Apple's foldable arrives with a 2nm chip built to run LLMs

Apple's foldable is real and it's called iPhone Duo — $1,999, a 7.6-inch inner display, preorders 16 October and shipping the 23rd. But the part that matters here is the silicon inside it.

The A20 Pro is the first high-volume smartphone processor built on TSMC's 2nm node, and Apple tuned it for models, not just frame rates. The Neural Engine is now dual 16-core — 32 cores, 2x the compute — and the neural accelerators run 8-bit floating point math twice as fast, which Apple pitches explicitly at third-party LLMs running on-device. Memory bandwidth is up 50% through what Apple calls the widest memory interface ever shipped in an iPhone. That last number is the quiet one: bandwidth, not raw FLOPS, is what usually chokes local inference.

The same chip powers the iPhone 18 Pro and Pro Max. On top of it sits the rebuilt Siri — personal-context search across your messages, mail and photos, plus onscreen awareness to act on what you're looking at — running locally where it can and falling back to Private Cloud Compute where it can't. The CPU gets two 20%-faster performance cores, the GPU a 7-core redesign worth up to 40%.

Why it matters: Apple just turned "run someone else's LLM on your phone" into a hardware spec instead of a research demo.