SoftBank CEO Torches Elon Musk’s Orbital Data Centers

On Tuesday, Masayoshi Son did something more useful than mock Elon Musk’s orbital data centers idea: he attacked the business case behind it. That timing matters because the AI infrastructure race is not waiting for decade-scale engineering experiments, and Son is hardly a cautious outsider.

The SoftBank founder and CEO told shareholders that space-based data centers won’t do much to cut costs and will take too long, according to TechCrunch. His sharpest line was also his clearest investment thesis:

“In the battle for AI, the next few years will be far more important than what might happen a decade or so from now.”

That is the core conflict. Musk is selling orbit as the next frontier for AI compute. Son is saying the winner gets decided before orbit becomes commercially useful at scale.

Tuesday’s shareholder question turned orbital data centers into a cost argument

Son’s response came during an annual shareholder meeting for SoftBank’s wireless and broadband division, after a question about whether the company planned anything similar to Musk’s space-based AI infrastructure vision. He praised Musk as a “remarkable agent of change,” but rejected the economics.

His argument was simple. The main benefit of putting data centers in space would be cheaper electricity. But Son said electricity is a small slice of AI infrastructure cost compared with chips and hardware. A Business Insider/Yahoo report based on SoftBank’s translation of his remarks said Son put electricity at about 7% of operating AI infrastructure costs, with chips and other expenses making up the remaining 93%.

That makes the orbital pitch less compelling. Even if space solar power reduced electricity costs, Son argued that transport, maintenance, networking, and latency costs would eat the advantage.

“He who strikes first wins,” Son said.

XOOMAR analysis: Son is not saying orbital data centers are physically impossible. He is saying they are irrelevant to the next phase of AI competition if they arrive years after the decisive compute buildout happens on Earth.

The cost stack is harsher in orbit than Musk’s pitch suggests

The orbital data center debate often starts with power. Space has abundant solar exposure, and AI data centers are consuming more electricity. The supplied Fortune material says global data centers consumed 448 terawatt-hours last year, more than Saudi Arabia, and U.N. researchers estimate that number will double by 2030, mainly because of AI demand.

That makes Musk’s idea understandable. It does not make it near-term infrastructure.

The economic problem is that a data center is not just a power bill. It is chips, hardware, networking, maintenance, upgrades, physical access, and uptime. Son’s 7% electricity figure cuts directly against the idea that cheap power alone can justify building compute in orbit.

The transport math is also brutal. Fortune cited NASA data showing that, as of 2018, launching material to low Earth orbit on SpaceX’s Falcon 9 cost $2,720 per kilogram. Fortune also cited Boon Ooi, a professor at Rensselaer Polytechnic Institute, saying that generating just one gigawatt of power in space would require about one square kilometer of solar panels.

That does not prove orbital data centers cannot happen. It does show why Son’s question lands. If the cost savings are mainly electricity, and the new costs include getting the hardware into orbit and keeping it working there, the burden of proof shifts back to Musk.

SpaceX’s incentive is not neutral

TechCrunch’s Sean O’Kane put the SpaceX angle bluntly on the Equity podcast. He said that when Musk talks about building “a constellation of satellites” that would need replacement every few years to create an “orbital data center,” he is:

“guaranteeing that much more business” for SpaceX.

That matters. SpaceX’s launch business benefits when more objects need to go to orbit. O’Kane also argued that SpaceX’s launch market share is heavily tied to Starlink, saying the company is “80 or 90% of the launch market globally” not only because it outperformed rivals, but because Starlink drives launch volume. Without Starlink, he said, SpaceX might be closer to “20% or 30% of the launch market, or 40%.”

Those are podcast comments, not audited market data. But the incentive analysis is fair. Musk’s orbital data centers would not just be a cloud strategy. They would be demand creation for launches.

