Apple has 10x more on reserve cash than the entirety of OpenAI when they trained GPT-4. I don't think OpenAI could have possibly spent $5B or more for training first version of GPT-4(which was trained before GPT-3.5 gained traction), which is a pocket change for Apple for such a core feature.