SoftBank's Son rejects the AI bubble thesis and floats a long horizon — but the capex math still has to close

Masayoshi Son is not in a mood for doubt. Speaking to reporters on 24 June 2026, the SoftBank Group founder dismissed the idea that the current AI investment cycle is a bubble as "blasphemy" and said he intends to keep running the Japanese conglomerate "into his 70s" to see the bet through, according to Nikkei Asia. The comments are a familiar Son register — maximalist, theological in tone, and deliberately offensive to the consensus of the moment. They are also a clear signal that the group's spending posture, already the most aggressive among the large diversified tech-holding companies, is not about to soften.

The structural read is straightforward. The world's major technology investors are, in aggregate, committing hundreds of billions of dollars to AI infrastructure — chips, model labs, data-center build-outs, and the power contracts that now travel with them. SoftBank is one of the loudest and earliest actors in that race, and Son's job on a Tuesday morning is to keep the conviction narrative intact at a moment when several credible voices are beginning to ask whether the revenue curve can catch up with the capex curve. His answer is that it will, and that anyone who disagrees is, in effect, committing a category error against history.

The substance behind the rhetoric

Son's argument is the same one he has been making since the 2017 launch of the SoftBank Vision Fund: that the world is entering a step-change in compute demand, and that the entities that finance the supply of that compute will own the platform layer of the next economic cycle. The empirical question is whether compute demand is growing on the trajectory he implies, or whether it is being front-run by suppliers who need a narrative to justify their own order books. The Nikkei Asia report does not break out any new revenue figures; it frames the interview as a posture statement, not a financial update. The harder numbers — operating cash flow at SoftBank's portfolio companies, the cash drain from the Vision Fund's remaining holdings, the dollar cost of the group's outstanding AI commitments — are not in the report and would have to be read off the group's most recent earnings disclosures to be evaluated properly.

That absence is itself the story. Son is choosing to speak when he has a thesis to defend, not when the books close. The next test of whether the bet is working will be the next set of quarterly results, and the cycle between those reports is precisely the window in which a founder of Son's communicative style can move the mood music. The "blasphemy" line is built to do that work — to put the burden of proof on the sceptics and to frame caution as a kind of cowardice.

The space-data-center rejoinder

A second, less-noticed comment from the same day cuts in the other direction. The same Son, addressing the much-discussed question of orbital data centers, said there is "little merit" to building them, per a post on X by Unusual Whales on 23 June 2026, summarising Son's reported position. The remark is significant because it cuts against one of the more viral framings of the AI build-out — the idea that the compute shortage on Earth will be solved by lifting racks into low orbit, where solar power is continuous and cooling is free. Son is, in effect, telling that story to calm down. The structural implication is that the AI capex cycle is, in his view, an Earth-bound infrastructure problem with a known engineering and power-grid solution. The space-data-center pitch, attractive as it is to venture-backed space startups and to the financial press, is not, in his telling, where the next several hundred billion dollars of capital expenditure will be deployed.

This is worth holding alongside the "blasphemy" line. The same man who says sceptics of the AI build-out are wrong is also telling investors that the most futurist version of the build-out — the one that tends to attract speculative capital — is not the one he is backing. That is a more disciplined position than it first appears, and it is a small piece of evidence that the SoftBank thesis is not a pure momentum trade.

What the capex math actually has to show

A serious evaluation of the Son position requires a few reference points that the Nikkei Asia and Unusual Whales items do not supply. The aggregate capex commitments of the major US hyperscalers and chipmakers over the next several years run into the hundreds of billions of dollars annually; the comparable figure for SoftBank's own commitments is harder to pin down because much of it is routed through the Vision Fund, through Arm's licensing and royalty stream, and through co-investment vehicles with Middle Eastern capital. The question of whether AI demand is on a trajectory to absorb that capital at acceptable returns is, at the time of writing, genuinely open. The bear case is that the demand curve is real but more concentrated than the capex curve implies — that the marginal dollar of compute is going to a smaller and smaller set of model developers and a smaller and smaller set of inference workloads, with the result that the supply side gets paid in bursts, not in a smooth annuity. The bull case is the one Son is selling: that demand is broad-based, that the current model-generation step will be followed by another, and that the platforms that own compute will be the platforms that own the next decade of the software industry.

A third, less comfortable possibility is the one the sources do not address directly: that the AI capex cycle is real, and broadly well-directed, but that the rate of return on the earliest cohort of bets will be poor — that the winners of the cycle will be the second and third movers, not the first, and that SoftBank's role as a first mover is therefore a strategic risk rather than a strategic certainty. Son's rhetoric is built to push that reading out of the frame. The next round of portfolio marks at the Vision Fund is what will actually move it.

Stakes and what to watch

If Son is right, the consequences extend well beyond SoftBank. The Japanese conglomerate is, in practice, a pricing signal for the rest of the AI-capital complex — its willingness to commit to chip-of-the-cycle orders and to anchor data-center joint ventures is one of the inputs that other capital allocators use to size their own exposures. A clear softening of that posture would be felt in chip suppliers, in power-grid planning, and in the secondary market for AI-linked private credit. A hardening of the posture — which is what the 24 June comments amount to — does the opposite, and is one of the reasons the AI capex narrative has retained its gravity even as public-market multiples have started to compress.

The two things worth watching in the near term are concrete. First, the size and structure of the next major chip-order cycle at any of the leading AI accelerator suppliers; a large SoftBank-anchored order would be a confirming data point for the Son thesis, while a quiet one would not be. Second, the disposition of the data-center power question, which is increasingly the binding constraint on AI capex in the United States, Europe, and parts of East Asia. Son's dismissal of the orbital data center pitch implies that he expects the power question to be solved, or at least managed, on the ground. If that expectation turns out to be wrong, the "blasphemy" framing will not be the part of the interview that ages well.

What remains uncertain is the share of SoftBank's AI exposure that is actually incremental to the broader market capex cycle, as opposed to co-invested with it. The sources do not break that out. They also do not specify the size of the next round of SoftBank commitments or the timeline over which they are expected to be deployed. The interview is, in other words, a posture document — a useful one, given how much of the AI narrative is currently driven by posture — but not a balance-sheet event. The balance-sheet event is what the next earnings report will be, and the gap between Son's vocabulary and that report's numbers is the part of the story this publication will be watching most closely.

Desk note: Monexus framed the Son comments as a posture event with two separable claims — a bullish capex stance and a sceptical view of orbital data centers — rather than as a pure bubble-versus-non-bubble argument, which is the line most wire coverage has taken. The interesting question is not whether Son is enthusiastic about AI; the wire already shows that. The interesting question is what he is not enthusiastic about, and what that tells us about the shape of the next capex cycle.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://t.me/NikkeiAsia
• https://t.me/nikkeiasia
• https://en.wikipedia.org/wiki/SoftBank_Group
• https://en.wikipedia.org/wiki/Masayoshi_Son
• https://en.wikipedia.org/wiki/Artificial_intelligence_bubble