中国版"星际之门"获全球关注！远景引领全球AI基建新范式

Summary 1: The Xinghe Base is the article's core project information.

1. The project is planned at 3 GW, comparable to the Abilene core campus of the U.S. "Stargate" project; the first 120 MW phase will be delivered within one year at only one-fifth of U.S. cost, and its data center capacity has already been fully booked by internet giants.

2. SemiAnalysis conducted an on-site visit to the Xinghe Base in Ulanqab in September this year and called its construction speed, project scale, green-power ratio, and vertical integration capability impressive.

Summary 2: Actionable takeaways center on green power and power-supply technology.

1. Through self-built wind farms, dedicated lines, energy storage systems, and a source-grid-load-storage model, the project plans green-power coverage above 80%, with wind capacity at roughly 1.75x data center load and four hours of battery storage.

2. The project uses the world's first commercially operated integrated SST 800V high-voltage DC power system, converting 10kV AC directly into 800V DC and eliminating low-voltage transformers, UPS, and rectification stages.

3. The project has no central chillers; liquid-cooled and air-cooled servers share one water loop and rely on Ulanqab's cold climate for natural cooling. Inner Mongolia has planned 25 GW of data center projects, indicating that compute-power integration is moving from individual projects to a regional industry.

Summary 1: Envision gains strong brand endorsement through an international research firm's report.

1. SemiAnalysis is a well-known semiconductor and AI industry research firm. Its first public on-site visit report on a Chinese data center calls the Xinghe Base China's version of "Stargate" and describes Envision as a leading integrator of new energy systems and a frontrunner in AIDC power systems.

2. The report uses "impressive" to summarize construction speed, project scale, green-power ratio, and vertical integration; this third-party assessment can be converted into brand trust assets.

Summary 2: Product and pricing competitiveness support brand positioning.

1. The brand narrative can highlight compute-power integration and end-to-end vertical integration: self-built wind farms, dedicated lines, energy storage, low-voltage distribution, 800V DC architecture, and prefabricated substations, with roles spanning manufacturer, design integrator, and developer.

2. At one-fifth of U.S. cost, with first-phase delivery within one year and data center capacity fully booked by internet giants, its product R&D and delivery capabilities have been validated by major customers.

3. The world's first commercial SST 800V high-voltage DC system and source-grid-load-storage green-power coverage above 80% can serve as high-end product selling points for AI data center customers.

Summary 1: The growth market and demand shifts are clear.

1. The Xinghe Base is planned at 3 GW, with 120 MW in the first phase delivered within one year; data center capacity has been fully booked by internet giants. Inner Mongolia has planned 25 GW of data center projects, indicating sustained and expanding market opportunities in AI computing infrastructure and green-power support.

2. Demand is shifting from standalone data centers to compute-power integration. With planned green-power coverage above 80%, coordination is required across wind farms, energy storage, dedicated lines, SST 800V DC, prefabricated substations, liquid cooling, and air cooling.

Summary 2: Learnable models and entry points for cooperation.

1. Envision's vertical integration and source-grid-load-storage model are worth studying: selecting sites with better wind resources, installing wind capacity at roughly 1.75x load, and using four-hour battery storage to shift surplus power; using the Tianji weather large model to forecast wind output and the Tianshu energy large model to dispatch storage.

2. In terms of cooperation, its three roles—manufacturer, design integrator, and developer—mean there are entry points in equipment supply, system integration, engineering construction, energy dispatch, and O&M. At one-fifth of U.S. cost, low-cost and fast delivery are key to competitiveness.

3. Competitive warning: major customers have already locked up data center capacity, so latecomers need to secure green power, equipment, and customers earlier.

Summary 1: Product manufacturing and design demand centers on new energy and AIDC power systems.

1. The Xinghe Base requires self-built wind farms, dedicated lines, energy storage systems, low-voltage distribution, 800V DC architecture, and prefabricated substations. It also uses the world's first commercially operated integrated SST 800V high-voltage DC power system, converting 10kV AC directly into 800V DC and eliminating low-voltage transformers, UPS, and rectification stages.

2. For cooling, there are no central chillers; liquid-cooled and air-cooled servers share one water loop and rely on Ulanqab's cold climate for natural cooling. This creates supporting design demand for liquid cooling, air cooling, water loops, and data center structural components.

Summary 2: Business opportunities and digitalization implications.

1. The project is planned at 3 GW, with 120 MW in the first phase; Inner Mongolia has planned 25 GW of data center projects. There are order opportunities in wind power, energy storage, SST, prefabricated substations, DC distribution, and liquid cooling equipment. At one-fifth of U.S. cost, factories need high cost-performance and rapid delivery capabilities.

