Perplexity CEO Aravind Srinivas: People want to use their own local hardware

Summary

Perplexity CEO Aravind Srinivas discusses the company's push into local and hybrid AI compute, including a portable computer built with NVIDIA on DGX Spark and a model-routing system that shifts sensitive tasks to local hardware. He argues the industry needs hybrid cloud/local orchestration rather than a binary view on data-center overbuilding. Srinivas also highlights power permits and HBM/memory as key bottlenecks for scaling AI agents.

• Perplexity launched a fully local AI implementation with NVIDIA on DGX Spark.
• A second hybrid compute version routes privacy-sensitive tasks to local models and frontier tasks to cloud.
• Srinivas says local models still trail frontier cloud models, so hybrid orchestration is the right approach.
• Finance users show interest in air-gapped local AI deployments.
• Data-center buildout bottlenecks include power permits and HBM/memory.
• Srinivas argues tapping local power, compute, and memory will be needed to scale always-on AI agents.