Someone who has been wrong before and still won gives them a certain kind of confidence. That’s how Masayoshi Son looks. The 68-year-old CEO of SoftBank famously lost more paper wealth than almost any single investor during the dot-com crash. This week, he spoke to shareholders and set a goal that most people would have thought was crazy: to increase SoftBank‘s net asset value to 1 quadrillion yen, or $6.4 trillion, in sixteen years.
When you think about how SoftBank recently passed Toyota to become Japan’s most valuable company, the number looks different. This isn’t a man speaking from the side. He has moved real money and taken real risks before, and now he’s doing it again, but this time the engine is artificial superintelligence.
Son’s plan has more than one plan going at the same time. Early in 2025, he promised $19 billion to the Stargate project with OpenAI. This was part of a plan to invest $500 billion over four years in U.S.-based AI infrastructure. Arm Holdings is mostly owned by SoftBank, and its chip designs are quietly used in a lot of processors around the world. Most recently, SoftBank paid $5.4 billion to buy ABB Robotics and closed a $4 billion deal with DigitalBridge. It’s easy to see the pattern: build the physical foundation before the demand really comes in.
Son also went off on a tangent at the meeting of shareholders for SoftBank’s wireless division, and it was aimed right at Elon Musk. Musk and SpaceX have been pushing orbital data centers as a long-term way to lower the energy costs of AI computing. The idea is that cheap solar power in space could help pay for the high electricity bills that slow down operations on Earth. Son carefully took apart the argument. He said that electricity costs are only a small part of the total cost of running a data center. More money is spent on semiconductors and hardware. Putting things into orbit, keeping them in good shape, and dealing with the delay in communication between Earth and space are also costs. Son agreed that Musk was a huge deal in the business world, but he was very clear about when things would happen. He said that what an orbital data center might have to offer in ten years will not matter nearly as much as what they do now.
Keep in mind that the pasts of these two men are not simple. In 2017, they got together to talk about SoftBank’s possible investment in Tesla. The talks broke down because of disagreements about who owned what. When Stargate was announced in January 2025, Musk publicly asked SoftBank if it had actually gotten anywhere near the money it had promised. What SoftBank did in response wasn’t words; it bought other companies. The competition is real, but it’s not being fought with press releases.
The issues of energy at stake in this debate are really important. In 2024, data centers around the world used about 415 terawatt-hours of electricity, which is about 1.5% of all the electricity used in the world. That number could more than double by 2030, mostly because AI will be doing more work. Son says that infrastructure on Earth can meet this demand in months, but orbital systems won’t be commercially viable on a large enough scale for a while yet. A Morningstar analyst recently said that SpaceX’s IPO, which raised about $75 billion in June and put the company’s value at almost $1.77 trillion, included a premium of about $72 per share for risky future bets like orbital computing. The same analyst said there was a 43% chance that those orbital data centers would never be able to make money.
Son’s frame is easier to understand. “He who strikes first wins,” he told the shareholders, which could be seen as both a plan and a warning. Even in the best case scenario, it’s still not clear if the goal of a $6.4 trillion net asset can be reached. But seeing Son make a promise of this size and detail makes it hard not to take the goal seriously. He’s not telling us what will happen. He wants to buy enough of it to make a difference.