Perplexity CEO Aravind Srinivas said AI models running directly on phones and computers could reduce reliance on large data centers. He explained that current AI depends on centralized GPU-powered facilities, but smaller, efficient on-device models could enable faster processing and improved privacy. Srinivas highlighted “test time training,” where local AI learns user workflows over time. He questioned heavy investments in centralized data centers if edge AI adoption accelerates.
Perplexity CEO Aravind Srinivas said AI models running directly on phones and computers could reduce reliance on large data centers. He explained that current AI depends on centralized GPU-powered facilities, but smaller, efficient on-device models could enable faster processing and improved privacy. Srinivas highlighted “test time training,” where local AI learns user workflows over time. He questioned heavy investments in centralized data centers if edge AI adoption accelerates.