Apple's 2nm A20 Pro bets the iPhone franchise on on-device AI

Apple used its "Surprise and Shine" event on 9 September to put a new chip, not a new feature, at the centre of its pitch to investors and buyers: the A20 Pro, the first high-volume smartphone chip built on TSMC's 2-nanometre process. It ships in the iPhone 18 Pro, the iPhone 18 Pro Max, and Apple's first foldable, the iPhone Duo, which starts at $1,999. iPhone 18 Pro prices rise $100 across the range. Apple also refreshed its smartwatches and launched AirPods 5; the standard iPhone 18 will not arrive until spring 2027.

The chip's headline number is the Neural Engine: an added engine takes it to 32 cores, which Apple says delivers twice the compute of its predecessor with new 8-bit floating point support. The CPU pairs two "supercores," 20% faster than before, with four efficiency cores, all carrying neural accelerators. The GPU moves to a seven-core layout, up to 40% faster, with double the FP8 performance for running language models directly on the phone, backed by 50% more memory bandwidth. Apple says the result is 40% more sustained performance than the A19 Pro and double the A18 Pro.

Apple has also committed to making some iPhone chips at TSMC's Arizona fabs, tying its most advanced silicon to US manufacturing capacity rather than relying solely on Taiwan.

An Apple executive called the A20 Pro "the ultimate chip for running advanced on-device models," adding it has "the widest memory interface we've ever shipped in an iPhone," according to MacRumors.

The launch lands as Apple works to catch up in the AI race; in July it briefly overtook Nvidia as the world's most valuable company, according to CNBC.

Apple is making a clear strategic choice: rather than chase cloud-based AI infrastructure against Microsoft, Google and Nvidia, it is putting the compute in the customer's pocket. Doubling Neural Engine performance and FP8 throughput signals a push to run serious language models locally, which would let Apple sell AI capability as a hardware upgrade cycle rather than a subscription or data-centre cost. That plays to its strengths in chip design and retail pricing power, evident in the $100 price rise it is willing to pass on to buyers.

The risk is execution and timing. The iPhone Duo at $1,999 tests whether foldables can be a mainstream Apple category rather than a niche, and the delay of the standard iPhone 18 to spring 2027 stretches the product cycle. Securing 2nm capacity first, and tying some production to TSMC's Arizona fabs, gives Apple a supply-chain edge over rivals still queuing for the same node, but it also raises the stakes: if on-device AI does not visibly outperform cloud rivals, the chip advantage will not translate into share gains.