Bernstein’s September 2026 assessment describes a meaningful advance for Huawei, not parity with Apple. Its Kirin 9050 Pro reportedly edges past the older A17 Pro in one CPU benchmark, while the newer A20 Pro retains a substantial overall lead. Bernstein estimates the Kirin uses technology equivalent to a 7nm process; reporting describes it as made without EUV lithography.
1
8
What the Apple comparisons show
Against the 3nm A17 Pro: The Kirin 9050 Pro scores higher in Geekbench 6 multicore, according to reporting on Bernstein’s assessment. That measures work spread across CPU cores, not every aspect of chip or phone performance. The available reporting does not establish a Kirin win in Geekbench 6 single-core or provide a comparable set of single-core scores for all three chips.
1
28
Against the 2nm A20 Pro: Bernstein’s reported estimate puts the Kirin about 30% behind overall. That is not presented as a Geekbench 6 single-core result, and it should not be treated as one. Nor should Geekbench 7 scores reported elsewhere be mixed with the Geekbench 6 multicore comparison.
6
12
15
On Bernstein’s broader measure, Huawei’s estimated mobile-chip lag has fallen from roughly four years in the previous generation to three years—a one-year improvement, not a claim that it has caught Apple’s latest processor.
1
How LogicFolding could help without EUV
Huawei’s LogicFolding architecture, part of its Tau Scaling Law approach, places logic in connected layers and aims to shorten signal paths rather than relying only on smaller transistors. It is a possible way to improve chip design while access to EUV manufacturing equipment remains restricted. The reported benchmark result alone cannot show how much of the gain comes from stacking rather than other design changes.
24
29
38
Huawei reports that transistor density rose from about 155 million to 238 million transistors per square millimeter, approximately 55%, compared with the Kirin 9030 Pro. At comparable performance, it also reports power reductions of 66% for the NPU, 58% for the GPU and 41% for a CPU performance core. These are company-reported, component-level measurements—not independent findings that the whole phone uses less power in every workload.
22
What still needs testing
Independent, like-for-like Geekbench 6 runs would make the Apple comparison clearer, particularly for single-core performance. Sustained speed, temperatures, power use in shipping phones, manufacturing yield and production cost also remain important open questions. A multicore benchmark win and Huawei’s reported component measurements do not settle them.
1
21
22