SoftBank's Masayoshi Son rejects Elon Musk's orbital data centers as AI race heats up on Earth

SoftBank CEO Challenges Space-Based AI Compute Solutions

Masayoshi Son, the billionaire founder and CEO of SoftBank Group, has publicly dismissed Elon Musk's vision for orbital data centers, arguing that the AI race will be won with terrestrial data centers rather than extraterrestrial infrastructure. Speaking at an annual shareholder meeting for SoftBank's mobile unit on June 23, 2026, Son told investors that the costs of space-based AI compute solutions would likely exceed any benefits, and that the battle for AI dominance will be decided much closer to home1

1

2

2

. "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 stated, emphasizing the urgency of building Earth-based AI infrastructure now rather than betting on orbital facilities that may take years to materialize3

3

.

The Economics Behind Rejecting Orbital Data Centers

Son's skepticism rests on a straightforward cost analysis. While acknowledging that orbital data centers could reduce electricity costs through direct solar power access, he pointed out that power represents only a small fraction of total data center operating expenses4

4

. The bulk of costs come from hardware, chips, servers, and other infrastructure. Any savings on electricity would be offset by the substantial expenses of launching and maintaining equipment in space, plus the communication delays inherent in operating servers hundreds of kilometers above Earth1

1

. According to NASA data from 2018, SpaceX's Falcon 9 rocket still cost $2,720 per kilogram to transport materials into low Earth orbit3

3

. Experts have noted that generating just one gigawatt of power in space would require approximately one square kilometer of solar panels, creating massive weight and launch cost challenges3

3

.

SoftBank's AI Infrastructure Push on Earth

While calling Elon Musk a "remarkable agent of change," Son made clear that SoftBank will focus on building "formidable" data center capacity on Earth1

1

. "He who strikes first wins," he declared, underscoring his belief that immediate action trumps long-term orbital ambitions3

3

. SoftBank has committed approximately $65 billion to OpenAI and is a major backer of the Stargate project, which aims to build AI infrastructure across the United States4

4

. In January 2025, Son visited the White House to announce SoftBank's initial commitment of $19 billion for the Stargate project, which plans to invest a total of $500 billion over four years3

3

. The company's telecom unit is also preparing forays into neocloud, AI-focused cloud infrastructure, and data center storage battery markets in the U.S., with the Japanese neocloud business scheduled for launch this year4

4

.

SpaceX's Orbital Ambitions Face Scrutiny

The timing of Son's comments is notable, coming just weeks after SpaceX completed a historic IPO that briefly made Musk the world's first trillionaire. However, SpaceX shares have already begun to fluctuate, briefly dipping below the opening-day price of $150 on Tuesday morning before rebounding1

1

. In its IPO filing, SpaceX estimated a $28.5 trillion total addressable market, with roughly $26.5 trillion expected to come from AI computing alone1

1

. Musk has said SpaceX is working toward launching 100 gigawatts of computing power into orbit annually with solar-powered AI satellites1

1

. Analysts at Morningstar have warned that SpaceX's IPO was overvalued, assigning just a 7% probability to the best-case scenario where the company successfully uses reusable rockets and orbital data centers to capture significant data center compute market share1

1

. Morningstar analyst Nicolas Owens noted that investors were paying a $72 premium on SpaceX's initial share price for the chance to participate in its orbital data center ambitions3

3

.

Technical Challenges of Cooling and Power Dissipation

Beyond the economics, significant technical hurdles remain for orbital data centers. Scientists have raised questions about the physics of cooling and power dissipation in the vacuum of space2

2

. The heat generated by dense clusters of chips must be dissipated somehow, and without air to carry it away, thermal management becomes a genuinely difficult problem2

2

. These technical challenges tend to be absent from promotional materials but represent fundamental obstacles to making space-based facilities viable. SpaceX itself has acknowledged these hurdles, with one filing noting that orbital data centers may not prove feasible2

2

.

Why the AI Race Timeline Matters

Son's argument hinges as much on timing as on economics. He believes the AI race will be decided in the next few years, long before orbital data centers could become operational2

2

. This makes them a distraction from the competition that actually counts. At SoftBank's annual shareholder meeting, Son set a target of raising the company's net asset value to 1 quadrillion yen ($6.4 trillion) over the next 16 years, fueled by its pursuit of artificial superintelligence, a system he claims could be 10,000 times smarter than a human3

3

. Son dismissed talk of an AI bubble as "blasphemy against AI" and said he has no plans to retire soon3

3

. He also noted that AI competition is intensifying but said there is room for "ten-fold, a hundred-fold" growth for OpenAI and rivals like Anthropic and Google4

4

.

The Energy Crisis Driving Space Ambitions

The debate over orbital versus terrestrial data centers reflects a real crisis in AI computing infrastructure. Global data centers consumed 448 terawatt-hours of electricity last year, more than Saudi Arabia's entire consumption, and U.N. researchers estimate that number will double by 2030 to meet growing AI demand3

3

. Data centers on Earth are colliding with limits on grid capacity, water, and land, which is why space has been proposed as a potential solution2

2

. Both SpaceX and Jeff Bezos' Blue Origin have announced plans to build and launch orbital data centers to overcome these energy and space constraints4

4

. Bezos has predicted that humans could start building massive data centers in space within the next two decades1

1

. However, Son's position is that the fix is more expensive than the problem, at least on the timescale that matters for winning the AI race2

2

.