Core Viewpoint - The recent merger between Haiguang Information and Zhongke Shuguang is a significant event in the semiconductor industry, marking a shift from fragmented competition to a more systematic approach in domestic computing power, potentially creating a giant with a market value of nearly 400 billion yuan covering the entire industry chain from chip design to server manufacturing and computing power services [2][3]. Group 1: Merger Details - Haiguang Information plans to merge with Zhongke Shuguang through a share swap, which will enhance the domestic server CPU market share, where Haiguang holds 53.6% and ranks first, while also having over 30% of the domestic GPU market, ranking second [3]. - The merger is expected to accelerate the domestic replacement process, leveraging Zhongke Shuguang's strengths in high-end computing, storage, security, and data centers [3]. Group 2: Market Impact - The combination of Haiguang's DCU chips with Zhongke Shuguang's liquid-cooled servers is anticipated to improve cost-effectiveness and challenge the market shares of international giants like NVIDIA and AMD in AI training [3][4]. - The merger may gradually erode the market share of Intel and AMD in China, where their combined market share exceeds 70%, due to the technological synergy and ecosystem created by the new entity [4]. Group 3: Industry Chain Effects - The merger will likely compress the market bargaining power of mid-tier foundries, as the new entity will create a complete ecosystem that reduces reliance on external partners [5]. - Foundries like Hon Hai (Foxconn) and Quanta Computer may face declining bargaining power and profit margins due to the integrated ecosystem formed by the merger [5][9]. Group 4: Cloud Computing Sector - The merger could lead to price wars in the cloud computing sector, as the new entity may offer competitive mixed cloud solutions, potentially increasing pressure on third-party cloud service providers like UCloud and QingCloud [10][12]. - Despite the growth in revenue driven by the demand for computing power, net profits in the cloud computing market have not kept pace, indicating a challenging environment for existing players [11][12].

Core Insights - Qualcomm is leveraging its experience in low-power CPU design from the mobile sector to enter the cloud computing market with a focus on cost-effective and energy-efficient server chips [1][2] - The Falkor CPU architecture is designed to meet performance benchmarks while maintaining low power consumption and silicon area requirements [4][7] - The Centriq 2400 series chips integrate up to 48 Falkor cores on a 398 mm² die, with a thermal design power (TDP) of 120 watts, indicating a power consumption of less than 2.5 watts per core [7][8] Performance and Architecture - Falkor is a 4-wide aarch64 core that runs on a 64-bit Arm instruction set, inheriting features from Qualcomm's previous mobile cores [4] - The architecture includes a unique L0 and L1 instruction cache setup, providing higher instruction cache capacity compared to contemporary cores [9][13] - Falkor's performance in SPEC CPU2017 tests shows a significant lead over Arm's Cortex A72, with a 21.6% advantage in integer operations and a 53.4% lead in floating-point operations [74][76] Memory and Cache Design - The L2 cache of Falkor is 512 KB with 8-way set associative design, serving as an intermediate cache between L1 and the on-chip network [49] - The Centriq's L3 cache capacity totals 60 MB across 12 slices, with a latency of 40.9 ns, which is relatively high compared to competitors [61][65] - Falkor's memory subsystem is designed to handle high bandwidth, with a theoretical memory bandwidth of 128 GB/s [69] System Architecture - The Centriq architecture does not support multi-socket configurations, limiting it to a maximum of 48 cores, which is a strategic decision to focus on mainstream cloud applications [71][72] - The interconnect topology utilizes a segmented ring design, allowing for efficient data transfer between cores and caches [56][58] Future Outlook - Qualcomm aims to re-enter the server market with new CPU offerings, building on the lessons learned from the Centriq experience [90][91] - The company is exploring partnerships for AI solutions, indicating a strategic pivot towards integrating AI capabilities into its server offerings [91]