🛰️ Overview #
Beijing’s Municipal Science and Technology Commission and the Zhongguancun Science City Management Committee have announced an ambitious Space Data Center system targeting a 700–800 km twilight orbit. The system aims for gigawatt-level power, hosting million-card server clusters, echoing Google’s recently revealed Project Suncatcher.
🏗️ System Architecture #
The proposed system consists of three major subsystems:
- Space computing power
- Relay transmission
- Ground control
Each Space Data Center module is designed with:
- ~1 GW power capacity
- Ability to support million-card computing clusters
- Scalable, constellation-based deployment
📅 Development Roadmap #
🚀 Phase I (2025–2027) – “Space Data, Space Compute” #
- Breakthroughs in energy supply and heat dissipation for orbital data centers
- Development and testing of experimental satellites
- Build Phase I computing constellation
- Target capability:
- 200 kW total power
- 1000 POPS computing performance
🛰️ Phase II (2028–2030) – “Ground Data, Space Compute” #
- Advances in on-orbit assembly & construction
- Reduced launch and operational costs
- Deployment of Phase II constellation
🌌 Phase III (2031–2035) – “Space-Based Main Compute” #
- Mass production of satellites
- Networked launches
- On-orbit docking to form large-scale, modular Space Data Centers
❄️ Natural Advantages of Space Data Centers #
Space-based computing infrastructure offers inherent benefits:
☀️ Energy #
- Unlimited solar power
- No ground-based power grid constraints or land limitations
🧊 Cooling #
- Low-temperature space environment allows high-efficiency heat dissipation
- Significant reduction in energy costs for cooling compared to Earth
🛰️ Current Progress: Chenguang-1 #
The consortium has already:
- Achieved multiple core technological breakthroughs
- Completed development of the first experimental satellite “Chenguang-1”
- Entered final assembly and testing
- Planned launch window: late this year or early next year
🌞 Google’s “Project Suncatcher” #
Space-based AI compute is not China’s idea alone—Google has announced its own program:
🌐 Objective #
A large-scale AI data center in space powered by near-continuous solar energy.
🧠 Key Technologies #
- Satellites equipped with custom TPU (Tensor Processing Unit) chips
- Vast satellite network for distributed compute
- Solar energy in orbit yields up to 8× efficiency over ground-based solar farms
Sundar Pichai emphasized the abundance of solar power in space:
“The energy emitted by the sun is 100 trillion times greater than the total electricity generated by humanity.”
⚡ Why Space? #
As AI models grow exponentially in size:
- Compute demand rises
- Energy consumption becomes unsustainable
- Ground grids face increasing strain
- Environmental costs escalate
Space-based systems offer:
- Almost uninterrupted sunlight
- Far lower reliance on batteries
- Radically higher energy generation efficiency
🚀 Google’s Deployment Plan #
Google is taking a phased, pragmatic approach:
🛰️ Phase I (2027) #
- Partnership with Planet
- Launch of two prototype satellites
- Mission goals:
- Test TPU hardware in the space environment
- Validate machine learning performance
- Demonstrate optical inter-satellite links for distributed ML tasks
Space may soon become the next battleground for large-scale AI computing—driven by energy, efficiency, and the race toward exascale and beyond.