I would bet large sums of money that Apple is waiting to make a hardware play. When there's a sufficiently capable and intelligence-dense LLM, they will bake it into custom silicon and ship "the first MacBook with on-device AI, powered by our new I1 chip". Imagine Siri being powered by an LLM running entirely on-device at 10,000+ tokens/sec.

Most consumer tasks don't require a frontier model, and (beyond the app store) Apple isn't interested in being a channel through which a frontier model provider like OpenAI can sell subscriptions to their own model.