Masayoshi Son, the founder and CEO of SoftBank, argued at the company’s June 23 shareholder meeting that Elon Musk’s plan to build data centers in orbit is the wrong answer to the AI compute shortage — and he is not the only one who thinks so.

Son said building orbital data centers won’t do much to cut costs and will take far too long when the real competition is happening right now. “The battle for AI, the next few years will be far more important than what might happen a decade or so from now,” he said, according to Bloomberg, which first reported his remarks.

Why the Criticism Is Unusual

Son is not the most obvious voice for skepticism about bold, expensive technology bets. SoftBank backed WeWork, invested in a wave of pre-revenue startups that later collapsed, and has pledged up to €75 billion to build data centers in France. His own shareholder pitch decks have included projections stretching decades into the future.

TechCrunch noted the irony directly: Son is criticizing a multi-decade infrastructure bet while himself being deeply invested in data center projects on Earth. SoftBank has significant commitments to terrestrial AI infrastructure, which gives it a financial interest in the outcome of this debate.

Sam Altman, the CEO of OpenAI, has also expressed skepticism about the orbital data center concept, though Altman and Musk have a long and complicated history that complicates any clean reading of his views on Musk’s business bets.

What SpaceX Is Pitching

The orbital data center concept involves building a network of satellites that host computing capacity, using Starlink’s connectivity infrastructure as the backbone. Musk has framed it as a way to place compute capacity outside national borders and avoid the regulatory and permitting obstacles that slow data center construction on the ground.

Critics, including Son, point out that satellites need to be replaced every few years, making this a form of ongoing revenue guarantee for SpaceX‘s own launch business. SpaceX holds 80 to 90% of the global commercial launch market, largely due to Starlink. Every orbital data center is also another rocket manifest.

The Broader Compute Race

The debate sits inside a broader moment where everyone with access to compute is finding ways to rent it out. AI chipmaker Groq raised $650 million in June 2026 and is leasing inference capacity to other companies. SpaceX has signed compute deals with Google, Anthropic, and Reflection AI since the latter’s IPO earlier this month. The field is crowded and moving fast.

Son’s comments reflect a growing consensus that Earth-based infrastructure will carry the AI buildout for at least the next decade, whatever gets built in orbit after that.