
智通财经APP消息显示,AI算力产业链正步入年内密集的事件催化期。2026中国算力大会定于9月11日至13日在河北省廊坊市召开。此次大会以“共织算力网 智启新将来”为核心主题,致力于构建一个集产业交流、创新赋能、供需对接与生态共建于一体的国家级综合平台,旨在全面推动我国算力产业向高质量、集约化及普惠化方向迈进。
8月底,英伟达交出了一份堪称完美的2027财年第二季度财报。该季度公司实现营收962亿美元,同比增幅达106%;其中数据中心业务贡献了890亿美元,同比大增117%,在总营收中占比约92.5%。更让市场振奋的是,英伟达在财报电话会议中首次披露了2028财年营收将增长约70%的中长期指引。CEO黄仁勋直言,公司过去从未提前一年发布业绩指引,这反映出“如今具备了更强的可见度”。此外,英伟达来自云服务领域的未完成订单金额已超过2万亿美元。这份财报的价值不仅限于单季度的亮眼业绩,更重要的是拓宽了全球AI资本支出的预期天花板。
根据集邦咨询最新发布的AI服务器产业研究报告,受AI基础设施建设需求激增的驱动,预计2026年全球九大云端服务供应商(CSP)的资本开支总额将实现约90%的同比增长,且2027年预计仍将保持约50%的增速。
8月22日,2026绿色算力(人工智能)大会在内蒙古自治区呼和浩特市盛大召开。会议期间,呼和浩特与乌兰察布两市携手火山引擎、寒武纪等领军企业,成功集中签约13个重大项目,总投资规模高达1361亿元。这些签约项目广泛覆盖了芯片检测、装备制造、算力中心及词元工厂等多元领域。在大会开幕式上,中国信息通信研究院副院长敖立正式发布了《绿色算力发展研究报告(2026年)》。该报告强调,全国一体化算力网正在加速成型,大模型正推动词元经济蓬勃崛起。数据显示,截至2026年3月,国内日均词元调用量已从2024年初的约1000亿飙升至约140万亿,增幅超过千倍。
在政策端,近期中央网信办等七部门共同印发了《促进数字化绿色化协同转型发展实施思路(2026-2030年)》(简称《实施思路》)。文件提出,要提升算力设施的绿色属性,推广并应用液冷散热、余热回收、低功耗芯片等创新技术,同时探索800V高压直流供电架构在算力设施中的应用,以推进超百千瓦单机柜的部署。此外,文件还强调优化算力设施的绿色运营模式,聚焦数字孪生仿真与多模态感知等核心技术,深化人工智能算法在绿色运营场景中的创新实践,从而建立覆盖算力设施全生命周期的智能化管理体系,实现能效的实时监测、预测与动态调节,以及机房的智能运维。
值得关注的是,随着人工智能大模型的不断演进,AI应用场景正逐渐从简单的聊天和搜索,拓展至智能体、自动驾驶及具身智能等复杂领域。一个日益明朗的趋势是:AI产业的竞争焦点正由“模型竞争”向“算力竞争”转变。在此背景下,算力租赁正逐渐演变为衔接算力基础设施与AI应用的核心枢纽。
一方面,全球AI应用引发的Token消耗量持续攀升,算力需求正加速释放;另一方面,高端GPU供应依然紧缺,且数据中心建设、GPU采购及网络电力等基础设施的投入极为庞大。对众多企业和机构来说,自建算力中心所面临的资金压力与技术门槛不断攀升。因此,通过租赁模式获取算力资源,正成为一种更具灵活性的优选方案。
与此同时,近期英伟达系列GPU的租赁价格频频传出上涨消息,这从供需终端的角度,再次印证了全球算力供不应求的紧张局面。
中信建投在最新研报中指出,AI算力已全面跨入“推理主导”的新阶段,应用端高Token密度场景的爆发式增长拉动了需求猛增,而供给端则呈现高端资源持续紧缺、中低端逐步结构性饱和的分化态势。此外,地缘管制与产业链整体成本的攀升,进一步拉高了高端算力的准入门槛。算力租赁业务正加速放量,迎来量价齐升的千亿级高景气黄金期,预计2026年市场规模将跨越2600亿元大关。产业正从“粗放扩张”加速向“高端溢价与精细运营”转型,马太效应促使资源加速向头部企业集中。与此同时,商业模式也正从传统的裸金属租赁,向Token分成及Token工厂一体化服务演进,叠加锁长单与金融杠杆的赋能,正全面重塑产业确定性并拓宽增长空间。
渤海证券研报认为,国内头部国产大模型企业正不断深化芯模协同开发,将国产算力芯片确立为支撑模型推理的核心硬件底座。随着国内云服务商资本开支的回升,国产算力在投资结构中的占比有望稳步提升,算力租赁及IDC厂商将深度受益于智算中心新建与扩容所释放的业务增量红利。
长城证券此前发表观点称,AI算力行业的长期成长逻辑十分清晰。中科曙光的国产算力路线正加速推进,合肥智算集群等项目有力印证了产业实践的成效。随着高密度算力建设的推进,液冷技术有望实现渗透率的快速提升,该机构持续看好国产算力及液冷产业链的发展机遇。
