The 2nm mobile chip race is led by Apple, with the A20 Pro entering mass production while MediaTek and Qualcomm pursue tape-out efforts; TSMC and Samsung are expanding 2nm wafer capacity, positioning Apple as a dominant customer for these advanced nodes.

MediaTek holds about 31% of the global smartphone chip market by mid-2026, with Qualcomm and Apple at roughly 23% and 19% respectively, and Apple expected to secure a large share of 2nm capacity amid intensifying competition.

Consumers stand to gain from on-device AI delivering longer, private, low-latency interactions without cloud dependence, though this comes with higher immediate power draw and battery considerations.

PrismML drives Bonsai 27B’s on-device fitting, contrasting with Qualcomm’s approach of embedding compression features directly into hardware.

GPU strategy remains central: Qualcomm’s Adreno Neural Fusion adds AI compute to the GPU pathway, while Arm’s Mali G2-Ultra NX treats neural rendering as a hardware-native capability, signaling a shift in mobile GPU competition.

Qualcomm is developing a Hexagon NPU for Mixture-of-Experts models up to roughly 30B capacity, featuring a broad precision range (INT2 to FP16) and a redesigned memory hierarchy to keep per-token compute low across many parameters.

For buyers, ecosystem fit matters: iPhone users can leverage Bonsai 27B today, while Android users will watch Qualcomm’s disclosed roadmap and upcoming Snapdragon Summit details.

Foldables and premium design remain a battleground for Huawei and Xiaomi as they vie with iPhone in flagship segments, while AI features and advanced packaging seek differentiation.

Predictions include mainstream MoE on Android by 2027, dense compression favored by Apple, GPU-embedded AI cores becoming a competitive arena, Samsung’s Exynos credibility being tested, and growing relevance of model-compression vendors.

2nm/3nm efforts by Xuanjie and other local designs under GAA constraints face scale and export-control hurdles; Xiaomi notes economies of scale require tens of millions of units annually.

The core distinction is dense compression on Apple’s side versus sparse MoE on Qualcomm’s side, with trade-offs in throughput, latency, and hardware/software integration.

On-device AI is a sizable, fast-growing market, with 2026 on-device generative AI around $6.2B and projected to nearly $95B by 2035; broader on-device AI is expected to reach about $75.5B by 2033.