Masayoshi Son's $5 Trillion 'Bet': Why the AI Bubble Argument is Absurd, But Warrants Caution

This story starts with a metaphor about "brain capacity."

Masayoshi Son of SoftBank, the man famous for "big bets" in tech investment, recently threw out another number that made everyone take notice. He said that by 2040, the AI industry will need $5 trillion in annual investment. How big is that number? Among countries with top global GDP rankings, only the US and China have annual economic totals exceeding it. He also bluntly evaluated: talk of an AI bubble is utterly absurd.

I understand his logic and partially agree with his judgment. But as a journalist who has covered the AI beat for 5 years, I want to say more: behind what he calls "absurd," there may lie a more dangerous signal.

Conclusion first, argument later.

The conclusion is: Masayoshi Son argued for AI's long-term golden age from the "demand side," but it is precisely this rhetoric of "inevitability" that can lead the market into blind optimism about the "short-term path." We should accept the long-term trend of AI as infrastructure, but must remain vigilant against the illusion that "throwing money wins."

His argument centers on an analogy between "primate brains" and "silicon brains."

In Masayoshi Son's speech, there's a vivid metaphor: the human brain (primate brain) has 10 billion neurons, equivalent to one parameter. Future AI large models will have "neuron" counts far exceeding humans. To achieve this level of intelligence, what is needed? Computing power, energy, and massive capital.

The data chain he provided is:

• Today's ChatGPT: Already shows "seeds of intelligence" surpassing certain human capabilities.
• Future AGI (Artificial General Intelligence): Requires models with 10 trillion, or even 100 trillion parameters.
• Infrastructure supporting all this: From chip manufacturing (TSMC's 3nm/2nm fabs) to power supply (nuclear-power-level data centers), to global fiber optic networks. Each item is astronomically expensive.

[!note] Masayoshi Son's "Input-Output" Deduction

Let's do a simple comparison to see if his $5 trillion/year figure is "outrageous":

- Global Semiconductor Market Size (2023): ~$500 billion

- Global Power Market Investment Scale (2023): ~$3 trillion

- Global Data Center Investment Scale (2023): ~$200 billion

Masayoshi Son's $5 trillion is almost equivalent to doubling or more than doubling all three major markets above over the next 15 years. He believes AI will create value far exceeding these costs, thereby attracting capital.

But the question is, is this "inevitability" over-packaged?

My view is that Masayoshi Son's discourse has an implicit premise: We must go to any lengths to achieve AGI. For an investment institution like SoftBank, this premise holds. Because SoftBank plays the role of "selling shovels"—whoever finds gold, the shovel seller profits. SoftBank owns Arm (chip IP) and Vision Fund (invested in OpenAI, Nvidia, etc.); he is one of the largest "infrastructure builders" in the AI ecosystem.

So, shouting out $5 trillion is essentially a manifesto of "creating demand." He needs the market to believe that future AI will require this much money, so his investments (like Arm's valuation) have support, and his next funding round (like Vision Fund II) has a story to tell.

Thus, we must beware of another form of "bubble."

The "bubble" Masayoshi Son criticizes refers to views that AI investment is overheated and valuations are too high. But in my view, the real bubble risk stems precisely from this "grand gesture, grand narrative" style of storytelling.

• First, it masks the current "inefficiency" and "immaturity" of AI applications. Many companies spend billions buying GPUs, but actual application scenarios (like customer service, code generation) don't have high ROI (Return on Investment). If everyone believes "throwing money always pays off," it leads to massive capital flooding into inefficient experiments, causing resource waste.
• Second, it ignores the slowdown of "Moore's Law." Masayoshi Son's $5 trillion is based on the assumption that "computing demand grows 80% annually." But in reality, chip processes are approaching physical limits, and costs rise sharply after 3nm. If the slope of the computing growth curve slows down over the next 5 years, the $5 trillion figure shifts from "demand" to "inventory."
• Third, it creates an anxiety of "must win." This narrative hijacks the entire industry. If all companies believe "we must invest $5 trillion to not fall behind," eventually only a few may survive, while the vast majority of startups die burning cash. This doesn't look like a "golden...

Original link: https://www.ithome.com/0/976/507.htm