Masayoshi Son questions Musk's orbital AI data centers, calling the economics upside-down

SoftBank founder Masayoshi Son used SoftBank’s annual shareholder meeting on Tuesday to poke a hole in Elon Musk’s vision of space-based AI data centers. The punchline was blunt: Son asked, "What's the point? What's the benefit of building AI data center in space?" Then he answered his own question with numbers. He argued that electricity accounts for only about 7% of the cost of operating AI infrastructure, while chips required to run and train AI models, plus other expenses, make up the remaining 93%. In other words, if the selling point is lower power bills, Son says orbit does not move the needle enough to justify the complications.

Son also made it clear that lower electricity costs are not the whole bill. Even if electricity is cheaper in space-based setups, Son said any savings would be outweighed by additional maintenance, networking, and latency-related costs that come with operating data centers in orbit. He added that the economic and technical tradeoffs of space-based AI infrastructure could take years to work out. That is the strategic tension: the future of AI compute is moving fast, but the time horizon for proving an orbital approach could be painfully slow.

This isn’t a random tech hot take. It is a capital allocation argument in a moment when nearly every major AI stakeholder is racing to lock in compute capacity and supply chains. Musk has been among the most vocal advocates for putting data centers in space as a potential solution to AI’s soaring energy demands. Earlier this year, SpaceX said it aims to build a “constellation of a million satellites that operate as orbital data centers” and that it had begun hiring engineers to make the vision a reality. The broader idea is that distributing compute and networking in space could help scale AI without being as bottlenecked by terrestrial power constraints.

But Son’s critique flips the usual narrative. If electricity only represents about 7% of AI infrastructure operating costs, then a strategy built primarily on energy savings risks being the wrong lever. He pointed to chips needed to run and train AI models, and other expenses, as the dominant cost drivers at about 93% of total operating costs, leaving less room for an electricity-based thesis to carry the business. From a board and finance perspective, it is a reminder that cost structure determines which improvements matter. Lower energy price per unit matters most when energy is the majority cost. If it is not, the savings can get swamped by the cost of making the system work in a much harder environment.

Son’s other concern is systems reality, not just accounting. Operating data centers in space is not simply “build and connect,” it is “build, maintain, network, and keep latency acceptable under orbital constraints.” Son explicitly cited maintenance, networking, and latency-related costs as additional burdens for space-based data centers. That is a second-order issue executives should care about: even if the hardware works, the services have to perform with the speed and reliability AI workloads demand, and any added delay can translate into lost throughput, worse user experience, or higher operational complexity.

The push for orbital compute is not confined to SpaceX. The source notes that tech billionaires like Amazon founder Jeff Bezos and Google CEO Sundar Pichai have also embraced the idea of space-based data centers to scale AI. Pichai called the notion a “moonshot,” but said that “when you truly step back and envision the amount of compute we're going to need, it starts making sense and it's a matter of time.” That “time” is doing a lot of work. Son’s remarks land on the opposite side of that timeline, emphasizing that the tradeoffs could take years to figure out while the AI industry is actively spending today.

Son’s positioning matters too. After criticizing space-based data centers, he praised Musk as a pioneering entrepreneur. Then Son pivoted to SoftBank’s own strategy: SoftBank is determined to position itself at the forefront of the AI boom. He said, “The winner will be decided in the next some years,” and argued that instead of focusing on “the space where we have no idea what will happen in terms of AI-related business,” SoftBank wants to take a “near-sighted perspective,” becoming a first-comer in AI-related businesses. That framing tells you what is at stake for Son personally and for SoftBank as an organization: the company wants to bet in areas where the path from investment to opportunity is clearer.

The debate also has a visible fault line across major AI leaders. The source says OpenAI CEO Sam Altman slammed the concept of putting data centers in space as “ridiculous.” Altman said in February, “We are not there yet.” He added, “There will come a time. Space is great for a lot of things. Orbital data centers are not something that's going to matter at scale this decade.” Put Altman’s timeline next to Son’s “could take years” comment, and you can see a broader skepticism pattern: the technology ambition is high, but the “at scale” moment may be far enough away that investors and operators should not over-weight it in near-term planning.

For executives, board members, and capital allocators watching AI infrastructure evolve, Son’s remarks are a reminder that megatrends are not just about grand visions. They are about cost shares, engineering realities, and timelines that survive contact with operations. If you are underwriting AI compute expansion, you have to ask which bottleneck matters most today. Son’s answer is that electricity is not the main cost driver, and orbit adds complexities that could erase any benefit. The question now is not whether orbital ideas are imaginative. It is whether they are economically decisive on the timelines that investors and AI companies actually live by.