AI has equipped clean energy with a "smart brain", including weather prediction, power prediction, network control, energy storage scheduling, digital twins, etc., turning intermittent wind and solar power into dispatchable energy, ultimately forming a "computing power chasing green power and green power surrounding computing power adjustment" computing power collaborative closed loop.

In the middle of this year, China's photovoltaic installed capacity reached 1286GW, surpassing coal-fired power for the first time and becoming the largest installed power generation capacity in China; However, the global installed capacity of electrochemical energy storage is only 700GWh so far. Compared with the objective demand of 15TWh by 2050, it is currently only 5% completed, and there is still at least 19 times the growth space in the future. ”Zhao Wei, Senior Vice President and Chief Scientist of Sunshine Power Supply Group

On September 23rd, at the Yunqi Conference held in Hangzhou, Zhao Wei, Senior Vice President and Chief Scientist of Sunac Power (300274. SZ), provided the above two sets of data. He judged that AI computing power, as a huge new electricity load, happens to coincide with the window of rapid development of clean energy, and the two largest and most concerned fields of global investment are rushing in both directions.



AI给清洁能源装上了“智慧大脑”，包括气象预测、功率预测、构网控制、储能调度、数字孪生等在内把间歇性的风光变成可调度能源，最终形成“算力追着绿电跑、绿电围着算力调”的算电协同闭环。

“今年年中，我国光伏装机达到1286GW，首次超过煤电，成为我国第一大发电装机；而全球电化学储能装机至今只有700GWh，对照2050年15TWh的客观需求测算，目前仅完成5%，未来还有至少19倍的增长空间。”阳光电源高级副总裁、集团首席科学家赵为

9月23日，在杭州举行的云栖大会上，阳光电源（300274.SZ）高级副总裁、集团首席科学家赵为给了上述两组数据。他判断，AI算力作为新增的巨大用电负荷，恰好叠加在清洁能源高速发展的时代窗口之上，全球投资规模最大、最受关注的两个领域正在双向奔赴。



Photovoltaic Summit and Energy Storage Gap

In Zhao Wei's view, carbon neutrality has become a global consensus and direction of efforts. In the future, clean energy technologies dominated by light, wind, hydrogen, and electric vehicles will contribute more than 60% of carbon reduction, and the demand for clean energy will be at least 3 to 4 times the current installation volume. The surpassing of coal-fired power by photovoltaic installed capacity is not only a turning point in the energy structure, but also a milestone for green electricity to enter thousands of industries and households.

Zhao Wei also quoted the astronomical civilization classification quite scientifically: the standard for judging civilization is not weapons and population, but the energy scale that civilization can mobilize. Today, humans are probably at level 0.73 and have not yet reached the complete planetary civilization. The new productivity paradigm brought by AI is becoming a solver for civilization level problems, while the demand for energy is rapidly increasing.



光伏登顶与储能缺口

在赵为看来，碳中和已是全球共识和努力方向，未来实现过程中，以光、风、氢、电动车为主导的清洁能源技术将贡献60%以上的减碳量，清洁能源需求至少是当前安装量的3至4倍。光伏装机超过煤电不仅是能源结构的转折点，更是绿色电力走进千行百业、千家万户的里程碑。

赵为还颇为科幻地引用了天文学上的文明分级：评判文明的标尺不是武器和人口，而是文明能够调动的能量规模，今天人类大概处在0.73级，尚未达到完整的行星文明，AI带来的新生产力范式正在成为文明级问题的求解器，同时对能量需求快速抬升。



In his view, clean energy provides a green energy base for AI, and GW level wind and solar energy storage can meet the mandatory requirement of 80% green electricity proportion in intelligent computing centers. Without cheap and stable green electricity, computing power cannot be scaled up; Power electronics undertake efficient energy supply, efficiently delivering green electricity to AI chips through 800VDC, solid-state transformers (SST), and wide bandgap devices.

On the contrary, AI has equipped clean energy with a "smart brain", including weather prediction, power prediction, network control, energy storage scheduling, digital twins, etc., turning intermittent wind and solar power into dispatchable energy, ultimately forming a "computing power chasing green power running, and green power surrounding computing power adjustment" computing power collaborative closed loop.

