Perplexity announces hybrid AI system that decides what runs locally or in the cloud

The more consequential detail isn't the hybrid routing itself but who is doing it. Perplexity is a search and answer product, not an infrastructure vendor, and building an inference orchestrator puts it in direct competition with the cloud providers and device chipmakers it currently depends on.

This move lands in the middle of a hardware land grab we've been tracking closely. Nvidia's RTX Spark pitch (covered June 1 via The Decoder) was explicitly about making local agent inference practical on Windows devices, and Perplexity's orchestrator is exactly the kind of software layer that would sit on top of that hardware. The two stories together sketch a plausible near-term stack: capable edge silicon at the device level, with an orchestration layer deciding what stays local and what routes up. OpenAI's AWS distribution deal from the same week adds pressure from the other direction, since cloud-native inference gets cheaper and more accessible just as edge compute matures. Perplexity is betting it can own the routing logic before any single hardware or cloud vendor locks that layer down.

Watch whether Perplexity publishes latency and cost benchmarks broken down by task type within the next two quarters. Without that, the routing intelligence claim is unverifiable, and the product is just a client-side load balancer with better marketing.