Report Interpretation
The report argues that Huawei has used LogicFolding on a 7nm-class DUV process to attain multicore performance associated with 3nm-class mobile chips. Bernstein expects rising advanced-logic and packaging demand to favor SMIC, NAURA and Piotech.
Summary
Bernstein sees Huawei's Kirin 9050 Pro as a major LogicFolding breakthrough for China's semiconductor supply chain
The report argues that Huawei has used LogicFolding on a 7nm-class DUV process to attain multicore performance associated with 3nm-class mobile chips. Bernstein expects rising advanced-logic and packaging demand to favor SMIC, NAURA and Piotech.
- Kirin 9050 Pro is Huawei's first commercially shipped LogicFolding chip, launched in the Mate XT2 on September 7, 2026.
- Bernstein says the chip's multicore performance exceeds Apple A17 Pro, while single-core performance remains around earlier N4/N5-class chips.
- Geekbench 7 scores rose 14% in single-core and 20% in multicore performance versus Kirin 9030 Pro.
- LogicFolding is presented as a route to lower power through stacked dies and shorter communication paths.
- SMIC, NAURA and Piotech are identified as principal supply-chain beneficiaries and are rated Outperform.
Report Interpretation
Overview
Bernstein examines Huawei's Kirin 9050 Pro, the first commercial implementation of its LogicFolding architecture, and argues that the technology materially improves performance and efficiency despite 7nm-class DUV manufacturing. The report frames the launch as an underestimated advance for China's advanced-logic and packaging ecosystem.
Core views
Huawei launched its second-generation tri-fold Mate XT2 on September 7, 2026, featuring the Kirin 9050 Pro, which Bernstein describes as the first commercially shipped LogicFolding chip. The central conclusion is that LogicFolding enables a 7nm-class, DUV-manufactured chip to deliver overall performance associated with 3nm-class mobile processors. Bernstein argues that this narrows Huawei's performance gap with Apple to roughly three years from about four years in the prior generation, although the Kirin 9050 Pro remains about 30% below Apple's contemporaneous A20 Pro in performance. The report attributes the advance to stacking two dies and shortening critical transistor communication paths. This creates timing slack that can be used to reduce voltage at the same workload or raise clock speed for higher performance. Huawei's roadmap indicates transistor density rises from 155 to 238 MTr/mm², a 55% increase. The Kirin 9050 Pro uses a nine-core 1+2+4+2 CPU design with SMT, raises peak P-core frequency from 2.75 GHz to 3.1 GHz, employs a Maleoon 955 GPU, and includes an NPU capable of running a 30-billion-parameter on-device mixture-of-experts model. Bernstein uses third-party Geekerwan testing and standardized Geekbench comparisons to assess the chip. Geekbench 7 results were 1,813 single-core and 8,159 multicore, versus 1,596 and 6,802 for Kirin 9030 Pro, representing gains of 14% and 20%, respectively. The report says multicore performance exceeds Qualcomm Snapdragon 8 Gen 3 and is comparable with MediaTek Dimensity 9400, while its implied Geekbench 6 multicore result exceeds Apple A17 Pro. However, it explicitly acknowledges that single-core performance remains constrained by transistor-level process limits: an approximately 2,000 Geekbench 6 single-core score is similar to Snapdragon 8 Gen 2, Dimensity 9200, and Apple A14. The report emphasizes that hardware-software co-optimization improves actual user experience beyond synthetic scores. In gaming tests on tri-fold devices, Mate XT2 broadly matched Samsung's Galaxy Z TriFold using Snapdragon 8 Elite. In Genshin Impact, Huawei achieved 59.4 FPS versus Samsung's 59.8 FPS while rendering at 1920x1346 versus 1518x966; its 1% low was 52.7 FPS versus 46.9 FPS. In Ananta, Huawei recorded 39.4 FPS versus 38.9 FPS, while in Wuthering Waves it recorded 38.2 FPS versus 37.7 FPS. The report notes that Mate XT2 used more power in some tests and that the measured performance improvement versus Mate XTs reflects the combined effects of HarmonyOS 7, the new GPU and LogicFolding, rather than isolating each contributor. At matched workloads, Huawei laboratory results cited by Bernstein show power reductions of 66% for the NPU, 58% for the GPU, and 41% for the P-core versus Kirin 9030 Pro; Geekbench 6 CPU power at the same 2.75 GHz clock was 23% lower. The report explains that parallel workloads such as GPU and NPU operations benefit more because additional transistors can create more processing engines operating at lower voltage and frequency. At turbo settings, NPU throughput rises 141%, GPU frame rate rises 42%, and P-core performance rises 18%, but power density can exceed that of the planar predecessor. DSP is identified as a warning case: power fell 25% while area shrank 40%, raising power density by 24%. For the supply chain, Bernstein believes SMIC manufactures the LogicFolding chip on its 7nm DUV process and should be the principal foundry beneficiary as adoption grows. Each Kirin 9050 Pro is said to double logic-wafer consumption versus a planar Kirin 9030 Pro. NAURA supplies etch and deposition equipment to SMIC's 7nm process, while Piotech is described as China's leader in wafer-to-wafer bonding tools for advanced packaging. Bernstein expects the currently proprietary technology to spread gradually to other local fabless companies as the ecosystem matures, supporting demand for advanced logic and packaging.
