News · 2026-07-19

Masayoshi Son Says the AI Economy Will Need $5 Trillion a Year by 2040

SoftBank founder Masayoshi Son used his SoftBank World keynote to put a number on his AI conviction: by 2040, he said, roughly 20 percent of global GDP will flow into AI, and building the infrastructure to support it will require about $5 trillion of investment every year. In the same breath he dismissed the idea that AI is a bubble as, in SoftBank's own transcript, an outrageous and foolish question. It is a maximalist scale thesis, delivered as the industry debates whether AI capital spending has raced ahead of AI profits.

Key facts

• The forecast: about 20 percent of 2040 global GDP moving into AI, which Son put at roughly 7,000 trillion yen in annual GDP, and roughly $5 trillion (about 800 trillion yen) per year in infrastructure investment.
• The scale picture: 100 trillion AI agents, 1 billion humanoids, about 3 terawatts of AI data-center power demand, and 'quetta'-scale compute.
• Who and when: Masayoshi Son, at the July 14 SoftBank World keynote, reported by SoftBank on July 16, 2026.
• Primary source: SoftBank's official keynote report.

Some context for the numbers. Son's case is a pure scale argument: if AI becomes the substrate of the economy — running 100 trillion software agents, powering a billion humanoid robots, drawing three terawatts of electricity, which is a power draw on the order of entire national grids — then the spending required to build it out looks proportionate rather than reckless. His rebuttal to bubble-watchers is that today's capex only looks excessive if you ignore the size of the value he expects AI to create by 2040. He compared the skeptics, per SoftBank's report, to people who dismissed airplanes or automobiles without ever having used them.

The interesting part is that SoftBank's own investor materials describe a more disciplined financing plan than the headline suggests. In its FY2025 investor Q&A, the company says AI-related enterprise businesses are expected to become future growth drivers, with full-scale expansion from FY2027 onward, and — importantly — that it will not assume all AI investment sits on its own balance sheet. It plans to lean on off-balance-sheet arrangements, partner collaboration, and external capital to keep financial discipline. SoftBank also describes a 'Neo Cloud' approach: building chips and systems into its own data centers and offering them as cloud services, rather than only leasing land and buildings to hyperscalers.

That matters because it reframes the $5 trillion figure. It is not a claim that SoftBank will spend it; it is a claim about the total ecosystem buildout Son expects, which he wants SoftBank positioned across. And the ecosystem really is spending heavily. TSMC's official Arizona release, for instance, describes a second fab under construction, about $40 billion of investment across two fabs, more than 10,000 construction workers, and an expected 10,000 additional high-tech jobs when complete. None of that proves Son's 2040 forecast, but it confirms that the semiconductor and infrastructure buildout he is pointing at is real, capital-intensive, and already underway.

The honest counter-case, which SoftBank frames as the view Son is rebutting, is straightforward: AI spending is too large and too early relative to realized profits, and a forecast reaching to 2040 is unfalsifiable in the near term. A 20-percent-of-GDP claim is exactly the kind of round, enormous number that is impossible to check and easy to headline. It should be read as a strategic bet and a positioning statement, not a verified projection.

Why it matters: SoftBank is one of the largest allocators of AI infrastructure capital in the world, so how Son frames the opportunity shapes real chip, power, and data-center investment. It also sharpens the central tension of this cycle — the tug between scaling laws that reward ever-larger buildouts and the sober question of when that spending turns into profit. The caveat is the whole point: everything past the confirmed keynote text is forecast, and the $5 trillion is a vision of an ecosystem, not a budget.

Primary source, verified: read the paper →

Ground Truth. (2026, July 19). Masayoshi Son Says the AI Economy Will Need $5 Trillion a Year by 2040. Ground Truth. https://groundtruth.day/news/masayoshi-son-5-trillion-a-year-ai-thesis.html

@misc{groundtruth:masayoshi-son-5-trillion-a-year-ai-thesis,
  title  = {Masayoshi Son Says the AI Economy Will Need $5 Trillion a Year by 2040},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {jul},
  url    = {https://groundtruth.day/news/masayoshi-son-5-trillion-a-year-ai-thesis.html}
}

Topics: industry · softbank · infrastructure · markets · compute · investment