Masayoshi Son questions orbital data centers

Masayoshi Son is publicly poking holes in the latest Silicon Valley moonshot: putting data centers in orbit. At a recent SoftBank shareholder meeting, Son said space-based compute would not meaningfully cut costs and would take too long to matter, according to TechCrunch’s recap of an Equity podcast discussion.

The timing of Son’s skepticism lands in the middle of a very terrestrial bottleneck. As TechCrunch notes, the AI industry is “compute constrained,” while building new capacity on Earth has become slower and harder—caught up in power, permitting, and local opposition. That friction is part of what makes orbital data centers attractive as a narrative: if land, water, and grid connections are scarce, then the sky looks like a blank sheet of paper. But orbital hardware is not exempt from the same basic arithmetic; it simply moves it into launch costs, replacement cycles, and the logistics of maintaining equipment that cannot be serviced like a warehouse full of racks.

On the podcast, hosts also pointed out who stands to benefit even if the concept never becomes a mainstream alternative to ground-based facilities. A plan built around constellations of compute satellites implies frequent replenishment, which translates into steady launch demand—business that would flow to SpaceX if it is both the builder and the launcher. The conversation also framed the enthusiasm as part of a broader “neo-cloud” scramble, where companies and investors search for new infrastructure stories as traditional cloud economics tighten and AI workloads grow more expensive to run.

Son’s argument is essentially a race-against-the-clock claim: the next few years matter more than a promise a decade out. That is an awkward message for a market that rewards grand timelines and large addressable-market estimates, but it matches how AI spending is currently being allocated—toward capacity that can be deployed quickly and billed immediately. In that environment, orbital compute risks becoming a financing pitch that benefits the launch and satellite stack long before it benefits customers looking for cheaper inference.

Son did not offer a technical roadmap for how space-based data centers would beat Earth-based ones; he questioned why they would. For now, the easiest way to monetize the idea may be to sell the first launches rather than the first discounted compute hours.