Global computing power ecosystem competition intensifies, AI market trends towards heterogeneous integration.

date
13:43 16/01/2026
avatar
GMT Eight
The global competition of AI computing power ecosystem is becoming increasingly intense, reshaping the landscape.
Huachuang Securities released a research report stating that according to the China Institute of Communications, it is estimated that the global computing power scale will exceed 16ZFlops by 2030, with the proportion of intelligent computing power increasing from 63% in 2023 (equivalent to 875EFlops in FP32) to over 90% in 2030. The current global computing power ecology competition has intensified, and the AI market may tend towards heterogeneous integration. With the shift in AI computing power demand from "general" to "specialized", tech giants have established UCE and UALink to mimic NVIDIA Corporation's InfiniBand and NVLink, reshaping the market competition landscape with "open interconnection + specialized chips". Against this backdrop, attention should be given to the localization process of computing power in China. Key points from Huachuang Securities are as follows: Global AI computing power revolution, sparking a race for acceleration chips 1) Rapid growth in global intelligent computing power demand: According to the China Institute of Communications, the global computing power scale will exceed 16ZFlops by 2030, with the proportion of intelligent computing power increasing from 63% in 2023 (equivalent to 875EFlops in FP32) to over 90% by 2030. 2) NVIDIA Corporation's computing power layout takes the lead: According to Gartner's data, AI servers equipped with GPUs are expected to account for about 71% of the market by 2024, with NVIDIA Corporation holding a market share close to 90% and AMD around 8%. 3) Optimization of intelligent computing costs driving customized solutions: With the explosion in demand for inference on the application side, Alphabet Inc. Class C, Amazon.com, Inc., Microsoft Corporation, Meta, and other large-scale cloud factories are accelerating the development of ASIC chips. 4) Professional semiconductor design companies holding key ecological positions: In the ASIC market, Broadcom Inc. has established long-term partnerships with ultra-large scale customers such as Alphabet Inc. Class C, Meta, Amazon.com, Inc., holding a 55%-60% market share to rank first, with Marvell Technology, Inc. holding a 13%-15% market share to rank second. NVIDIA Corporation's data center business breakthrough, deepening the CUDA "moat" 1) Building a full-stack computing infrastructure: NVIDIA Corporation's full-stack solution includes the basic CUDA programming model running on all NVIDIA GPUs, along with hundreds of domain-specific software libraries, software development kits (SDKs), and application programming interfaces (APIs). 2) Breaking "Moore's Law", Vera Rubin computing platform released. At the CES conference in January 2026, NVIDIA Corporation redesigned Vera CPU, Rubin GPU, ConnectX-9 network card, BlueField-4 DPU, NVLink-6 switch chip, and Spectrum-6 optical Ethernet switch chip to achieve ultimate synergy from chip to platform level. 3) CUDA becomes the "glue" of NVIDIA Corporation's software and hardware ecosystem: CUDA computing heavily relies on NVIDIA Corporation's GPU architecture for acceleration, with a cumulative download volume exceeding 53 million times, the CUDA ecosystem having 5 million developers worldwide, 40,000 companies, and thousands of generative AI companies. 4) Data center business driving high revenue growth: The data center business has surpassed the gaming business starting from the 2023 fiscal year, becoming the core driver of revenue growth for the company. Corresponding revenue in the 2025 fiscal year reached $115.19 billion, a year-on-year increase of 142.4%, accounting for 88.3% of total revenue. Broadcom Inc., the "hidden champion" in the field of AI-specific chips 1) Broadcom Inc. rises, focusing on ASIC chips: Broadcom Inc. focuses on Ethernet and PCIe as core technologies, leading in the commercial Ethernet switching field and deeply participating in the construction of ultra-large-scale AI. 2) Hardware and software synergy driving high-speed revenue growth for Broadcom Inc.: In the 2025 fiscal year, Broadcom Inc.'s semiconductor business revenue reached $36.86 billion, accounting for 58% of total revenue, with AI business revenue reaching $20 billion, a year-on-year increase of 65%; infrastructure software revenue was $27 billion, benefiting from widespread adoption of VCF, and seeing a year-on-year increase of 26%. Intensified global computing power ecology competition, AI market may trend towards heterogeneous integration 1) NVIDIA Corporation maintains its lead in global semiconductor sales: According to the World Semiconductor Trade Statistics Organization (WSTS), in the third quarter of 2025, global semiconductor sales reached $208.4 billion, with NVIDIA Corporation ranking first with revenue of $57 billion. 2) ASIC competition accelerating towards ecological competition: With the shift in AI computing power demand from "general" to "specialized", tech giants have established UCE and UALink to mimic NVIDIA Corporation's InfiniBand and NVLink, reshaping the market competition landscape with "open interconnection + specialized chips". 3) AMD enhances ecological development, chip performance expected to significantly improve: In June 2025, AMD's MI350 series chip has reached roughly the same level of floating-point performance as NVIDIA Corporation's B200. AMD plans to launch the MI400 series and MI500 series in 2026 and 2027 respectively, promising to achieve a 1000-fold AI performance improvement within four years. 4) Large-scale cloud providers strengthening vertical integration: Alphabet Inc. Class C, Meta, Amazon.com, Inc., Microsoft Corporation, and others are advancing computing power autonomy through self-developed chips, aiming to reduce external dependencies, optimize cost structures, and promote industry development towards specialization and customization. Investment suggestions: Intensifying competition in the global AI computing power ecology, reshaping the landscape Recommendations for A-share listed companies: Cambricon Technologies Co., Ltd. (688256.SH), Hygon Information Technology Co., Ltd. (688041.SH), Moore Thread-U Co., Ltd. (688795.SH), Talkweb Information System Co., Ltd. (002261.SZ), Inspur Electronic Information Industry Co., Ltd. (000977.SZ), Dawning Information Industry Co., Ltd. (603019.SH), Foxconn Industrial Internet Co., Ltd. (601138.SH), Zhongji Innolight Co., Ltd. (300308.SZ), Eoptolink Technology Inc. (300502.SZ), China Mobile Limited (600941.SH), China Telecom Corporation (601728.SH), etc. Recommendations for U.S.-listed companies: NVIDIA Corporation (NVDA.US), Broadcom Inc. (AVGO.US), Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR (TSM.US), Oracle Corporation (ORCL.US), Advanced Micro Devices, Inc. (AMD.US), Arista Networks, Inc. (ANET.US), Dell Technologies, Inc. Class C (DELL.US), Marvell Technology, Inc. (MRVL.US), Vertiv Holdings Co. (VRT.US), Digital Realty Trust, Inc. (DLR.US), CoreWeave, Inc. (CRWV.US), Credo Technology Corporation (CRDO.US), Nebius, Inc. (NBIS.US), Super Micro Computer, Inc. (SMCI.US), etc. Risk Warning Unpredictable technological innovation; intensified industry competition; political risks such as those from the GEO Group Inc.