In March 2026, the outline of the 15th Five-Year Plan was officially released, and "deepening the 'Eastern Data and Western Computing' project, building a multi-level computing facility system and a national integrated computing network" was included in the highest-level national planning document. In April 2026, the Political Bureau of the Central Committee proposed the construction direction of the "six networks", incorporating the computing network (water network, new power grid, computing network, new-generation communication network, urban underground pipe network, and logistics network).
Almost at the same time, market rumors suggested that China plans to invest about 2 trillion RMB over five years to build a national AI computing network. In response, the National Development and Reform Commission (NDRC) stated at a press conference in June 2026 that this specific figure was not confirmed, but explicitly clarified that "market forces will play a decisive role in the specific construction process; we will work with relevant departments to strengthen overall coordination and policy guidance." The strategic position of computing infrastructure is beyond doubt.
01. From the 15th Five-Year Plan to the Computing Internet
The computing network is a systematic project driven by national will.
The 15th Five-Year Plan places it in the chapter on Digital China construction, requiring the promotion of cloud-edge-end collaboration and intelligent computing cloud services to reduce the computing costs of the whole society. Supporting policies continue to be implemented: the "Action Plan for High-Quality Development of Computing Infrastructure" issued by the Ministry of Industry and Information Technology (MIIT) and five other departments proposes that by 2025, the computing scale will exceed 300 EFLOPS, with intelligent computing accounting for 35%; the "Action Plan for Computing Interconnection" issued by the MIIT in May 2025 clarifies that by 2028, the standardized interconnection of national public computing will be basically realized, forming a computing internet.
The intensive implementation of policies is transforming into visible construction achievements. In May 2025, China Telecom, China Mobile, China Unicom, and the China Academy of Information and Communications Technology jointly launched the construction of the computing internet test network. By the end of March 2026, China's intelligent computing scale reached 1.882 million P, 2.5 times that of the same period last year; by the end of 2025, 145 trunk optical cables were built among the 8 major computing hubs, forming the "8+10+3" spatial layout of "8 national computing hubs + 10 national computing clusters + 3 computing-power-electricity coordinated development areas".
02. From 100,000-Card Clusters to World-Class Computing Center Clusters
Against this backdrop, the construction of intelligent computing centers is in full swing.
On July 9, 2026, the core node of the National Supercomputing Internet in Zhengzhou announced full-load operation, and the country's first 100,000-card-level super-intelligent converged computing resource pool was officially put into use, capable of providing over 100,000 cards of domestic AI computing power externally. The construction pace of this node was extremely fast: trial operation started in February, deploying 3 sets of Sugon scaleX 10,000-card super clusters; on April 6, the 60,000-card AI for Science computing cluster was put into use; in July, the total computing power exceeded 100,000 cards. Its domestically produced ParaStor F9000 all-flash storage system won first place globally in both the production full-node and 10-node lists of the IO500 at the 2026 ISC High Performance Conference in Germany.
On July 10, Sugon announced the completion and operation of the fully domestic 100,000-card AI super cluster "Sugon 8000 (Dengfeng)", which was synchronously connected to the National Supercomputing Internet. It has completed the optimization of over 300 super-intelligent converged applications, covering more than 20 frontier fields such as large models and robotics, and the second system of the same scale has started development.
The three major telecom operators are also accelerating their computing layout. China Mobile's intelligent computing scale reached 92.5 EFLOPS (FP16, by the end of 2025), achieving full-specification computing capabilities from 100 cards to over 10,000 cards; China Unicom's intelligent computing scale has reached 45 EFLOPS, with a standard rack scale exceeding 1.1 million racks, and 7 100-megawatt-level AIDC parks built; China Telecom's self-owned and accessed total intelligent computing scale reached 91 EFLOPS (by the end of 2025), aggregating general large model corpus data of over 10 trillion tokens and high-quality datasets covering over 14 industries, with a total volume exceeding 500TB.
Beyond the operators, local intelligent computing centers are also being deployed. Among all local computing layouts, Inner Mongolia is forming a world-class computing center cluster relying on its cool climate and abundant wind and solar resources. Network giants Huawei, Tencent, and Alibaba, the three major telecom operators, and financial institutions such as the Agricultural Bank of China have settled here; Moonshot AI and DeepSeek are training AI models on servers in Helinge'er and other regions of Inner Mongolia; Sinopec is training its industrial large language model "Great Wall" here; and NEV (New Energy Vehicle) manufacturers such as XPeng are also training algorithms for autonomous driving here.
