Apple plans to renew its push for on-device AI, leaning on 15 years of custom silicon experience to run models locally on iPhones, Watches, and Macs rather than in expensive data centers, according to The Information. As part of an existing deal with Google, Apple is using a version of Gemini to train a smaller distilled model that runs on its devices. Apple has also considered acquiring Liquid AI, a Cambridge, Mass. startup specializing in local AI. Complex queries will still go to the cloud, where some new Siri requests will run on Gemini inside Google Cloud, with Apple recently approving Nvidia's confidential compute privacy technology for the setup. Apple spent $12.72B on capex last year, well below Meta's $72B and Microsoft's $88B. Analyst Richard Kramer of Arete Research estimates Apple has $50B worth of on-device compute “funded by users.”
