在人工智能快速发展与国家算力基础设施战略加速推进的背景下，国产AI加速芯片及其云服务能力建设已成为保障技术自主可控、支撑数字经济发展的关键环节。近年来，国内多家科技企业纷纷投入自研GPU/NPU研发，并尝试构建从芯片、集群到云服务的全栈能力。然而，受限于硬件性能、软件生态、系统集成及商业化运营等多重因素，真正实现规模化落地的厂商仍属少数。本报告聚焦同时具备自研AI加速芯片、万卡级AIDC集群管理能力与对外云服务商业化能力的厂商，系统分析当前市场格局、核心竞争要素与发展瓶颈。
在人工智能算力需求呈指数级增长的行业背景下，构建具备自主可控能力与高效运营水平的GPU云平台，已成为支撑国家科技创新与产业智能化转型的关键基础设施。其核心竞争力由硬件自研与云平台综合能力两大维度共同构筑。
以华为昇腾、百度昆仑芯等为代表的自研GPU，凭借其针对AI推理、视频编解码、大数据分析等特定场景的深度定制架构，以及从芯片指令集到上层软件框架的全栈垂直优化，已在真实业务负载中展现出超越通用芯片的显著性价比优势。
Against the backdrop of rapid advancements in artificial intelligence (AI) and the accelerated implementation of national strategies for computing infrastructure, domestically developed AI acceleration chips and their associated cloud service capabilities have become critical components in ensuring technological self-reliance and supporting the growth of the digital economy. In recent years, numerous Chinese technology companies have invested heavily in the independent development of GPUs/NPUs and are striving to build full-stack capabilities spanning from chips and computing clusters to cloud services. However, due to constraints related to hardware performance, software ecosystems, system integration, and commercial operations, only a limited number of vendors have successfully achieved large-scale deployment. This report focuses on vendors that simultaneously possess three key capabilities: independently developed AI acceleration chips, management expertise over AIDC (AI Data Center) clusters at the scale of tens of thousands of cards, and commercially viable external cloud services. It provides a systematic analysis of the current market landscape, core competitive factors, and existing bottlenecks.
As industry demand for AI computing power continues to grow exponentially, building GPU cloud platforms that are both technologically autonomous and operationally efficient has become essential infrastructure for driving national scientific innovation and industrial intelligent transformation. The core competitiveness of such platforms is jointly shaped by two dimensions: hardware self-reliance and comprehensive cloud platform capabilities.
Domestically developed GPUs—represented by Huawei Ascend and Baidu Kunlunxin—have demonstrated significant cost-performance advantages over general-purpose chips under real-world workloads. This stems from their deeply customized architectures tailored for specific AI scenarios such as inference, video encoding/decoding, and big data analytics, as well as their full-stack vertical optimization extending from chip instruction sets to upper-layer software frameworks.
