Our Take
Son raises a fair timing objection, but he's not impartial: SoftBank has massive bets on terrestrial data centers and stands to lose if orbital alternatives work.
Why it matters
The compute shortage is driving wild speculation about solutions. When executives attack rival visions, follow the money—their objections often reflect what threatens their own business, not objective truth.
Do this week
Infrastructure teams: audit your data center roadmap assumptions against both the timeline and the financial interests of who's making claims about future capacity.
SoftBank's CEO publicly doubts orbital data centers
Masayoshi Son, founder and CEO of SoftBank, argued at a recent shareholder meeting that Elon Musk's vision for orbital data centers won't substantially reduce costs and will take years to deploy. His core objection: "In the battle for AI, the next few years will be far more important than what might happen a decade or so from now."
Son's skepticism carries weight. He runs a company with a well-documented appetite for bold bets, yet he's publicly questioning what others have begun to treat as inevitable. The dissent matters because it reflects broader uncertainty in the industry about whether space-based infrastructure solves a real problem or simply chases novelty because terrestrial constraints are severe.
Musk has pitched orbital data centers as a way to bypass earthbound limitations: environmental permitting delays, power grid bottlenecks, and NIMBYism. An orbital constellation would require regular satellite replacement, which would in turn guarantee sustained launch demand for SpaceX.
Everyone talking about orbital data centers has something to sell
The skeptic here is not impartial. SoftBank has invested heavily in terrestrial data center projects. If orbital alternatives prove viable, Son's existing bets lose optionality and value. Similarly, Musk's pitch directly benefits SpaceX's launch business; Sam Altman (OpenAI CEO) has rolled his eyes at the idea, and he too has baggage—a complicated history with Musk and OpenAI's own infrastructure spending to protect.
This pattern repeats across the industry. Groq, facing compute constraints, is pitching its own rental model. SpaceX signed its first post-IPO compute deal. Every actor in the current compute squeeze is "talking their own book," as one TechCrunch editor put it. The industry is so constrained that anyone with idle capacity or a plausible path to new capacity is positioning it as a business.
Son's objection is valid on its merits: orbital infrastructure requires years to build, test, and scale, while AI demand accelerates in months. But his timing critique doubles as a defense of terrestrial data center economics, which SoftBank owns.
Pin your infrastructure decisions to timelines, not hype
When evaluating compute sourcing or data center partnerships, separate the technical claim from the financial incentive. Ask: who profits if this solution succeeds, and who profits if it fails? Son's concern about timeline is real. Orbital launch cadences and satellite replacement cycles are not fiction. But his conviction that terrestrial dominates the next five years is also self-interested.
The safe move is neither to dismiss orbital outright nor to wait for it. Build modular assumptions about capacity. Negotiate multi-year terrestrial contracts with clear exit clauses tied to actual orbital deployment timelines. Require vendors—whether they pitch space or ground—to stake their claims on verifiable milestones, not vision statements.