Perplexity AI Inc. is developing a software platform designed to split artificial intelligence processing between local personal computers and remote cloud servers to meet rising demand for AI compute capacity. The company presented the concept during Computex in Taipei, with its chief executive explaining how the system would operate.
Aravind Srinivas, Perplexity's CEO, described the platform as functioning like an air-traffic controller for AI tasks. In real time the system decides which portions of AI workloads can be executed on a user’s PC and which segments require the higher horsepower of cloud-based servers, Srinivas said in an interview with Bloomberg Television.
The stated objective of the architecture is to lower the cost of running AI-enabled applications. Srinivas said the platform is intended to deliver greater efficiency in compute consumption and to mitigate the expense associated with routing all processing through large centralized models.
"You don’t want all your compute centralized in servers and everything running through the largest models," Srinivas said. "You’re already reading reports of how people are freaking out about their cost. Some people are spending half a billion dollars per month. What you actually want is for efficient value per watt per user."
Srinivas introduced the platform alongside Intel Corp. CEO Lip-Bu Tan. The article notes that Intel leads the market for PC processors. Perplexity said the system will operate with other technology as well, including processors from Nvidia Corp., and Srinivas said the company intends to remain "chip agnostic."
Perplexity, which operates an AI search product, reported revenue growth from $100 million to $500 million and said headcount rose 34%, according to a message Srinivas posted on X in April. Srinivas added that improvements from larger competitors helped spur Perplexity's expansion.
"Every time any of the AI gets better, our unified system also gets better because we route across all of them," he said.
The announcement frames the platform as a cost-focused routing layer that can combine local PC compute with cloud resources, while remaining compatible with multiple chip vendors. Details on technical implementation, deployment timelines, or commercial availability were not provided in the remarks summarized in the article.
Top-line takeaway: Perplexity is building a hybrid routing platform for AI workloads to allocate tasks between PCs and cloud servers, announced at Computex, with the aim of lowering compute costs and supporting multiple processor ecosystems.