Analysis framework
Bernstein combines Huawei disclosures and laboratory data with third-party Geekerwan tests, comparing Kirin 9050 Pro against its predecessor and competing mobile SoCs under standardized CPU, GPU, NPU and game-testing conditions. It then links the technical mechanism of die stacking and lower-voltage operation to likely demand for foundry capacity, semiconductor equipment and advanced-packaging tools.
Methodology notes
Advanced-logic and advanced-packaging supply-chain transmission
The report connects Huawei's LogicFolding adoption to higher wafer demand at SMIC and to equipment and bonding-tool demand for NAURA and Piotech.
Cross-platform benchmark and real-device performance testing
Bernstein compares standardized Geekbench, SPEC CPU, Steel Nomad Light, INT8 throughput and game results, while noting that synthetic benchmarks do not capture all software, thermal and system-level effects.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- SMIC (981.HK; 688981.CH)Expected major beneficiary as the foundry Bernstein believes manufactures LogicFolding chips on a 7nm DUV process.
- Strengths
- Each Kirin 9050 Pro is said to double logic-wafer consumption versus a planar Kirin 9030 Pro.
- Risks
- Technology adoption depends on manufacturing cost and available capacity.
- NAURA (002371.CH)Expected beneficiary from supplying etch and deposition equipment to SMIC's 7nm process.
- Strengths
- The report identifies NAURA as a major equipment vendor for the relevant process.
- Risks
- Demand depends on continued advanced-logic production and adoption.
- Piotech (688072.CH)Expected beneficiary from wider LogicFolding adoption and associated advanced packaging.
- Strengths
- Described as the domestic leader in wafer-to-wafer bonding tools.
- Risks
- Demand depends on wider use of the proprietary LogicFolding ecosystem.
Key data
- Mate XT2 launchSeptember 7, 2026Huawei launched the Kirin 9050 Pro in its second-generation tri-fold smartphone.
- Mate XT2 starting priceRMB19,999General sales were set to begin on September 12, 2026.
- Mate XT2 overall performance claim42% above Mate XTsHuawei attributes the improvement to the upgraded chip and HarmonyOS 7 optimization.
- Geekbench 7 score1,813 single-core / 8,159 multicoreVersus 1,596 / 6,802 for Kirin 9030 Pro; increases of 14% and 20%.
- GPU performance1,378 versus 993Steel Nomad Light score, up 39% versus Kirin 9030 Pro.
- NPU throughput67.7 TFLOPS versus 27 TFLOPSINT8 throughput, up 150% versus Kirin 9030 Pro.
- Iso-performance power reduction66% NPU, 58% GPU, 41% P-coreHuawei laboratory comparison with Kirin 9030 Pro.
- Transistor density238 MTr/mm² versus 155 MTr/mm²Huawei roadmap indicates a 55% increase attributed to stacking.
Impact & implications
Bernstein views LogicFolding as evidence that China can improve advanced-chip performance without moving beyond 7nm-class DUV manufacturing. The report expects broader ecosystem adoption over time to increase demand for SMIC's foundry capacity, NAURA's process equipment and Piotech's wafer-to-wafer bonding tools.
Risks
- Manufacturing cost and available manufacturing capacity remain uncertainties for broad LogicFolding adoption.
- The first-generation folded DSP reduced power by less than area, increasing power density by 24% versus the planar predecessor.