SoftBank CEO Masayoshi Son just shot down Elon Musk’s orbital data center vision: “What’s the point?” Power is only 7% of AI infra cost. The real battle is terrestrial. “He who strikes first wins.” Two of the biggest AI investors in the world disagree on where to put the servers.
Masayoshi Son — SoftBank Shareholder Meeting
“In the battle for AI, the next few years will be far more important than what might happen a decade or so from now. He who strikes first wins.”
The Debate
Musk’s thesis: Put data centers in orbit. Solar power is unlimited in space. No land costs. No grid constraints. Starlink provides the connectivity. SpaceX provides the launch. The ultimate vertical integration.
Son’s rebuttal: Power is only ~7% of AI infrastructure cost. Chips and other expenses are the other 93%. Any power savings from space get eaten by launch costs, maintenance, communication latency, and the sheer complexity of operating servers in orbit. Build on Earth. Build now. Build fast.
Son called Musk a “remarkable agent of change” — then dismissed his orbital plan and said SoftBank will focus on “formidable” terrestrial data center capacity.
The key insight: Son’s 7% number is the data point that matters. If power is only 7% of AI infra cost, then Wall Street’s $11.6B bet on power IPOs is betting on a small slice of the cost stack. And Musk’s orbital thesis — built entirely on solving the power problem — is solving a problem that accounts for less than 1/10th of the total. Chips are the real cost. RAMageddon is the real crisis.
Two Visions of AI Infrastructure
Musk — Build in Space
• Unlimited solar power
• No land or grid constraints
• SpaceX launches + Starlink connectivity
• Timeline: ~10 years
Son — Build on Earth, Now
• Power is only 7% of cost
• Chips are 93% — solve that first
• “He who strikes first wins”
• Timeline: next 2-3 years
The Structural Read
SON IS PROBABLY RIGHT FOR NOW
The AI race’s critical window is the next 2-3 years — model capability is doubling every 3 months. Whoever has the most compute NOW wins. Orbital data centers are a 2035 story. SoftBank’s $100B+ Stargate investment in terrestrial data centers is the near-term play that actually matters.
MUSK IS PROBABLY RIGHT LONG-TERM
If AI compute demand keeps growing exponentially, Earth’s power grid can’t keep up. Data centers already consume 3-4% of US electricity, projected to double by 2030. At some point, terrestrial power becomes the binding constraint. Musk is building for that inflection — the question is whether it arrives before or after the current AI race is decided.
THE SPACEX STOCK DECLINE IS SON’S PROOF
SpaceX fell from $225 to $153 post-IPO. Son’s implicit argument: the market is telling you Musk’s vision is overpriced. SoftBank’s terrestrial bet is the pragmatic position. The $60B Cursor deal and orbital data center plans may be brilliant strategy — or may be the hubris that takes SpaceX public at too high a price.
The Bottom Line
Masayoshi Son and Elon Musk agree on one thing: AI infrastructure is the most important investment of the decade. They disagree on where to build it. Son says Earth — power is only 7% of cost, chips are the real constraint, and the race is won in the next 2-3 years. Musk says orbit — unlimited power, no grid limits, and SpaceX is the only company that can get there. The market is siding with Son for now: SpaceX stock is down 30% from its peak while SoftBank doubles down on terrestrial compute. But Musk has been right about the long-term before. The question is whether “long-term right” matters when the race is decided in the short term.
Sources: TechCrunch, Bloomberg, Fortune — June 2026