Masayoshi Son has a reputation for making the boldest possible bets in tech — he famously committed to spending trillions on AI and has said that AGI could arrive sooner than anyone expects. But on one front, the SoftBank founder is drawing a sharp line: he doesn’t buy the idea of building data centers in space.

Speaking at SoftBank Corp.’s annual shareholder meeting, Son was asked whether SoftBank planned to pursue anything similar to Elon Musk’s push for orbital data centers. His response was blunt. The main rationale for space data centers is cheaper electricity, but electricity costs represent only a small fraction of what it costs to run a data center — dwarfed by the cost of hardware like chips. And whatever savings might come from solar power in orbit get eaten up by the cost of launching equipment into space, the logistical nightmare of maintenance, and communication delays between the satellite and Earth. He called Musk a “remarkable agent of change,” but made clear SoftBank would be building its data center capacity on the ground. “He who strikes first wins,” Son said — and winning, in his view, means moving quickly on infrastructure that exists now.

SoftBank has skin in this game. The company has committed around $65 billion to OpenAI and has pledged hundreds of billions more toward data centers and related infrastructure globally. Son also acknowledged the intensifying competition in AI, though he said there was enough room for OpenAI, Anthropic, and Google to all grow significantly. AI is still early, he argued, with room for “ten-fold, a hundred-fold” expansion ahead.

The space data center debate has been gaining momentum across the industry for the past several months. Musk has been the most visible proponent, arguing that ground-based power infrastructure simply cannot keep pace with AI’s energy demands, and that satellites running on solar power could solve the problem at scale. SpaceX unveiled its AI1 satellite in June — a 70-meter orbital data center with 150 kW of AI compute capacity — and the project sits at the center of SpaceX’s pitch to investors as an AI company, not just a rocket company.

Google is also in the picture. Its Project Suncatcher aims to send TPUs into orbit, with plans to have them operational by early next year. Then there’s Starcloud, a startup backed by both Google and NVIDIA, which has already trained an AI model in space using an NVIDIA GPU. Even India has entered the race, with Pixxel and Sarvam collaborating on an orbital data center satellite called Pathfinder, carrying data center-class GPUs and scheduled for launch as early as Q4 2026. China has been moving quickly too, with state-backed initiatives coordinating across government, universities, and private companies to build space computing infrastructure.

OpenAI CEO Sam Altman has been decidedly skeptical of the whole thing. When asked on a podcast whether space data centers would provide meaningful compute for OpenAI in the next two to three years, his answer was a flat “no.” Same for five years. When the interviewer pushed to ten years, Altman laughed and said he wished Elon luck. Musk fired back on X: “He’s right… for OpenAI” — suggesting that the inability to use orbital compute would eventually put OpenAI at a disadvantage.

It’s worth noting the position Altman is in when he says this: SoftBank, OpenAI’s largest backer, is now saying much the same thing. Son’s $65 billion commitment to OpenAI, and the Stargate data center program he’s partly bankrolling, are all bets on terrestrial infrastructure. That alignment between investor and company on this specific question isn’t coincidental — it reflects a shared view that the next few years of the AI race will be decided by who can build the most compute on Earth, fastest, not by who can figure out the engineering of running GPUs in orbit.

Son’s framing at the shareholder meeting was unambiguous: “In the battle for AI, the next few years will be far more important than what might happen a decade or so from now.” Space data centers, in his assessment, are a decade-out proposition at best — and the AI race won’t wait.