Masayoshi Son says nuclear fusion will power AI within 15 years

Signals Inbox·July 14, 2026·Nuclear

Masayoshi Son says fusion could replace gas as the main power source for AI data centers within 15 years. The bold part is not just the timeline. He is pairing it with a forecast that AI data centers could need 3 terawatts by 2040, roughly 1.8 times today’s total global electricity use.

Q1What did Masayoshi Son actually say?

At SoftBank World 2026, Son said gas would power the first wave of AI infrastructure, but fusion could become the main source within about 15 years. He also forecast 3 terawatts of data center power demand by 2040. This is a prediction from SoftBank’s founder, not a new power contract or a working reactor.

Q2Why does the 3-terawatt number matter?

Because it is enormous. Son’s estimate is roughly 1.8 times current global electricity consumption. The IEA’s nearer-term forecast is much smaller but still huge: data center electricity use could rise from about 415 terawatt-hours in 2024 to around 945 terawatt-hours by 2030. Son is basically arguing that today’s grid buildout is only the opening act.

Q3Why fusion instead of solar, wind, or normal nuclear?

AI data centers want power all day, every day, in a small area. Solar and wind are cheap but variable, so they need storage, transmission, and backup. Fission is steady but slow to permit and build. Fusion promises compact, clean, always-on power with less long-lived waste. The catch is simple: nobody has yet sold commercial fusion electricity to the grid.

Q4Is Son’s 15-year timeline realistic?

Possible, but very aggressive. Helion says it will supply Microsoft with at least 50 megawatts from a fusion plant in 2028. Google has agreed to buy 200 megawatts from Commonwealth Fusion Systems in the 2030s. Those deals show real customer demand, but they are still promises. The hard part is moving from one pilot plant to thousands of reliable power units.

Q5What changed in the last year?

Fusion money accelerated. Industry investment rose sharply, while supply-chain spending increased 24% in 2025. Big tech also moved from vague interest to actual purchase agreements. That matters because AI companies are no longer treating power as a utility bill. They are signing contracts years ahead and helping finance entirely new energy technologies.

Q6What is the real tension here?

AI needs power now, while fusion may need another decade or more. The IEA expects gas and coal to supply over 40% of new data center electricity demand through 2030. So the uncomfortable bridge is fossil fuel first, fusion later. Son’s clean-energy future could therefore require a very dirty buildout before the technology arrives.

Q7So should we take this seriously?

Take the bottleneck seriously, not the deadline. Son has a reason to talk big because SoftBank is betting heavily on AI infrastructure. But the core point is solid: chips are no longer the only limit. Power generation, grid connections, turbines, transformers, and permits are becoming strategic assets. Fusion wins only if it becomes cheaper and faster than all those alternatives.