XOOMAR analysis: this is the cleanest way to read the hype. Musk’s claim sits at the intersection of AI compute scarcity and SpaceX launch economics. Even if the cloud business case remains unproven, the narrative can support investor enthusiasm around future launch demand.

Son’s skepticism cuts deeper because SoftBank is already all-in on AI

This is not a conservative industrialist dismissing speculative tech. Son built his reputation on enormous, risky bets. TechCrunch’s Kirsten Korosec called it “very ironic” that Son is the skeptic, given SoftBank’s “long history of wild bets.”

SoftBank is also deeply exposed to the AI infrastructure race. The supplied Fortune material says SoftBank is one of the main backers of Stargate, the OpenAI-led infrastructure project that plans to invest $500 billion over four years, with Son visiting the White House in January last year to announce SoftBank’s initial $19 billion commitment. Free Malaysia Today’s supplied report also says SoftBank has committed about US$65 billion to OpenAI and pledged to direct hundreds of billions into data centers and related infrastructure globally.

That makes Son’s dismissal more pointed. He is not rejecting AI buildout. He is rejecting orbit as the place to fight the next round.

This debate also fits the pressure showing up elsewhere in tech hardware and consumer markets. XOOMAR has tracked adjacent cost spillovers in Apple Price Hikes Dump AI Data Center Costs on Buyers and AI Data Centers Turn RAM Prices Against Cheap New PCs. The common thread is compute scarcity. When AI infrastructure becomes the bottleneck, every cost center starts moving.

June’s podcast debate shows every AI forecast has a balance-sheet shadow

TechCrunch framed the Son comments inside a wider Equity podcast discussion that also touched on OpenAI’s custom chips, Groq’s new $650 million funding, and the rise of “neo-clouds.” The point was not just orbital data centers. It was the scramble to rent, build, or package AI compute while supply remains tight.

TechCrunch’s Anthony Ha made the broader point: executives often describe a future that favors their own businesses. Musk’s orbital data center vision would be good for SpaceX. SoftBank’s grounded infrastructure strategy favors SoftBank’s current AI commitments. Sam Altman, who has also criticized orbital data centers as “ridiculous” and said they are not likely to matter at scale this decade, has his own history with Musk and his own OpenAI infrastructure agenda.

That does not mean everyone is lying. It means none of these views are detached.

For investors, this distinction matters. The supplied Fortune material says SpaceX went public earlier this month in the biggest IPO in history, and Morningstar analyst Nicolas Owens wrote that investors were paying a $72 premium on SpaceX’s initial share price for exposure to its orbital data center ambitions. That is exactly how long-dated infrastructure stories work: even before deployment, they can become valuation options.

The next proof point is not a vision deck, it is deployable compute

The near-term test for orbital data centers is not whether Musk can make them sound inevitable. It is whether SpaceX or partners can show a working economic path that beats Earth-based AI campuses on cost, timing, or performance.

For now, the source record supports Son’s view more than Musk’s. Earth-based data centers can be built now. SoftBank is backing that route with huge commitments. Son’s case is that chips, hardware, and speed to deployment matter more than theoretical power savings in space.

A credible challenge to that thesis would need hard evidence: lower total cost per unit of compute, acceptable networking and latency costs, workable maintenance plans, and a deployment timeline that matters before the current AI buildout cycle passes. Without that, orbital data centers remain a capital-market story wrapped around a launch-demand story.

The practical takeaway for AI developers, cloud buyers, fintech firms, and enterprise tech leaders is blunt. Treat orbital compute as a signal of how desperate the AI infrastructure race has become, not as a procurement plan. The next several years still belong to chips, power contracts, terrestrial data centers, and companies that can bring capacity online before the decade-scale dreams arrive.

The Bottom Line

• Son’s critique challenges the economics behind orbital data centers, not just the engineering hype.
• If electricity is only a small share of AI infrastructure costs, space-based compute may offer limited near-term savings.
• The debate highlights how quickly AI infrastructure decisions are being made while longer-term space concepts remain uncertain.