Landmark orders in other regions have also been frequently secured. Hubei Mobile's Computing Network Jianghan Plain Big Data Center (Phase I) launched EPC bidding in June 2026, with a total investment of 88.8231 million RMB; the Qinghai Fangzhou Cloud Green Computing Center, with a budget of 1.004 billion RMB, was won by the China Construction Eighth Engineering Division, relying on Qinghai's clean energy to create highly efficient and low-carbon computing facilities. Demand from head internet vendors continues to surge: ByteDance's domestic IDC demand in 2026 is about 1.5GW, Alibaba plans 2GW, and launched the first batch of 500MW bidding in April. Inference computing power has become a new battlefield: China Mobile spent 5.112 billion RMB in the second half of 2025 to centrally procure 7,058 inference AI servers, setting the record for the largest single AI server order of the year.
03. Comprehensive Capacity Expansion Across the Entire Chain: Optical Modules, PCBs, Power Chips, and Liquid Cooling
The surge in computing infrastructure is rapidly transmitting upstream along the industrial chain. From optical modules to PCBs (Printed Circuit Boards), and from power chips to liquid cooling systems, upstream links are generally facing tight situations with overflowing orders and urgent capacity.
Optical modules are the "highways" of the computing network, currently undergoing a generational switch from 400G to 800G and 1.6T. Tongling Xuchuang Guangtong Technology Co., Ltd. plans to invest in the "Phase I Project of Tongling Xuchuang High-Tech Zone Optoelectronic Industrial Park" and the "Phase II Project of Tongling Xuchuang High-Tech Zone Optoelectronic Industrial Park". The total investment for the two phases is 1.12 billion RMB, with full production in July 2026, producing a total of 8 million high-speed optoelectronic devices annually, supported by a 5,000-person capacity team. Dongshan Precision acquired Source Photonics, which is investing a total of 1.2 billion USD (about 8.1-8.7 billion RMB) to implement optical chip and optical module expansion projects in Changzhou and other places. Eoptolink has been continuously expanding capacity from Q2 to Q4 this year, with capacity climbing month by month. The expansion adopts an asset-light production line expansion model, mainly in overseas Thailand and supplemented by domestic Chengdu/Wuhan, focusing on 1.6T LPO/silicon photonics high-end computing modules. HGTECH is under construction with 271 million RMB and fixed assets of 3.095 billion RMB, an increase of 291 million RMB quarter-on-quarter in the first quarter, mainly due to the completion and transfer of the Wuhan park, Thailand Phase I plant, and optical module automated production lines to fixed assets. Hengtong Optic-Electric built an 8 billion RMB super optical module base in Suzhou. Phase I will be completed and put into production by the end of 2026, estimated to release an annual capacity of 4 million 800G + 1 million 1.6T silicon photonics modules, with flexible shared-line switching production for 800G/1.6T.
The demand for PCBs in AI servers shows an explosive trend of simultaneous volume and price increases. Currently, the AI server architecture has completed the transition from traditional CPU architecture to GPU and ASIC computing clusters, posing stringent requirements for the layer specifications, substrate performance, hole structure, line precision, and overall stability of PCBs, directly driving the market demand for high multi-layer boards and high-end HDI boards. WUS Printed Circuit's Thailand production base has completed capacity ramp-up and officially entered the stage of scaled and efficient operation, capable of fully undertaking high value-added PCB orders for AI servers and high-speed communications. Avary Holding plans to raise no more than 9.6 billion RMB, all of which will be used for the Huai'an Qingding AI server and high-speed optical module high-density interconnect HDI build-up board project, to fully expand high-end computing PCB capacity. Shenghong Technology focuses on high value-added PCB tracks such as AI computing equipment and data center switches. The four new plant projects in Huizhou have been implemented in batches and are currently in the stage of capacity ramp-up and mass production. Zhongjing Electronic plans to raise funds to promote two major projects: the Thailand intelligent PCB production base and the Zhuhai Fushan high-density PCB expansion. The Thailand base is expected to achieve initial production in the second half of this year. Shennan Circuits raised funds through private placement, focusing on the Wuxi AI computing electronic circuit products project, focusing on the R&D and production of high-speed, high-density, and high multi-layer computing-specific PCB products, precisely meeting the needs of AI hardware upgrades. Dongshan Precision's PCB business old plant renovation project has been successfully put into production, and new high-end capacity is steadily landing as planned, with the capacity structure continuously leaning towards high-end fields such as AI and high-speed communications.
