孫正義對決馬斯克談太空數據中心：這場辯論揭示的算力基礎設施

In a rare public clash between two of Asia's and America's most prominent tech investors, SoftBank founder Masayoshi Son dismissed Elon Musk's concept of orbital data centers as having "little value," arguing that terrestrial infrastructure — particularly in regions with abundant renewable energy — is the more practical path for the next generation of AI compute.

Knowledge point: what makes a good location for compute

The debate is not really about space vs. ground — it is about the fundamental constraints of data center design. The three most important factors are: energy (both quantity and cost), cooling (heat dissipation is the limiting factor for density), and latency (the time it takes for data to travel between the compute and the user).

Space-based data centers offer one theoretical advantage: unlimited solar energy and radiative cooling (space is cold). But they face severe disadvantages: launch costs, maintenance impossibility, and latency — a signal to low Earth orbit and back takes about 25-50 milliseconds, which is acceptable for batch processing but problematic for real-time AI inference. Ground-based data centers in places like Iceland (geothermal cooling), the Pacific Northwest (hydro power), or the Middle East (solar) can achieve excellent energy efficiency without the latency penalty.

The real strategic question

Son's critique reflects a broader industry debate about "compute geography." As AI models grow exponentially, the physical location of data centers becomes a strategic asset. Countries are competing to host them through energy subsidies, tax incentives, and regulatory frameworks. The knowledge lesson: where computation happens is becoming as important as who owns the chips. Understanding the physics of latency, cooling, and energy transmission is essential literacy for anyone following the AI industry.