Author: Xia Mo, Editor: Chao Yi
Globally, an unprecedented wave of AIDC (AI Data Center) infrastructure construction is sweeping across the world.
According to projections by market research firm Dell'Oro Group, global data center capital expenditures will approach the milestone of USD 1 trillion in 2026 and reach USD 1.7 trillion by 2030. Goldman Sachs offers a more optimistic forecast, estimating that cumulative capital expenditure on computing power, data centers, and electricity between 2026 and 2031 could be as high as approximately USD 7.6 trillion.
The industry outlook is clear, yet severe challenges remain. As AI computing density continues to rise, traditional data center power supply architectures face multiple challenges regarding efficiency, space utilization, and dynamic load response. Consequently, 800V high-voltage DC power supply has become a key technological direction for next-generation AI infrastructure.
Institutional estimates indicate that switching a 2.5MW data center from traditional UPS to an SST solution can save over CNY 500,000 in annual electricity costs, free up distribution space for deploying dozens of additional racks, and reduce copper usage by tons. For any AIDC project, this represents not just incremental improvement, but a systematic reconstruction starting from the underlying architecture.
With commercialization accelerating, SST is on the eve of explosive growth. Huaan Securities predicts the global SST market size will surge from CNY 2.82 billion in 2026 to CNY 101.04 billion in 2030. Industrial Securities' forecast is even more bullish, analyzing that the global off-cabinet power supply market space will be approximately CNY 271 billion in 2030; among which, SST penetration based on installed capacity is estimated at around 40%, corresponding to an installation demand of approximately 51.9GW. Calculated at an average price of CNY 3.4/W, its market size would be approximately CNY 176 billion.
Amidst this multi-billion trend, an increasing number of new energy and tech giants are attempting to stake their claim on this strategic high ground. However, in reality, most enterprises are still stuck in the product promotion stage, with only a handful possessing true delivery and operational capabilities. Comprehensive data from third-party sources such as Gaogong Industry Research (GGII) and 24Tide Industry Research Institute (TTIR) shows that while over 80 companies globally have released SST products, fewer than five have completed third-party certification and project delivery, fewer than three have achieved grid-connected operation, and only one has implemented commercial operation in large-scale AI data centers.
This is merely one facet of competition in the global AIDC ecosystem; the real battle has just begun. We believe that in the AIDC infrastructure wave, if enterprises wish to stand at the forefront, what is tested is no longer single-product technology or delivery capability, but whether the enterprise possesses simultaneous capabilities in SST, battery cells, BMS, PCS, power electronics, system integration, energy management, and global delivery.
Whoever builds the next-generation power architecture for AI data centers will likely hold the strategic initiative in this infrastructure investment cycle.
Tech Giants Flock to SST: The "Global First" Goes to China
Globally, an increasing number of energy and tech giants are concentrating their efforts on the SST track.
Overseas giants include Eaton, Schneider Electric, ABB, and Siemens. Within China's grid sector, there is Zhongdian Purui under NARI Technology; in the transmission and transformation equipment sector, there are China XD Electric, TBEA, and TELD; in the PV and storage camp, there are Envision and Sungrow Power; automotive-grade power module and electronic control companies include Inventchip Technology and Leadrive Technology; and low-voltage electrical appliance and connector players like Tianzheng Electric and Luxshare Precision have also entered the field.
In a research report, Sinolink Securities broke down the competitive barriers of SST into three levels: the ability to participate in defining product specifications, medium-voltage system engineering capability, and high-frequency power electronics productization capability. These correspond to different time scales: short-term speed of prototype validation, mid-term accumulation of system-level reliability, and long-term modular mass production capability. "Stable operation of SST requires accumulated practical experience. Jumping from 10kV to 35kV is difficult; enterprises with rich practical experience in SST system project delivery are better positioned to seize the first-mover advantage in AI data center demand."
In the SST and high-voltage DC intelligent computing power supply roadmap, several manufacturers have formed differentiated layouts.
In November 2025, Delta Electronics, together with Meituan and Chindata, released an intelligent DC power supply system solution based on SST, becoming one of the early publicly disclosed commercial cases in the industry. Its single power cabinet outputs 1MW, occupies only 1m², and achieves 98.5% efficiency. In March 2026, Delta further launched a containerized SST DC mobile intelligent computing center, integrating CPU, GPU, and SST into a unified delivery unit. Currently, Delta's SST has completed long-term network testing in data centers of leading cloud vendors.
Envision and Sungrow Power are also at the forefront of R&D and industrial layout in 800V DC power supply systems and SST technology. Particularly, Envision holds an absolute leading position domestically and globally in the commercial operation of large-scale AI data centers.
On November 2, Envision Energy announced that its integrated 800V high-voltage DC power supply system successfully achieved end-to-end connectivity at the Envision Ulanqab Galaxy Base. As the world's first integrated 800V high-voltage DC power supply system put into commercial operation in a large-scale AI data center, the system consists of the Envision Tianshu Controller, Solid-State Transformer (SST), Battery Energy Storage System (BESS), and DC/DC converters, pioneering a new model for AI data center power distribution.
