SoftBank's Masayoshi Son: Musk's Orbital Data Centers Won't Decide the AI Race

SoftBank founder Masayoshi Son has publicly rejected Elon Musk's idea of building orbital data centers, calling it pointless in practical terms. According to Son, the winner of the race for AI supremacy will be determined by computing infrastructure on Earth, not by server farms in orbit.

A clash of two visionaries

The idea of orbital data centers emerged as a response to a real bottleneck in AI development: ground-based server capacity is constrained by limited power grids, a shortage of cooling water and a lack of suitable land. Musk, who heads SpaceX and runs the AI company xAI, promotes orbital computing as an unconventional workaround. By his logic, servers above the atmosphere would not compete with cities for electricity, would not need approval from local authorities and could, in theory, tap unlimited solar energy.

Son categorically disagrees. The SoftBank founder, whose company has invested hundreds of billions of dollars in technology companies around the world, from Alibaba to Arm, is convinced that the technical difficulties and enormous costs of orbital infrastructure outweigh any potential advantages. In his view, the AI race will be won by whoever scales computing capacity faster and more cheaply, and that is only possible on Earth.

Why ground infrastructure wins

Under Son, SoftBank has consistently bet on terrestrial AI infrastructure: the company is a strategic partner of leading American AI giants and a participant in large-scale data center projects in several regions of the world. Here, his criticism of Musk is not just talk but a position backed by concrete investment decisions.

The key reasons space data centers lose out to those on Earth:

• The cost of launching payloads into orbit remains high, even with SpaceX's reusable rockets
• Cooling servers in a vacuum with sharp temperature swings is an unsolved engineering problem
• Latency in sending data from orbit to Earth is critical for real-time AI applications
• Repairing and replacing hardware in space is incomparably harder and more expensive than in ground-based data centers
• Ground capacity scales faster: new regions and gigawatts come online in months, not years

Context: the battle for power and land

Behind this public dispute lies a acute industry problem. Data centers for training and inference of large language models consume enormous amounts of electricity: the largest facilities already draw as much as small cities. A shortage of land near power hubs and the difficulty of obtaining construction permits for data centers have become among the main constraints for the entire AI industry. This is what pushes some entrepreneurs toward unconventional solutions, from building small nuclear reactors next to server halls to the concept of orbital computing clusters.

Musk promotes the idea of "servers in the sky" through several of his companies: SpaceX would handle transporting hardware to orbit, and xAI would manage computing workloads. However, the biggest players in the AI industry (Microsoft, Google, Amazon, Meta) are choosing a different path: aggressive expansion of ground-based hyperscale infrastructure. New data centers are being built at record speed in several regions at once. SoftBank is moving in the same direction, and Son's position directly reflects that strategy.

"The AI race will be won by whoever wins on Earth,"

Masayoshi Son said in an interview with Bloomberg.

What this means

The debate between Son and Musk exposes a fundamental split in strategic bets: scale proven but resource-hungry ground infrastructure, or invest in fundamentally new solutions with high risk and high potential. For now, the industry is voting with its money for Earth: trillions of dollars of AI infrastructure capital spending are going there. Son's position is not mere skepticism but a strategic statement about where the fate of the AI race will be decided.

*Meta is recognized as an extremist organization and banned in Russia.