相关概念股:
智谱(02513):9月初,招银国际发布研究报告指出,智谱(02513)的年化经常性收入增长态势稳健且前景广阔,公司新增用户规模与参与度的各项数据进一步巩固了该机构的乐观判断。基于更为强劲的年化经常性收入表现,招银国际将2025至2028年的收入复合年增长率预测从199%大幅上调至275%,目标价由1503.9港元上调至1985港元,并维持“买入”评级。该机构认为,智谱在模型迭代速度、智能水平与单次任务成本之间实现了良好的平衡,有望继续占据大语言模型行业的Pareto前沿位置,为长期商业化变现留出充足空间。研报数据显示,智谱在11个月内完成了6次GLM迭代,智能指数从32跃升至60;MaaS平台注册用户数激增144%达到740万,付费日活用户大增603%,代币调用量年内累计增长超40倍,同时API平均售价提升约101%。截至2026年8月,公司开放平台及API的年化经常性收入已达16亿美元,较7月初的10亿美元进一步攀升,管理层预期年底将冲击24亿美元。国内10万卡芯片集群的部署以及推理栈的重构,使得单代币成本年内累计下降约80%,算力乘数同比提升14倍,开放平台毛利率在2026年上半年达到24.6%。招银国际重申将2025至2028年收入复合年增长率预测由199%上调至275%,并预计2028年总收入将达381亿元人民币。
天数智芯(09903):其中期业绩显示,得益于云端AI推理算力需求的集中释放,智铠系列实现了突破性增长。今年上半年,该系列收入约为6.54亿元,同比狂飙651.8%;毛利约2.33亿元,毛利率达35.6%,已成为推动天数智芯营收增长的核心引擎。而面向高端训练场景的天垓系列则维持了稳健的增长步伐,上半年收入约2.62亿元,同比增长38%。
金山云(03896):近年来,公司不断加强AI云服务能力的构建,促进传统云计算业务向AI基础设施服务转型。作为国内首批独立云服务提供商,金山云在过去两年里全面拥抱AI:借助小米与金山生态的天然场景优势,AI相关收入持续放量,智算云服务已成长为增长的主要驱动力;同时,公司不断加大AI基础设施投资,夯实从IaaS到MaaS的全栈服务能力。伴随企业对AI应用需求的日益增长,那些兼具云平台实力、客户基础及生态体系的企业,有望在AI产业化浪潮中开拓出全新的增长空间。
粤港湾智算(01396):粤港湾智算(01396)公布了截至2026年6月30日止6个月的中期业绩,收入实现了十倍的跨越。期内实现营业收入约人民币25.6亿元,同比飙升约1,020.9%,成功迈上营收新台阶,展现出强劲的成长动能。盈利能力也实现飞跃:期内利润约人民币2.81亿元,相当于去年全年净利润(约人民币0.56亿元)的5倍,盈利水平实现跨越式跃升并创下历史新高。算力规模方面,公司FP16稠密算力规模突破5万PFLOPS,稳居行业第一梯队,构筑了十万卡级超大算力集群的坚实底座。
本文来源:https://i.ifeng.com/c/8wDWaeo7SXV
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
According to a report by Zhixin Securities, the AI computing power industry is entering a concentrated window of catalytic events for the year. The 2026 China Computing Power Conference is scheduled to take place from September 11 to 13 in Langfang City, Hebei Province. The conference, themed "Weaving the Computing Power Network, Enlightening the New Future," aims to serve as a national-level comprehensive platform for industry exchange, innovation enablement, supply-demand matching, and ecosystem building, driving the country's computing power industry toward high-quality, intensive, and inclusive development.