AI server power chips are rigid demands for computing hardware. Data shows that in 2025, global AI server shipments were about 2.46 million units, a year-on-year increase of 24.3%. Industry data shows that from 2025 to 2027, the AI server power market will grow exponentially, with the scale rising from 7.4 billion USD to 32.5 billion USD, and a CAGR of 110%. Previously, the high-end AI server power chip market was long monopolized by overseas giants. In this round of continuous price increases and prolonged delivery cycles for overseas chips, the supply chain security demands of head OEMs and cloud vendors have increased, and they are actively accelerating the introduction of domestic power chips. Joulwatt's 90A high-current intelligent power stage is suitable for high-performance computing power supply scenarios such as servers and GPUs; SG Micro has broken through 90A high-current DrMOS, 48V DC-DC, high-precision monitoring AFE, and other AI core power chips, cutting into the supply chains of Ascend, domestic head server ODMs, and high-speed optical modules; MORNSUN focuses on high-current intelligent power stages, launching MK6841, MK6850, and other 90A high-power DrMOS products. Adopting advanced packaging and optimized topology, it significantly improves power density and heat dissipation capabilities, effectively reduces board layout area, and precisely adapts to the high-density power supply needs of AI servers. To meet the industry's capacity needs, leading special process chip manufacturer SiEn Integrated announced the investment in a new project to expand production, building a 12-inch digital-analog mixed-signal chip production line with a monthly capacity of 50,000 wafers. SiEn Integrated stated that the launch of the Phase IV project marks that, on the basis of consolidating its two core advantage markets of NEVs (New Energy Vehicles) and industrial control, the company is systematically extending to the two high-growth tracks of AI server power and optical interconnection, forming a dual-wheel drive pattern of "stabilizing the foundation with core main business and expanding boundaries in emerging tracks". SiEn Integrated will further expand the capacity of high-end PMICs (Power Management ICs) to cope with the market demand brought by the global AI computing explosion.
As the power per rack leaps from 6kW to 18kW or even over 100kW, and new intelligent computing clusters require a PUE of ≤1.25, liquid cooling is becoming a "must-have". China Mobile's Hohhot Data Center has built the world's largest single liquid-cooled intelligent computing center for operators. The total project investment is 4.66 billion RMB, deploying about 20,000 AI acceleration cards. The localization rate of AI chips exceeds 85%, including 19,800 training cards and nearly 10,000 inference cards, covering five types of local chips such as Iluvatar, Huawei Ascend, Biren, and Kunlunxin. The intelligent computing scale reaches 6.7 EFLOPS (6.7 billion billion floating-point operations per second). The project adopts energy-saving solutions such as cold plate liquid cooling technology and intelligent busbars, with a PUE as low as 1.15 and a green power usage ratio of 55%.
04. Computing Power is National Power, Chain Linkage is the Future
From the 15th Five-Year Plan including the computing network in national major projects, to the official launch of the 100,000-card cluster at the core node of the National Supercomputing Internet; from the rise of the world-class computing center cluster in Inner Mongolia, to the 88.82 million RMB EPC bidding for Hubei Mobile's Jianghan Plain Big Data Center and the winning bid for the 1.004 billion RMB Qinghai Fangzhou Cloud Green Computing Center; from Yunnan Germanium's expansion of 450,000 pieces of indium phosphide substrates, to Sanan Optoelectronics' mass production of 6-inch InP optical chips. China's AI industrial chain is undergoing a profound transformation empowered by the triple boost of "policy drive + demand pull + domestic substitution". In the future, with the connection of the computing internet and the continuous expansion of green computing hubs such as Inner Mongolia, upstream capacity will be gradually released.