The overnight sensation "Space Data Center"

Wallstreetcn
2025.12.17 03:25
portai
I'm PortAI, I can summarize articles.

The core advantages of space data centers focus on four dimensions: energy, cooling, latency, and scalability. Morgan Stanley and Deutsche Bank both released research reports, pointing out that the main challenges faced by space data centers are engineering technology challenges rather than physical limitations. Deutsche Bank believes that space data centers provide entirely new incremental opportunities for launch and satellite manufacturing companies, and if successful, the constellation scale will reach hundreds to thousands of satellites in the 2030s

With the surge in demand for artificial intelligence and the structural bottlenecks faced by ground data centers in terms of energy and cooling, the concept of space data centers is moving from science fiction to reality. SpaceX has confirmed its IPO plans for next year and is seeking a valuation of $1.5 trillion, with Musk strongly supporting the deployment of data centers in space. Wall Street investment banks are beginning to seriously examine the feasibility of this concept.

Recently, Morgan Stanley and Deutsche Bank released research reports almost simultaneously, stating that the main challenges facing space data centers are engineering and technical challenges rather than physical limitations. Google, OpenAI, and Blue Origin are all exploring related technologies, with Google’s Project Suncatcher planning to launch prototype satellites in collaboration with Planet Labs by 2027. OpenAI's Sam Altman had considered acquiring the rocket company Stoke Space, while Eric Schmidt actually acquired Relativity Space, partly due to interest in space data centers.

Analysts point out that space data centers can achieve 40% higher solar energy intensity in orbit compared to the Earth's surface, generating power continuously for 24 hours, while achieving passive cooling through a vacuum environment. The transmission speed of laser links is over 40% faster than ground fiber optics. These advantages theoretically provide the potential to address the pain points of ground data centers.

Deutsche Bank predicts that if technology validation is successful, the scale of related satellite constellations will reach hundreds to thousands in the 2030s, creating entirely new market opportunities for launch and satellite manufacturing companies.

Technical Advantages Drive Space Layout

The core advantages of space data centers focus on four dimensions: energy, cooling, latency, and scalability. In suitable orbits such as dawn-dusk sun-synchronous orbits, solar panels can utilize solar energy around the clock. Due to the lack of atmospheric filtering, energy intensity is 40% higher than on the ground, allowing operators to generate 6-8 times the energy output while avoiding complex ground power grids and battery backup systems.

In terms of cooling, the cooling systems of ground data centers account for 40% of energy consumption and require large amounts of water cooling and piping systems. Jensen Huang, CEO of NVIDIA, recently pointed out that in a 2-ton GPU rack, 1.95 tons is cooling equipment. In space, heat can be dissipated into the vacuum through passive radiators on the satellite's shadowed side, eliminating the need for traditional cooling equipment.

The latency advantage comes from the transmission speed of laser links in a vacuum being over 40% faster than ground fiber optics, thanks to the avoidance of glass refractive index effects and the indirect paths of ground cables. Images, weather, and climate monitoring data generated by satellites in orbit currently need to be transmitted to the ground for processing, consuming large bandwidth and being slow. Edge computing can enable real-time processing in orbit.

Cost and Engineering Challenges to be Resolved

Despite the clear theoretical advantages, space data centers still face dual challenges of cost and engineering. **Rocket launch costs remain high, with the reusable Falcon 9 commercial launch priced at about $70 million. Even considering a 40% gross margin, the actual cost still reaches $30 million, with a cost of about $1,500 per kilogram. According to the Google Project Suncatcher white paper, launch costs need to be reduced to below $200 per kilogram to be feasible, which requires SpaceX's Starship to achieve regular operational status **

Thermal management issues are more complex in the space environment. Although the temperature in space is low, the vacuum environment is a perfect insulator, and heat can only dissipate through radiation (slowly) rather than convection (quickly). The power density of GPUs is extremely high, generating concentrated heat in a small area, and large AI clusters require massive passive heat sinks, necessitating breakthroughs in heat sink design.

Radiation threats accelerate chip aging, with cosmic rays and high-energy protons continuously bombarding satellites. Google’s white paper simulations show that TPU logic cores have strong radiation resistance, but high-bandwidth memory (HBM) experiences errors at lower doses. While wrapping servers in lead or aluminum can solve this, it increases the mass of the satellites.

Space maintenance is highly impractical, and satellites need to use higher-grade "space-grade" hardware to ensure longevity, driving up costs. Although there are proposals for orbital transfer vehicles (OTVs) for maintenance, the cost of equipment capable of replacing components like advanced robotic arms is prohibitively high.

Starlink's Grand Vision

Deutsche Bank predicts that Starlink will surpass 9 million users by the end of 2025, doubling year-on-year, demonstrating strong growth momentum. Musk stated that Starlink will develop an improved version of the V3 satellite data center, equipped with high-speed laser links and a downlink transmission capacity of 1 Tbps, but must be launched by Starship due to size constraints.

Earlier this month, Musk mentioned on the X platform that 1 million tons of satellites would be launched each year, with each satellite having a power of 100 kilowatts, resulting in an annual AI computing power increase of 100 gigawatts. Assuming the V3 satellite weighs between 1200-2000 kilograms, this implies an astonishing scale of 500,000 to 800,000 satellites. Analysts find this number hard to believe and expect that the final Starlink V4 or V5 data center satellites will be significantly larger and more powerful.

Furthermore, Musk plans to build a satellite factory on the Moon, utilizing Tesla's Optimus humanoid robots to launch satellites to lunar escape velocity using electromagnetic railguns, eliminating the need for rockets. The Moon's gravity is only one-sixth that of Earth, and there is no atmospheric drag, requiring less energy to deploy satellites. Musk believes this could ultimately achieve over 100 terawatts of AI computing power, propelling humanity towards a Type II Kardashev civilization.

Investment Opportunities Emerge

From a market size perspective, Deutsche Bank believes that space data centers provide entirely new incremental opportunities for launch and satellite manufacturing companies. Hyperscale cloud service providers have ample capital and can easily fund future deployments, which was a concern for some previous LEO broadband constellation projects. Initial deployment scales are small to validate engineering and economic feasibility, expected to begin in 2027-28. If successful, the constellation scale could reach hundreds to thousands of satellites in the 2030s.

Within Deutsche Bank's coverage, three publicly listed companies are expected to benefit from the space data center concept. Planet Labs has partnered with Google to develop prototype satellites to be launched in 2027, testing TPU cooling and formation flying capabilities. Management stated that this is a competitive process, and the company has the capability to validate high-yield, low-cost satellite production, having built, launched, and operated over 600 satellites Rocket Lab can support the deployment of space data centers in various ways, including Neutron rocket launch services, and more importantly, utilizing its own bus to mass-produce satellites, as well as producing key components such as efficient solar panels and laser optical terminals. Intuitive Machines is acquiring Lanteris to obtain a bus platform capable of accommodating GPU/TPU payloads, with Lanteris having extensive experience in building satellites that require high power and heat dissipation.

On the private company side, the startup Starcloud has raised over $20 million in seed funding from Y Combinator, Andreessen Horowitz, and Sequoia Capital. Axiom Space plans to launch its first two orbital data center nodes by the end of 2025 and has raised over $700 million. Lonestar Data Holdings is developing lunar and space data center infrastructure, with its "Freedom" payload having landed on the moon via a SpaceX rocket