At the end of August, NVIDIA delivered a flawless quarterly report for its fiscal 2027 Q2. The company achieved revenues of $96.2 billion, representing a year-over-year increase of 106%. The data center business contributed $89.0 billion, surging 117% year-over-year to account for approximately 92.5% of total revenues. More importantly, NVIDIA announced for the first time during its investor conference a mid-term guidance indicating that revenues for fiscal 2028 are expected to grow by roughly 70%. CEO Jensen Huang noted that the company has never previously issued performance guidance a full year in advance, attributing this to "stronger visibility" in the current market. Additionally, NVIDIA reported unfilled orders from cloud service providers exceeding $2 trillion. This report is significant not only for its strong Q2 results but also for its ability to raise the global ceiling on expectations for AI capital expenditures.
The latest AI server industry research report by TrendForce predicts that global capital expenditures for the nine major cloud service providers (CSPs) will grow by approximately 90% in 2026 due to the surging demand for AI infrastructure, with a further 50% increase expected in 2027.
On August 22, the 2026 Green Computing Power (Artificial Intelligence) Conference was successfully held in Hohhot, Inner Mongolia. During the event, Hohhot and Ulanqab cities, in collaboration with companies such as Volcano Engine and Cambricon, signed a total of 13 major projects worth 136.1 billion yuan. These projects cover various domains including chip testing, equipment manufacturing, data centers, and token factories. At the opening ceremony, Professor Wei Peng, a specialist from the Chinese Academy of Sciences, emphasized that the large-scale application of AI models is driving the explosive growth of the token economy.
In terms of policy, the Central Cyberspace Affairs Commission, along with six other departments, recently issued the "Implementation Plan for Promoting the Synergistic Transformation of Digitization and Greening (2026–2030)." The plan outlines several key measures, including improving the green attributes of data centers by promoting technologies like liquid cooling, waste heat recovery, and low-power chips; exploring the use of 800V high-voltage DC power architectures; and optimizing operational models through digital twin simulation and multi-modal sensing to achieve real-time monitoring and dynamic adjustment of energy efficiency.
As large AI models continue to evolve, application scenarios are expanding from basic chat and search to more complex fields such as autonomous driving and embodied intelligence. The focal point of the industry is shifting from "model competition" to "computing power competition," with computing power rental services becoming a critical bridge between infrastructure and applications.
The global demand for tokens is rising, yet high-end GPU supplies remain constrained, and the costs associated with building and operating data centers are substantial. Leasing computing resources has emerged as a more flexible and efficient solution for many enterprises.
The recent price increases for NVIDIA GPUs further highlight the critical demand for computing power.
According to a report by Citic Securities, the AI computing power sector is transitioning into a "inference-dominated" phase, with rapid growth in application-end token density driving a surge in demand. The supply side is characterized by a shortage of high-end resources and a structural saturation of mid-low-end capabilities. The report forecasts that the computing power rental market will exceed 260 billion yuan by 2026, with a shift toward high-end services and refined operations.
Bohai Securities suggests that leading Chinese model companies are strengthening their integration with domestic chips, which are becoming the foundation for these models. The recovery in cloud service provider capital expenditures will further increase the proportion of domestic computing power in investment portfolios, benefiting rental and IDC providers.
Great Wall Securities notes that the long-term growth logic for the AI computing power industry is clear. The progress of Chinese companies like Sugon in constructing high-density clusters using liquid cooling technologies underscores the industry's potential.
Related Stocks:
Zhipu (02513): At the beginning of September, CMB International raised its price target for Zhipu from HK$1,503.9 to HK$1,985, based on strong annual revenue growth. The company has successfully balanced model iteration speed with cost efficiency, positioning itself well for future commercialization.
Tensilica (09903): Its mid-term results showed that the Zhikai series generated revenues of approximately 654 million yuan, a 651.8% year-over-year increase, driven by cloud AI inference demand. The Tianhai series, targeting high-end training, grew by 38% to 262 million yuan.
Kingsoft Cloud (03896): The company is actively transitioning its traditional cloud services toward AI infrastructure, leveraging its Xiaomi and Kingsoft ecosystem to drive growth in AI-related revenues and solidify its full-stack service capabilities.
GBA Computing (01396): GBA Computing reported a ten-fold year-over-year increase in revenue for the first half of 2026, reaching approximately 2.56 billion yuan. The company has established a leading position in the industry with a dense computing capacity exceeding 50 PFLOPS.
Source: https://i.ifeng.com/c/8wDWaeo7SXV If there are any copyright issues, please contact us at 13951431913#139.com or leave a comment on the right.