But there are three challenges. Firstly, it is expected that by 2030, the electricity consumption of data centers in China will reach 800 billion kilowatt hours, and the load of AI parks will reach hundreds of megawatts or even GW level. The internal bus tends to be 800VDC DC or even higher. Coupled with the national policy requirement that the proportion of green electricity in hub data centers should not be less than 80%, the constraints on total electricity consumption, power density, bus voltage, and green electricity proportion will be tightened. The power supply system should also consider efficiency, reliability, and low-carbon.



在他看来，清洁能源为AI提供绿色能源底座，GW级风光储可以满足智算中心80%绿电占比的硬性要求，没有便宜稳定的绿电，算力就无法规模化落地；电力电子承担高效供能，通过800VDC、固态变压器（SST）与宽禁带器件把绿电高效送到AI芯片。

反过来，AI给清洁能源装上了“智慧大脑”，包括气象预测、功率预测、构网控制、储能调度、数字孪生等在内把间歇性的风光变成可调度能源，最终形成“算力追着绿电跑、绿电围着算力调”的算电协同闭环。

但挑战有三，一是预计到2030年，我国数据中心用电量将达8000亿度，AI园区负荷达到数百兆瓦甚至GW级，内部母线趋向800VDC直流甚至更高，叠加国家政策对枢纽数据中心绿电占比不低于80%的要求，用电总量、功率密度、母线电压、绿电比例等约束同时收紧，供电系统要兼顾高效、可靠、低碳。



Secondly, solid-state transformers can directly convert 10kV or 35kV AC to 800V DC, significantly reducing weight and land occupation, and improving efficiency. However, they are still on the eve of transitioning from demonstration to scale, and a series of practical issues such as cost, on-site operational reliability, and standard system need to be continuously addressed.

Thirdly, there are four engineering barriers for AI to enter power electronics, including the scarcity of real fault samples, which need to be compensated by generating fault samples through digital twins; The device is controlled in real-time at the microsecond level and cannot be entirely handed over to AI. Traditional control is required for the underlying fast loop, while AI is responsible for slow loop optimization and prediction; There are risks associated with model extrapolation, and physical constraints and security mechanisms must be implemented for hard limiting; The lack of industry standards makes it difficult to determine the responsibility for black box failure, and there is a need for explainable AI and industry standard co construction.

Transforming computing power into power grid assets



二是固态变压器可直接将10kV或35kV交流转换为800V直流，大幅降低重量与占地、提升效率，但目前仍处在从示范走向规模化的前夜，需要持续解决成本、现场运行可靠性、标准体系的一系列现实问题。

三是AI进入电力电子存在四道工程门槛，包括真实故障样本稀缺，需依靠数字孪生生成故障样本弥补；设备是微秒级实时控制，不能全部交给AI，须由传统控制做底层快环、AI做慢环优化与预判；模型外推存在风险，必须做物理约束与安全机制硬限幅；行业标准缺失，黑盒失效责任难以认定，需要可解释AI与行业标准共建。

让算力变为电网资产



Zhao Wei summarized the integration of clean energy and AI into ten trends based on four dimensions: power supply architecture, topology control, operational lifespan, and system collaboration. The goal is to achieve more efficient power supply, smarter design, and more reliable operation, and to achieve deep synergy between computing power and energy.

The most certain change at the level of power supply architecture is DC conversion, and 800VDC will become the new standard for AI computing power supply. The medium voltage is centrally rectified by SST to output 800VDC directly to the cabinet, which is then reduced to chip voltage inside the cabinet. Compared with traditional AC links, it reduces losses, reduces copper materials, and improves reliability. The end-to-end efficiency target is set at over 94%, which will drive new industry opportunities such as SiC/GaN devices, high conversion ratio DC-DC, DC disconnect devices, and safety standards.

SST is expected to enter its commercial first year in 2027, with the core being the use of MFT intermediate frequency transformers to replace traditional 50Hz power frequency transformers. Data centers are the explosive scenario, and there is also a demand for new energy grid connection, overcharging, and flexible distribution networks. The advantages are a 30% land saving and 50% weight reduction, as well as natural compatibility with green power and energy storage access.



