SoftBank Group founder Masayoshi Son told an audience at a Tokyo conference on Monday that global spending on artificial intelligence infrastructure will need to reach $5 trillion a year by 2040, a prediction that brushed aside persistent warnings the sector is in a speculative bubble. Son argued the scale of compute required for artificial general intelligence (AGI) leaves no alternative, framing the current investment surge as a necessary ramp-up rather than a repeat of the dot-com excess.
His remarks, delivered at SoftBank World 2026, doubled down on the Japanese conglomerate’s conviction that AI represents the biggest technological shift in history. The $5 trillion annual figure far exceeds current levels as the estimated total worldwide IT spending is roughly $5.4 trillion in 2025. Son’s projection implies that only AI-related outlays alone will eclipse the entire tech budget of the planet within 14 years. He dismissed bubble talk directly, telling reporters that comparing today’s AI buildout to the late-1990s internet frenzy misunderstands the physical depth of the investment required.
Whether Son is correct is the subject of sharp debate. The case that a bubble is forming rests on a basic mismatch: hyperscale cloud providers and chipmakers are pouring hundreds of billions into new data centers and specialized processors, yet the revenue from AI services remains nascent. A June 28, 2026 BIS Annual Economic Report analysis found that returns on AI infrastructure investment are lagging behind the pace of spending. Bank of America Global Research calculated that global AI capex hit $480 billion in 2025 and could top $1.7 trillion by 2030, a trajectory some analysts warn overshoots realistic demand if enterprise adoption stalls. The International Monetary Fund flagged in a April 2026 Global Financial Stability Report that concentrated AI bets by a handful of large firms could amplify a correction akin to the telecom collapse of the early 2000s.
On the other side, proponents insist the sheer utility of AI will absorb the capacity. They argue that generative AI is a general-purpose technology, comparable to electricity or the internet itself, which demanded enormous upfront capital before productivity gains materialized. If AI diffuses across healthcare, manufacturing, education and government, $5 trillion a year might prove conservative. SoftBank itself is betting heavily through its Arm subsidiary and a network of portfolio companies, signaling that Son’s view is not merely rhetorical. A Goldman Sachs report from July 2026 noted that AI-driven efficiency improvements could add $4.4 trillion annually to global corporate profits over a decade, providing the cash flow to justify the investment. In this view, the real risk is underinvestment, not overcapacity.
For now, the truth likely lies between the poles. The magnitude of capital being committed has no modern precedent outside of wartime mobilization, which means both the upside and the downside scenarios are extreme. Son’s $5 trillion call will be tested by whether AI applications mature fast enough to fill the data centers now under construction.