
-
07:53
-
07:45
-
07:36
-
07:30
-
07:19
-
07:19
-
07:15
-
07:00
-
19:07
-
18:57
-
16:21
-
16:04
-
15:40
-
15:20
-
15:01
-
14:40
-
14:20
-
14:02
-
13:40
-
13:19
-
13:03
-
12:45
-
12:30
-
12:15
-
12:00
-
11:45
-
11:30
-
11:24
-
11:20
-
11:15
-
11:13
-
11:06
-
11:05
-
11:00
-
11:00
-
10:57
-
10:52
-
10:45
-
10:42
-
10:30
-
10:17
-
10:15
-
10:00