围绕供电架构、拓扑控制、运行寿命、系统协同四个维度，赵为将清洁能源与AI的融合总结为十大趋势，目标是供电更高效、设计更智能、运行更可靠，实现算力-能源深度协同。

其中，供电架构层面最确定的变化是直流化，800VDC将成为AI算力供电新标准。中压经SST集中整流输出800VDC直达机柜，机柜内再降至芯片电压，相比传统交流链路损耗下降、铜材减少、可靠性提升，端到端效率目标定在94%以上，并带动SiC/GaN器件、高变比DC-DC、直流开断器件与安规标准等新产业机会。

SST在2027年有望进入商用元年，其核心是用MFT中频变压器替代传统50Hz工频变压器，数据中心是引爆场景，新能源并网、超充、柔性配网同样有需求，优势是占地节约30%、重量减轻50%，且天然兼容绿电与储能接入。



Grid GFM can independently establish voltage frequency, support black start and island operation, and AI sinks to the converter body for millisecond level edge computing and adaptive parameter adjustment. Zhao Wei predicts that the penetration rate of grid connected inverters will exceed 40% by 2028 and gradually become a hard threshold for bidding for new energy grid connection and computing power grid connection.

Long term energy storage is listed as a future standard, and most lithium batteries only last for 2 to 4 hours, which cannot solve the energy storage needs of continuous days without wind or light, nor can it solve the seasonal gap, requiring a combination of multiple technologies. Iron air and iron liquid flow batteries are responsible for 100 hour level cross day energy storage, green hydrogen is responsible for cross season large capacity storage, natural gas is paired with SOFC solid oxide fuel cells for peak shaving, and the efficiency of cogeneration can exceed 90%. By relying on AI unified scheduling, a high proportion of clean energy can be reliably supplied throughout the year.

Zhao Wei stated that the ultimate goal is to promote the synergy of computing and electricity, transforming computing power from a burden on the power grid to an asset of the power grid. Green power direct supply can lock in a low price of 2 to 3 cents per kilowatt hour, directly reducing the total cost of ownership of computing power; EMS scheduling is divided into three layers based on response speed: millisecond second level relying on energy storage and power electronics for fast power regulation, minute hour level through staggered migration of computing tasks, and day season level through computing space selection and cross park scheduling.



构网型GFM可自主建立电压频率，支持黑启动与孤岛运行，AI下沉到变流器本体做毫秒级边缘计算、自适应调参。赵为预计，2028年构网型变流器渗透率将突破40%，并逐步成为新能源并网、算力并网招标的硬性门槛。

长时储能被列为未来标配，锂电大多只有2至4小时，解决不了连续多日无风无光的储能需求，更解决不了跨季节缺口，需要多技术组合。铁空气、铁液流电池承担百小时级跨日储能，绿氢负责跨季节大容量存储，天然气搭配SOFC固体氧化物燃料电池兜底调峰，热电联供效率可超过90%，依靠AI统一调度实现高比例清洁能源全年可靠供电。

赵为表示，最终指向的是算电协同，算力从电网负担变为电网资产。绿电直供可锁定每度2至3毛的低价，直接降低算力总拥有成本；EMS调度按响应速度分三层：毫秒-秒级依靠储能和电力电子快速功率调节，分钟-小时级通过算力任务错峰迁移，天-季节级通过算力空间选址与跨园区调度。



In terms of industrial implementation, Sunshine Power's specific product lines include AIDC intelligent computing center power supply, hydrogen production power supply, and grid type energy storage SST、 Virtual power plants and carbon management platforms have been developed, and a solution architecture of "AI for ALL, source load symbiosis, green power direct connection, and computing power collaboration" has been proposed, which can support the implementation of multiple scenarios such as intelligent computing centers, new energy bases, and virtual power plants.



产业落地上，阳光电源的具体产品线是AIDC智算中心电源、制氢电源、构网型储能、SST、虚拟电厂与碳管理平台等，并提出了“AI for ALL，源荷共生、绿电直连、算电协同”的解决方案架构，可以支撑智算中心、新能源大基地、虚拟电厂等多元场景落地。

(Using AI translation)

（使用AI翻译）