“Josh Wolfe predicts that 50% of AI inference queries will eventually be processed locally on-device using flash memory or models etched into silicon, rather than in the cloud.”