As understood by the author, the Envision Ulanqab Galaxy Base began operations in August this year, with a planned capacity of 3GW. Upon completion, it can support million-card parallelism and million-Petaflop computing power scale, with a green electricity direct connection ratio exceeding 80%.
In fact, Envision's commercial operation of this integrated 800V high-voltage DC power supply system in a large-scale AI data center holds significant strategic importance for the global AIDC industry. It demonstrates not only technical capability and commercial value in the SST field, but also the successful construction of a new generation power architecture for AI data centers through systematic synergy between different products such as the "Envision Tianshu Controller, Solid-State Transformer (SST), Battery Energy Storage System (BESS), and DC/DC."
In this commercial application under real GPU training loads, Envision focused on verifying three core capabilities: first, the full-condition stability of SST converting directly from 10kV medium voltage to the 800V DC bus; second, the support capability of DC energy storage under dual conditions of second-level power gaps and minute-level backup power supply; and third, the local absorption capability of GPU load fluctuations through millisecond-level coordination between SST and energy storage.
"AI data centers require an AIDC power system capable of coordinating renewable energy, energy storage, grid dispatch, and computing loads," said Huang Zhiyong, Senior Vice President of Envision Energy and President of the Energy Storage and AIDC Power System Product Line. "The successful powering of IT loads by this integrated 800V high-voltage DC power supply system marks that Envision's AI power system has closed the complete loop from green electricity access to computing loads. This is a critical step in moving computing-power coordination from policy advocacy to engineering reality."
It is evident that Envision is attempting to define the next-generation power architecture system and competitive rules for AI data centers.
Three Decisive Factors for the AIDC Industry: Full-Stack Self-R&D Capability, System Solution Capability, and Global Delivery Capability
After dissecting the structure and development direction of the global AIDC industrial ecosystem, we believe that full-stack self-research capability, system-level solution capability, and global delivery capability are the keys to winning the future AIDC infrastructure wave.
As is well known, AIDC faces global coordination involving power generation, grids, energy storage, and computing loads. It must meet high-reliability power supply requirements, improve green electricity consumption ratios, and adapt to the dynamic loads of high-power-density racks and GPU clusters. What AIDC tests is no longer the delivery capability of single products, but whether an enterprise simultaneously possesses system-level combat capabilities in SST, battery cells, BMS, PCS, power electronics, and system integration, along with global energy management, project operation capabilities, and global delivery capabilities.
In other words, whoever comprehensively builds the next-generation power architecture for AI data centers may hold the strategic initiative in this trillion-yuan level infrastructure investment cycle.
Currently, actions on the industrial side have clearly delineated strategic divisions.
One type of enterprise tends to quickly fill key links such as data centers, power electronics, and computing-power coordination through investment, cooperation, and industrial chain synergy, attempting to enter the AIDC market via ecological integration. Taking CATL as an example, it has recently made frequent deployments around the DC side of data centers, power electronics, computing infrastructure, and AI application ecosystems. Its logic lies in leveraging advantages in batteries and energy storage systems to extend downstream into data centers and computing-power coordination scenarios, building comprehensive competitiveness from energy storage equipment to energy usage scenarios.
Another type of enterprise emphasizes a self-research closed loop from underlying hardware to system scheduling. Envision has chosen the "full-stack self-research + scenario reconstruction" path. Its judgment is that what AI data centers truly need is no longer a single battery or SST product, but a complete power architecture capable of coordinating new energy generation, grid interaction, energy storage dispatch, and computing loads.
This aligns perfectly with Envision's existing capability foundation. Envision has long laid out wind power, energy storage, power electronics, and energy management systems, forming a technical system covering battery cells, BMS, PCS, SST, energy storage systems, and energy dispatch software. Based on the "Tianshu" energy large model, Envision also attempts to further enhance multi-energy coordination, load forecasting, and system dispatch capabilities.
At this year's Intersolar exhibition, Envision unveiled the next-generation power infrastructure solution within its AI power system, with SST and 800V DC energy storage systems being the core. SST can directly connect the medium-voltage grid with next-generation AI computing architectures, reducing traditional multi-stage conversion links and improving system efficiency; the 800V DC energy storage system addresses the dynamic load demands of high-power-density racks and GPU clusters, providing millisecond-level backup power, power smoothing, and energy dispatch capabilities.
Most critically, Envision is currently the first company globally to commercially operate this solution in a large-scale AI data center project.
As we understand, in the energy system of the world's largest AI computing super-unit built by Envision—the Envision Ulanqab Galaxy Base—SST handles main power and stabilizes bus voltage; DC energy storage acts as a buffer, filling second-level power gaps and providing minute-level backup and hour-level energy shifting; Tianji (meteorological large model) handles prediction, while Tianshu handles dispatch, functioning somewhat like the brain of an autonomous driving grid.
