Why the AI phone needs more memory

Every launch event this year has featured the same promise: assistants that rewrite emails, summarise calls and generate images without sending a byte to the cloud. What the presentations skip is what that requires.

A language model small enough to run on a phone still needs several gigabytes of memory resident at all times. Manufacturers have responded by making twelve gigabytes the new floor, and the cost of that memory is the single largest increase in this year's bill of materials.

Analysts expect the trend to continue. As models grow, so does the memory they need, and the phones that cannot keep up will be the ones that quietly lose the AI features from their spec sheets.