Not everyone is buying Elon Musk’s vision for orbital data centers. Speaking at a recent shareholder meeting, SoftBank founder and CEO Masayoshi Son poured cold water on the idea, arguing that building data centers in space will not meaningfully cut costs and will take far too long to address the immediate demands of the artificial intelligence race. “The next few years will be far more important than what might happen a decade or so from now,” Son stated, directly challenging the long-term timeline inherent in any space-based infrastructure play.
Why Son Says Orbital Data Centers Miss the Point
Son’s central argument hinges on the concept of urgency. In the current AI landscape, where compute constraints are the primary bottleneck for everything from model training to deployment, the industry faces a pressing need for data center capacity right now. Terrestrial data centers, despite facing challenges around power availability, land acquisition, and regulatory pushback, can be built and brought online within a few years. An orbital data center, by contrast, would require overcoming immense engineering hurdles, including the cost of launching and maintaining hardware, and the logistical challenge of replacing satellites every few years. For Son, a plan that delivers results a decade from now is not a solution to the crisis of the moment; it is a distraction from the near-term fight for AI dominance.
The Ironic Skeptic: SoftBank’s Own History of Wild Bets
The critique carries a layer of irony that was not lost on observers. SoftBank itself has a storied history of placing massive, speculative bets on futuristic technologies, from its outsized investments in WeWork to its Vision Fund’s aggressive pursuit of AI startups. As one industry analyst noted, it is striking to see Son playing the role of the skeptic, given that his own playbook often involves long-shot wagers on transformative ideas. Yet Son’s dismissal of Musk’s plan may signal a pragmatic realization that not all ambitious visions are equally viable, especially when weighed against the immediate, terrestrial realities of the AI infrastructure buildout. SoftBank has been pouring capital into on-the-ground data center projects, including a massive €75 billion commitment to build AI infrastructure in France, making it clear where the company believes the smart money lies in the short to medium term.
Behind the Plan: SpaceX’s Self-Interested Calculus
The debate also highlights a recurring theme in the AI industry: executives and companies are often “talking their own book.” Musk’s vision for orbital data centers, if realized, would be a monumental boon for his other venture, SpaceX. The plan would require launching a constellation of satellites that need frequent replacement, thereby guaranteeing a steady and massive stream of business for SpaceX’s launch operations. This dynamic is hard to overlook. Without Starlink, SpaceX would command a far smaller share of the global launch market. An orbital data center program effectively extends that captive demand indefinitely. This is not to say the idea lacks technical merit, but any assessment of its viability must account for the fact that its strongest proponent stands to benefit enormously—both from the demand it would create for launches and from the opportunity to lease out compute capacity in space, a business SpaceX is already pursuing in a smaller form on Earth.
The same self-interest applies to Son and his peers. SoftBank is heavily invested in terrestrial data centers, so it is natural for Son to advocate for solutions that align with his own company’s strategy. Sam Altman of OpenAI has similarly expressed skepticism about orbital data centers. With so much money at stake, there are no truly impartial observers. Every major prediction about the future of AI infrastructure is, to some degree, a reflection of the business interests of the person making it.
The Core Tension: Immediate Need vs. Distant Promise
The argument over orbital data centers ultimately distills to a single, unresolved tension. On one side, the AI industry is so compute-constrained that nearly any idea that increases capacity deserves a hearing. The promise of a limitless energy source (the sun) and a venue free from earthly NIMBY opposition is undeniably attractive. On the other side, the path to that future is prohibitively expensive, technologically uncertain, and—most critically—slow. For companies and investors trying to win the AI race today, a solution that arrives in a decade is barely a solution at all.
What This Means for the AI Infrastructure Race
For professionals tracking the AI industry, this debate underscores a critical distinction between near-term strategy and long-term vision. The immediate action for those building AI products or managing compute workloads is to watch where the capital is flowing now. SoftBank and others are doubling down on terrestrial data centers, meaning that most compute availability in the next few years will come from earthbound facilities. While orbital data centers remain a fascinating concept, the practical takeaway is clear: the most reliable path to securing AI compute capacity in the foreseeable future is through the traditional, ground-based infrastructure that companies like SoftBank are actively funding. The orbital play, while visionary, remains a distant prospect that does little to solve the capacity crunch facing AI developers today.
