Perplexity chief executive Aravind Srinivas presented the firm’s plans for AI platforms that manage several artificial intelligence models, applications and documents while weighing privacy, expense and speed. During Intel leader Lip-Bu Tan’s address at COMPUTEX 2026 on Tuesday, Srinivas presented Perplexity Computer, released earlier this year, as a platform able to employ as many as 20 AI models and direct operations across various applications and files. He noted the system forms a group of agents, applies up to 20 distinct AI models, and directs activity among models, applications and files within one unified platform. Srinivas explained that the agent coordination layer balances intelligence, precision, privacy and expense via hybrid agent-based processing. This setup permits smaller models to operate on local hardware. His comments reflect rising industry attention on performing select AI tasks on personal devices while relying on data centers for heavier workloads. He described current demonstrations as an early stage and predicted greater computing capacity in both data centers and local equipment. Srinivas called the development a key step for engineering in AI and semiconductors and acknowledged Intel’s collaboration. Tan earlier stated that AI is changing device usage and stressed on-device AI as a priority, noting progress with partners to increase device intelligence.
