On June 23rd, during the annual shareholder meeting of SoftBank Mobile, its telecom subsidiary, SoftBank Group founder Masayoshi Son didn’t hold back. He took a hard look at Elon Musk’s idea of building data centers in space and basically said, “No thanks.” Son’s verdict? The concept might sound cool with all that free solar energy, but when you do the math on total economics, it’s a losing game compared to sticking things on Earth.
During the Q&A session, Son addressed the crowd directly. Sure, the main selling point for space data centers is slashing electricity bills. But here’s the kicker: that’s only a tiny slice of the pie. According to the numbers Son laid out, power costs make up just about 7% of the total lifecycle expenses for an AI data center today. The heavy hitters? That’s where the real money goes—GPU chips, servers, land acquisition, building construction, and network bandwidth. These are the big budget items, not the power bill.
Son didn’t bash Musk personally; he called him a “brilliant change-maker.” However, he pointed out that while space saves on some power, the hidden price tag is astronomical. Think about the insane cost of launching every single piece of equipment into orbit. Add in the nightmare of maintaining hardware up there, plus the lag time from sending data back to Earth. These aren’t just hurdles; they’re roadblocks. Son was blunt: commercializing space data centers is a decade-long project. Meanwhile, the AI race will be decided in the next three to five years. We don’t have time to wait for the stars.
So, where is SoftBank putting its money? Right here on the ground. Son revealed that SoftBank has already dropped roughly $65 billion into OpenAI as a major shareholder. On top of that, they’re planning a massive 45 billion euro facility in France to build Europe’s biggest data center cluster. Even bigger news? They’re eyeing over $500 billion for a new AI infrastructure project in Ohio, USA, aiming to create the “largest scale” setup ever. The strategy relies on nuclear power, ample land, and government support to drive down costs through sheer scale. As Son put it, “We’re building a powerful computing network right here on Earth to give AI companies a rock-solid foundation.”
On the other side of the fence, SpaceX, led by Musk, has been painting a very different picture. In their filings, they’ve outlined a grand vision: using Starship to blast millions of tons of computing gear into orbit. The pitch? Unlimited solar energy and the vacuum of space for perfect cooling. A recent filing to the US Federal Communications Committee showed plans to deploy up to one million computing satellites between 500 and 2,000 km altitude. The system would run on solar power, use laser links for super-fast data transfer, and integrate seamlessly with the existing Starlink internet grid. Musk isn’t shy about his ambition; he calls this the first step toward a “Kardashev Type II civilization,” aiming to break the physical limits of Earth’s power supply for AI.
SpaceX has already designed the first generation of AI satellites, dubbed “AI1.” Each satellite carries swappable compute modules, huge solar panels, and advanced liquid-cooled radiators designed to dump heat efficiently in the vacuum of space. Musk’s timeline? He’s pushing to achieve an annual deployment rate of 1 gigawatt of space-based AI compute by the end of 2027, scaling up as reusable rocket tech matures. His theory is wild: if they can launch 1 million tons of payload annually, they could theoretically build a 100 GW solar AI satellite cluster, potentially reaching a terawatt (TW) of computing power.
To be fair, Son isn’t the only one raising red flags. Sam Altman, CEO of OpenAI, also threw cold water on the idea back in February. He told the world that Musk’s orbital data center concept is “unrealistic.” Between launch costs, economic challenges, and the sheer difficulty of fixing GPUs in orbit, Altman believes it’s unlikely to happen for another decade. That said, he did admit there’s future potential there; we just aren’t ready for it yet.
Right now, the industry vibe is pretty cautious. Most analysts agree that for the foreseeable future, traditional ground-based data centers will remain the undisputed kings of AI computing. If space data centers ever do take off, they’ll likely start in niche areas—like offline training tasks where speed doesn’t matter as much as raw scale—rather than replacing our current infrastructure anytime soon.
