02 Energy and Physical Infrastructure Layer
1. Industry Chain Positioning and Value Distribution
“The end of compute is electricity.” The energy and physical infrastructure layer is the most fundamental underpinning that keeps the entire AI industry chain running.
It is easily overlooked in traditional analytical frameworks, yet it has become the biggest physical bottleneck constraining AI capacity expansion. In the “smile curve,” although traditional electricity and rack leasing are utility-style low-margin businesses, clean energy (nuclear/geothermal) and liquid cooling temperature-control systems customized for high-density AI clusters are now experiencing explosive performance growth and valuation premiums.
2. Core Tracks and Technical Bottlenecks
- Energy Supply (Power Shortage Crisis): The power consumption of a 10,000-GPU cluster typically ranges from tens to hundreds of megawatts. In Europe and the United States, aging power grids and severe permitting delays for data centers pose major challenges. Suppliers with stable baseload clean energy sources such as nuclear power have become hot commodities courted by tech giants.
- Liquid Cooling Systems (High Thermal Density Challenge): As NVIDIA’s B200 chip pushes single-card power consumption beyond 1000W, traditional air-cooled data centers are hitting physical limits. Cold Plate liquid cooling is now being deployed at scale commercially, while the more aggressive Immersion cooling represents the ultimate solution for future data centers.
3. Localization Status and Trends
China holds a global absolute advantage in UHV transmission and wind/solar renewable energy construction, and domestic intelligent computing centers face virtually no structural power shortage. In the cooling sector, however, domestic companies such as 英维克 (Envicool) and 曙光数创 (Sugon DataEnergy) are rapidly iterating full-chain domestic liquid cooling solutions through national projects like “East-Data-West-Computing,” with technology levels now on par with or even surpassing overseas counterparts.
4. Benchmark Companies in This Layer
Envicool Technology Co., Ltd.
002837:SZ | ChinaCore Business: Provides full-stack cold plate liquid cooling solutions for data centers, covering cold plates, CDUs, piping, and intelligent control
⚔️ Moat: Self-developed high thermal conductivity micro-channel cold plates plus full-chain system integration capability, has supplied in bulk to leading AI computing clusters, with deep accumulation of engineering validation data
Key Competitors: Goaland, Vertiv, Sugon Data Energy
Shuguang Digital Innovation
872808:BJ | ChinaCore Business: Focuses on high-efficiency cooling technologies for data centers, particularly the research, development, and application of phase-change immersion liquid cooling systems.
⚔️ Moat: An absolute pioneer in domestic immersion liquid cooling technology, with extensive real-world deployment cases in national-level supercomputing centers and intelligent computing centers, leading peers in technical depth and engineering experience.
Key Competitors:
Green Revolution Cooling (GRC)
Unlisted company | United StatesCore Business: Provides single-phase immersion cooling complete solutions to reduce data center PUE
⚔️ Moat: World's first commercial single-phase immersion cooling system, with 140+ patents and high customer stickiness
Key Competitors: Submer, LiquidStack, Iceotope, Goaland
Vertiv
VRT:US | United StatesCore Business: Provide power, thermal management, and IT infrastructure solutions for data centers; a critical infrastructure supplier in the era of liquid-cooled heat dissipation.
⚔️ Moat: One of the world's largest suppliers of critical data center infrastructure, with products covering power distribution, thermal management (air cooling + liquid cooling), and monitoring systems, and deep expertise in liquid cooling system integration.
Key Competitors: