Hook
Masayoshi Son stood on a stage in Tokyo last week, his voice rising above the hum of servers that weren't yet built. He spoke of a world where artificial superintelligence would demand an annual investment of five trillion dollars into data centers, power grids, and humanoid robots. The number was precise — $5,000,000,000,000 every year until 2040. The audience nodded, some scribbled notes. I watched the livestream from my desk in Stockholm, my coffee growing cold, because I had heard this music before. In 2017, during the ICO mania, similar figures were thrown around for projects that promised to decentralize everything from file storage to human consciousness. The cadence was the same: a grand narrative, a clean number, a future so vast it justified any present sacrifice. The ghost in the machine that Son was tracing was not artificial intelligence; it was the oldest algorithm known to finance — the narrative of infinite growth.
Context
Son is the founder of SoftBank, a Japanese conglomerate that has positioned itself as the world's most aggressive technology investor through its Vision Funds. His most prized asset is Arm Holdings, the chip architecture company that powers nearly every smartphone and is now pivoting into data center AI accelerators. When Son talks about infrastructure, he is talking about the value of his own portfolio. But the context goes deeper. The crypto market I operate in has been bleeding for months — total value locked in DeFi has dropped 40% over the past quarter, and Layer2 solutions are fighting over crumbs of liquidity. In this bear market, survival matters more than gains. Son's prediction is not a market forecast; it is a fundraising pitch wrapped in a futurist's outfit. And it mirrors the very same narrative mechanics that built the crypto bubble of 2021: a story so compelling that investors forget to ask where the money comes from, or what happens if the story breaks.
Core
Let me dissect the narrative mechanism Son deployed, using the same tools I applied to audit the Ethos smart contract in 2017. That was my first lesson in trusting code over words. Son's prediction is a three-layer structure: a technical premise, a commercial promise, and a financial imperative. Each layer contains a fracture.
First, the technical premise. Son assumes that AI will evolve from today's large language models to artificial general intelligence and then to superintelligence within the next fifteen years. This is not just optimistic; it is unsupported by any peer-reviewed research. The current scaling laws — which state that model performance improves with exponentially more compute and data — are already showing diminishing returns. We are running out of high-quality training data, and the energy cost of training a single frontier model like GPT-5 is estimated to be over 100 gigawatt-hours. To reach superintelligence, we would need not just more of the same, but a fundamentally new architecture. There is no evidence such an architecture exists. Meanwhile, humanoid robots, which Son bundles as the physical endpoint, are still struggling to walk reliably on uneven terrain. The Boston Dynamics robots you see in viral videos are teleoperated for most tasks. The gap between a demo and a factory-floor workforce is measured in decades, not years.
Second, the commercial promise. Son claims that the revenue generated by superintelligence will justify the five trillion dollars of annual expenditure. But let us look at the numbers we have today. The entire global AI industry — including cloud AI services, software, and hardware — generated roughly $200 billion in revenue in 2025. To reach five trillion dollars in annual revenue by 2040, the industry would need to grow at a compound annual rate of over 30% for fifteen years, and maintain that growth while bearing the cost of the infrastructure. That would make AI the largest economic sector on Earth, surpassing energy, finance, and healthcare combined. There is no precedent for such a growth trajectory in any industry, not even in the early days of the internet. The internet's infrastructure investment peaked at around $500 billion annually during the dot-com bubble, and that was only after a decade of buildout. Son is asking us to believe in a tenfold increase on that peak, before any mass adoption of superintelligence has even occurred.
Third, the financial imperative. This is where the narrative becomes a weapon. Son is effectively setting a floor for AI infrastructure spending. By declaring that five trillion is the necessary amount, he forces every other player — from Microsoft and Google to sovereign wealth funds — to either match or lose the race. This is the classic prisoner's dilemma of capital allocation. If everyone invests, the narrative becomes self-fulfilling to some degree, even if the investment is wasteful. But if no one invests, the narrative collapses. Son is betting that the fear of missing out will outweigh the fear of overpaying. That is a bet he can win in the short term, but it creates a fragile equilibrium. I have seen this pattern before in DeFi's liquidity mining wars: protocols printed tokens to attract capital, and the capital flowed to the highest yield, but when the yield dried up, the liquidity vanished instantly. The same will happen here if Son's narrative stops being convincing.
To understand the depth of this narrative, I applied the same seven-dimensional framework I use to analyze blockchain protocols. Let me walk through the critical dimensions that Son's prediction fails to address, and where the cracks begin to show.
On the technology dimension, Son's path is a straight line from today's compute to superintelligence. But the blockchain industry has taught me that technological progress is rarely linear. In 2020, everyone was convinced that Ethereum's proof-of-work was the only secure consensus mechanism. Then The Merge happened, and the entire ecosystem shifted to proof-of-stake almost overnight. The technology that seemed fixed was upended. Son's assumption that we will scale today's GPU clusters to 100x larger sizes ignores the possibility that a new paradigm — such as neuromorphic computing, optical chips, or even brain-computer interfaces — could make current infrastructure obsolete. I have been tracking the rise of liquid networks for AI inference, and the first results suggest that smaller, specialized models running on edge devices can outperform massive cloud models for many tasks. If that trend continues, the demand for Son's hyperscale data centers may peak before they are even built.
On the commercialization dimension, the missing unit economics are deafening. Son never tells us who will pay for the superintelligence. Will it be consumers, companies, or governments? What price will they pay per query, per action, per robot hour? Without a pricing model, the revenue projection is a fantasy. In crypto, we learned that protocols with no sustainable revenue die. The same applies here. The only plausible large-scale payer is the advertising industry, but its global revenue is only about $600 billion per year. That cannot support a five-trillion-dollar compute bill.
On the investment narrative dimension, this is where Son's prediction functions as a classic pump mechanism. SoftBank's core asset, Arm, trades at over 100 times earnings, a valuation that can only be justified by a future where every device runs Arm-based AI accelerators. By declaring that the AI infrastructure market will be five trillion dollars annually, Son is essentially saying that Arm's addressable market is limitless. This is the same tactic used by crypto projects that announced partnerships with no real integration: the narrative itself becomes the product. But narratives have half-lives. The market will eventually demand capital expenditure receipts, not just keynote slides.
Contrarian
Now, let me offer the contrarian angle that most analysts miss. The five-trillion-dollar narrative, despite its absurdity, will have real consequences. It will drive capital into energy infrastructure, chip fabrication, and robotics, regardless of whether the superintelligence arrives. That capital is not fully wasted; it builds physical assets that can be repurposed. The data centers built today will be used for cloud computing, video streaming, and scientific research even if AGI never materializes. The nuclear reactors and solar farms built to power them will serve the grid for decades. The humanoid robots may be inefficient, but their development will advance automation in manufacturing and logistics. So even if Son's prediction is wrong by a factor of ten, the world will still end up with more compute, more energy, and more robots than it would have otherwise. The contrarian truth is that the narrative is dangerous not because it will lead to a bubble, but because it will accelerate a structural shift that we are not prepared for socially or geopolitically.
The blind spot here is environmental and social sustainability. The five-trillion-dollar annual infrastructure build would consume an estimated 10% of global electricity by 2040, even with efficiency gains. That electricity has a carbon cost unless it is fully renewable. But renewable energy infrastructure is also capital-intensive, and Son's plan does not allocate even a fraction of the five trillion to carbon capture or grid modernization. The result could be a massive increase in carbon emissions precisely when the world needs to decarbonize. Moreover, the labor displacement from humanoid robots is not factored into any social safety net. The narrative assumes that the wealth generated by superintelligence will trickle down, but history shows that technological windfalls are captured by capital, not labor. If Son's vision comes true, we will face a world of extreme inequality where the owners of the AI infrastructure — SoftBank, hyperscalers, and sovereign funds — control the means of production while billions of people are rendered economically obsolete. No amount of universal basic income can fix that if the infrastructure itself is owned by a few.
Another contrarian insight comes from the blockchain community: the promise of decentralized compute networks as a counterbalance. Projects like Render Network, Akash, and Bittensor propose to distribute AI computation across a peer-to-peer network, reducing reliance on centralized hyperscale data centers. In the shadow of Son's vision, these networks seem tiny — their total compute capacity is a fraction of a single Microsoft data center. But their value lies not in scale, but in resilience. A decentralized compute grid that is owned by thousands of independent operators cannot be shut down by a single government or corporation. If the future of AI is as powerful as Son claims, then the centralization of that power in the hands of a few entities is a massive risk. The contrarian bet is that the backlash against centralized AI will drive adoption of decentralized alternatives, even if they are less efficient. The narrative of "AI sovereignty" could become the next crypto bull market narrative, just as DeFi emerged after the 2018 ICO collapse.
Takeaway
I am listening to the silence between the blocks — the data points that Son's narrative conveniently ignores. The silence tells me that the five-trillion-dollar prediction is not a forecast but a signal. It signals that the largest institutional investors are preparing to pour capital into AI infrastructure at a scale that will reshape the global economy. For blockchain investors, the takeaway is not to chase the AI narrative in crypto directly, but to identify the points of friction. Where will the centralization of AI infrastructure create demand for decentralization? Which blockchain projects can provide verifiable computation, audit trails for AI decisions, or decentralized ownership of training data? The code is law, but trust is fragile — and a world where one man's keynote can move trillions is a world that desperately needs trustless systems.
The ghost in the machine is not superintelligence. It is the collective belief that infinite investment can overcome finite physics. We have seen this ghost before in crypto, in the tulip mania, in the South Sea Bubble. Each time, the narrative broke, but it left behind infrastructure that enabled the next wave. The five-trillion-dollar narrative will likely break too, but the data centers, the nuclear reactors, and the robots will remain. The question is who will own them, who will audit them, and how we will govern a world where the most powerful intelligence is locked inside privately owned servers. That is the question that no keynote can answer. The answer will be written in ledger light.