"We often use an analogy internally: the Tianshu controller is like a conductor leading an orchestra to play a concerto, where wind turbines, energy storage, SST, and loads each have their own parts, coordinating at different rhythms," Huang Zhiyong elaborated. The result is that Tianshu sets the plan, and the controllers of each device execute rapidly on-site. Ultimately, the GPU input meets the power curve required for computation, the grid sees a smooth curve, and the green electricity ratio reaches a higher level.
Of course, the principle is easy to explain, but actual implementation is far from simple. It should be noted that computing loads are almost non-adjustable, reliability requirements reach 99.995%, and grid backup is mandatory. Whoever can raise the green electricity ratio higher while ensuring safe and stable terminal power supply truly has the capability. Huang Zhiyong believes the specific difficulties lie in three areas:
Turning prediction into action: Seeing the weather is just the starting point. Accurately calculating how much power can be generated, arranging how to use it, and finally translating this into actions for each device—all steps must stay within the boundaries of power, battery state, temperature, and safety.
DC stability and protection: Several fast-acting power electronic devices are connected to a single busbar; they must be coordinated through simulation modeling and stability analysis. DC fault currents do not naturally cross zero, requiring isolation within milliseconds. With 800V entering the rack directly, safety and protection mechanisms must be redefined and resolved.
From "individual protection" to "joint judgment": Most devices are inherently designed to protect themselves first, cutting out immediately upon detecting issues. In an integrated system, every device must prioritize the overall situation.
Knowing theoretically that a system can run and actually tuning it under real loads are two different things. As we understand, the off-grid system of another large-scale AI data center built by Envision in Chifeng has been operating practically for over a year. Under 100% power electronic equipment conditions, voltage and frequency remained qualified, enduring multiple extreme weather events with no wind or sunlight. The difficulty and value of its SST-based 800V DC power supply system are self-evident.
In actual project verification, the value of Envision's AI power system has been fully corroborated by multi-dimensional quantitative data: In terms of green electricity supply, the system can provide over 60% reliable green electricity for computing loads and reduce comprehensive energy costs by more than 40%; in terms of grid connection efficiency, it can shorten the commissioning period by 3 to 6 months; in terms of grid capacity, by enhancing the effective stable capacity of existing grid resources, it can release up to 40% of available power.
On June 17, 2026, at the European tech event Viva Tech, Envision officially launched the "Mission Gobi" plan, announcing the construction of 5GW-scale green AI computing centers in global Gobi desert regions by 2030. From practical application in Ulanqab to the global 5GW green AI computing center plan, Envision is pushing AIDC power systems from single-project verification to scaled deployment, building stable, green, and economical new-generation infrastructure for the AI era.
Within the overall AIDC boom, there is an indisputable fact: the decisive point in this current competition lies not only in technology and hardware layers. What truly determines an enterprise's market position is a more fundamental question: who enters the architectural design system of AI giants first?
As Pulitzer Prize-winning journalist Gary Rivlin revealed in his book "Silicon Valley AI Dream," the harsh reality is that the training and inference costs of foundation models are astonishingly high; this is a game only giants can afford to play.
Facts confirm this. During an earnings call last August, Jensen Huang stated that the cost of building an AI data center with per GW capacity is approximately USD 50 billion. According to authoritative institutional data, global data center investment reached USD 770 billion in 2025, surpassing the entire photovoltaic industry's investment and upstream oil and gas exploration and development investments.
Additionally, according to public reports, the capital expenditure of the four major US cloud vendors reached USD 284.1 billion in the first three quarters of 2025; domestic enterprises Alibaba and Tencent Holdings spent over CNY 150 billion, among which Alibaba announced in 2025 that it would invest CNY 380 billion over the next three years for AI and cloud infrastructure construction, focusing on computing centers...
Enterprises such as Envision and Sungrow Power are accelerating the co-building of ecological systems with AI giants. For instance, in Chifeng, Envision collaborated with Tencent to create the world's first data center project achieving 100% direct green electricity supply. This project not only constructed a full-link green electricity supply system from power generation to computing power but also reduced the data center's comprehensive energy costs by more than 40%.
At the 2026 Yunqi Conference, Sungrow Power and Alibaba Cloud officially disclosed their computing-power coordination cooperation results. Both parties jointly established a Green Token Joint Innovation Center, exploring new infrastructure paths of source-load symbiosis and direct green electricity connection around the power supply architecture of computing centers.
Looking back at the history of global business development, trillion-yuan tracks never lack participants. In the future, whoever can truly coordinate new energy, grids, energy storage, and computing loads, and build an "industrial community" with top AI enterprises, will occupy a higher position in the AIDC infrastructure cycle and even possess the ability to change the industry landscape and trajectory.
Now, SST and the entire AIDC arena have warmed up globally, while differentiation and competition have just begun.