2. On digitalization, the Tianji physical-AI weather large model forecasts wind speed, wind direction, and wind output, while the Tianshu energy large model coordinates battery charging and discharging. This shows that manufacturers can use AI energy management to improve green-power utilization and system coordination.

3. Envision spans manufacturer, design integrator, and developer, so factories can seek supporting collaboration positions within its vertically integrated chain.

Summary 1: The industry trend is compute-power integration and AI infrastructure system integration.

1. Chinese companies are exploring a path different from the mainstream U.S. model, using new energy system integration as a foundation and promoting AI computing infrastructure through compute-power integration. SemiAnalysis describes Envision as a leading integrator of new energy systems and a frontrunner in AIDC power systems.

2. The project is planned at 3 GW, with 120 MW in the first phase delivered within one year at one-fifth of U.S. cost, indicating strong customer demand for rapid delivery, low cost, and high green-power ratios.

Summary 2: Customer pain points and solutions.

1. Green-power supply pain points are addressed through source-grid-load-storage: self-built wind farms, green-power coverage above 80%, sites with better wind resources, wind capacity at roughly 1.75x load, and four-hour battery storage; the Tianji weather large model forecasts wind output, and the Tianshu energy large model dispatches storage.

2. Power-supply efficiency pain points are addressed through the world's first commercial SST 800V high-voltage DC system, which converts 10kV AC directly into 800V DC and eliminates low-voltage transformers, UPS, and rectification. Cooling pain points are addressed through no central chillers, a shared water loop for liquid- and air-cooled systems, and natural cooling.

3. Service providers can offer solutions around wind farms, energy storage, SST, prefabricated substations, liquid cooling, energy large-model dispatch, and system integration. Vertical integration capability is key to competitiveness.

Summary 1: Platform operators have clear demand for low-cost, fast-delivery, green-power computing capacity.

1. The Xinghe Base is planned at 3 GW, with 120 MW in the first phase delivered within one year at one-fifth of U.S. cost. Data center capacity has already been fully booked by internet giants, showing that major customers are highly sensitive to scale, low cost, and rapid activation.

2. Inner Mongolia has planned 25 GW of data center projects. Platform operators can differentiate tenant recruitment around compute-power integration, source-grid-load-storage, and green-power coverage above 80%.

Summary 2: Operations management and trend judgment provide lessons.

1. Operationally, the project uses self-built wind farms, dedicated lines, energy storage, and a four-hour battery system, with wind capacity at roughly 1.75x load. The Tianji weather large model forecasts wind speed, wind direction, and wind output, while the Tianshu energy large model coordinates battery charging and discharging to improve green-power utilization.

2. Power supply uses the world's first commercial SST 800V high-voltage DC system, converting 10kV AC directly into 800V DC and eliminating low-voltage transformers, UPS, and rectification. There are no central chillers; liquid- and air-cooled systems share one water loop and use the cold climate for cooling.

3. On trend-risk avoidance: U.S. SST technology remains at the roadmap stage, whereas China has commercialized it. Platform operators should monitor technology-route maturity and the pace of customer lock-in to avoid falling behind in power architecture and green-power support.

Summary 1: A new industry development is that China's AI infrastructure is taking a compute-power integration path.

1. SemiAnalysis calls Envision's Ulanqab Xinghe Base China's version of "Stargate" and argues that Chinese companies are using new energy system integration as a foundation to promote AI computing infrastructure through compute-power integration, differing from the mainstream U.S. model.

2. The Xinghe Base is planned at 3 GW, with 120 MW in the first phase delivered within one year at one-fifth of U.S. cost and planned green-power coverage above 80%. Inner Mongolia has planned 25 GW of data center projects, indicating regionalization and scaling trends.

Summary 2: Technology routes and business models merit research.

1. Technology comparison: the world's first commercial SST 800V high-voltage DC system converts 10kV AC directly into 800V DC and eliminates low-voltage transformers, UPS, and rectification, while the U.S. remains at the roadmap stage. Over the past decade, about one-quarter of China's data centers used DC power, while about 97% in the U.S. still use AC UPS.

2. Business model: Envision spans manufacturer, design integrator, and developer, vertically integrating wind farms, dedicated lines, energy storage, low-voltage distribution, 800V DC, and prefabricated substations. Source-grid-load-storage coordination relies on site selection, 1.75x wind capacity, four-hour storage, and the Tianji and Tianshu large models.

3. New questions and implications: no central chillers and a shared water loop for liquid- and air-cooled systems have already been operated at scale in northern China, providing a sample for the U.S. chiller debate. Policy planning, green-power ratios, and power-supply standards are key areas for further research.