Forget the land, the future strongest battlefield for AI is in the "space data center," and Musk will create an ecological closed loop

Gavin Baker believes that space data centers will become the most important technological breakthrough in the next three to four years.

On December 9th, senior technology investor Gavin Baker elaborated in a recent podcast interview that from a first-principles perspective, space data centers outperform terrestrial data centers in every aspect.

According to Gavin Baker's assessment, it may take five to six years for space data centers to become the mainstream in deployed computing power. Until then, ground infrastructure will remain the primary support for AI development.

The advantages of space data centers stem from three core elements: energy, cooling, and chips. In the space environment, the cost structure and efficiency of these three factors will undergo fundamental changes. Satellites can remain in sunlight for 24 hours, with solar radiation intensity 30% higher than on Earth, resulting in a total irradiance six times that of Earth, and without the need for battery storage.

Baker pointed out that this development trend is closely linked to the technological advancements of SpaceX, a company owned by Elon Musk, and that the deep integration of SpaceX, Tesla, and xAI will provide unique advantages for the development of space data centers.

Technological Advantages of Space Data Centers

Gavin Baker analyzes from first principles that the core inputs for operating data centers are electricity, cooling, and chips. In the space environment, the cost advantages of the first two are particularly significant.

In terms of energy, satellites can maintain 24-hour sunlight through orbital design, with solar intensity 30% higher than that on the Earth's surface, resulting in a total irradiance six times that of Earth. More importantly, due to continuous power supply, systems do not need to be equipped with batteries, which represent a significant cost in terrestrial data centers. Therefore, the lowest-cost energy in the solar system is actually space solar energy.

Cooling systems occupy a large portion of rack weight and volume in terrestrial data centers, involving complex HVAC systems, cooling distribution units, and liquid cooling technologies. However, in space, cooling is almost free—it only requires installing radiators on the shadowed side of the satellite, utilizing the near absolute zero environment of space for efficient heat dissipation. This eliminates one of the most complex and costly components in terrestrial data centers.

In terms of network connectivity, space data centers also have advantages. Racks in terrestrial data centers are connected via fiber optics, essentially transmitting lasers through optical cables. In space, lasers can be used to connect satellites directly in a vacuum, which is faster than fiber optic transmission and can build a network that is faster and more coherent than terrestrial data centers.

From the user experience perspective, space data centers can also shorten data transmission paths. Current AI queries need to go through mobile phones, base stations, fiber optic networks, metropolitan aggregation facilities, and finally reach data centers, where calculations are completed before returning along the same path.

If satellites can achieve direct connections to mobile phones—Starlink has already demonstrated the ability to connect directly to cellular networks—data transmission will be greatly simplified, providing a lower latency and lower cost user experience.

Implementation Path and Technical Challenges

The main obstacle facing space data centers is launch capacity. A large number of Starships are needed to transport data center equipment into orbit, which is a key bottleneck in realizing this vision.

SpaceX maintains a leading position in this field, with Starship being the only launch vehicle capable of economically supporting the construction of space data centers.

Notably, Musk recently stated that Tesla, SpaceX, and xAI are moving towards integration, and Baker believes this provides a unique advantage for the development of space data centers.

In this integrated system, xAI will become the intelligent module of Tesla's Optimus robot, Tesla's vision system will provide perception capabilities, and SpaceX will build data centers in space, providing computing power support for xAI, Tesla, and other customers. This vertical integration creates significant competitive advantages, with each company creating value for the others.

For training models, due to the large scale, the application of space data centers may take a longer time.

However, for inference tasks, the application prospects of this technology are clearer. Each satellite can be viewed as a rack, or larger satellites can accommodate equipment for three racks. Baker stated that as technology matures, even training tasks will eventually migrate to space.