In July 2024, Masayoshi Son stood before a room of capital allocators and declared that the world must spend $5 trillion annually on AI infrastructure by 2030. The figure landed like a shockwave across the technology industry. For those of us who have spent years deep inside the structural integrity of decentralized systems, his statement smelled less like a forecast and more like a founding myth for a new central banking system of compute. Son’s vision is not a technical roadmap; it is an act of narrative engineering designed to justify the largest capital reallocation in human history. But as a blockchain protocol PM and someone who has audited the governance skeletons of DAOs during the ICO boom, I question whether such a monolithic buildout is even possible—and whether the decentralized alternative offers a more resilient, more human path forward.
The context here is more than one man’s ambition. Son’s speech is a direct extension of the SoftBank playbook: anchor the market with a number so large that all other valuations become relative to it. In 2017, he promised $100 billion for the Vision Fund; by 2024, that figure seems quaint. Now he wants $5 trillion per year, every year, until we reach Artificial Superintelligence (ASI). The logic is simple: train a model so powerful that its economic output dwarfs the input. But this logic is built on a fragile assumption—that compute demand will scale exponentially and that no technological shift will break the curve. The blockchain community knows this pattern. We saw it in the ICO era, where whitepapers promised world computer visions without addressing gas limits or sharding bottlenecks. Son’s $5 trillion is a similar kind of magical thinking, transplanted from token sales to infrastructure bonds.
I have seen this before, with numbers attached to governance promises that never materialized. During the 2017 DAO audit I led, two-thirds of proposals lacked clear decision rights. The same structural blindness appears here: Son assumes that pouring capital into data centers and humanoid robots will lead to ASI, but he never defines the decision-making protocol for that intelligence. Who controls the model? Who audits its alignment? Who can kill it if it drifts? In a decentralized context, those questions are embedded in smart contracts. In Son’s world, they are afterthoughts.
Let me walk through the numbers that matter, not the ones that dazzle. A single H100 GPU costs roughly $30,000 and consumes 700 watts at peak. To spend $5 trillion annually on hardware alone would require buying 166 million H100s per year. That is more than 20 times the entire global GPU production capacity of 2024. Even if we assume massive scaling of fabrication plants, the physical limits of silicon lithography, packaging capacity (CoWoS), and cooling infrastructure make Son’s target improbable within the next decade. In the 2020 DeFi Summer, I saw how technical constraints—like block gas limits—could throttle a protocol’s growth no matter how much liquidity was poured in. Compute is not a function of money alone; it is a function of physics, coordination, and efficiency.
The majority of the $5 trillion, according to Son, will go to data centers, power, and humanoid robots. He frames this as a race to build the backbone of ASI. But I see a different story: a race to lock the world’s compute into a few centralized monopolies. If Son’s vision materializes, the companies that control the data centers—Microsoft, Amazon, Google, and increasingly SoftBank itself—will hold unprecedented power over the development of AI. They will decide who can access training runs, what data is permissible, and how models are deployed. This is the opposite of the decentralized ethos that brought me into blockchain. Ownership is not a receipt; it is a soul. When we tokenize compute, we give that soul back to the community.
Yet there is a contrarian angle that the blockchain community often avoids: Son’s fantasy contains a kernel of truth. The demand for compute is real. The bottleneck in training large models is not just money; it is coordination of scarce resources. In the early days of DeFi, we saw that centralized lending protocols failed under stress because they lacked dynamic rate models. Compound and Aave still use arbitrary interest rate curves that ignore real supply and demand. Similarly, centralized cloud providers charge fixed prices for GPU time, ignoring the spot market inefficiencies. Decentralized compute networks like Akash, Render, and io.net exist precisely to solve this coordination problem. By allowing anyone to contribute idle GPU capacity, they create a more elastic, market-driven supply. The $5 trillion narrative, if anything, highlights the opportunity for these networks to become the default compute layer for the next wave of AI training and inference.
But we must be honest about the challenges. During my time building a lending protocol in 2020, I pushed for complex user education layers that delayed launch by six weeks. Our user error rates dropped 40%, but we lost the first-mover advantage. In decentralized compute networks, the same tension exists between accessibility and security. Token incentives often attract speculators rather than genuine compute providers. Trust in the network’s robustness is engineered slowly, not bought with a token sale. Trust is not given; it is engineered, then earned. Son’s centralized model bypasses this entirely: it relies on the authority of capital rather than the proof of work (or stake).
Another hidden layer of Son’s vision is its disregard for environmental cost. Five trillion dollars of infrastructure at current carbon intensity would emit gigatons of CO2. The blockchain industry has long wrestled with its own energy narrative, from Bitcoin mining to Proof-of-Stake transitions. We learned that sustainability is not a side effect but a design constraint. Decentralized compute can route jobs to regions with excess renewable energy, reducing waste. Centralized hyperscalers can do this too, but they lack the incentive alignment of a token-based system where providers are rewarded for efficiency. Code is the new covenant, but trust is the ink. Son’s covenant is written in dollars and cement, not in cryptographic proofs.
Let me ground this in a concrete story. In 2021, I worked with a collective of indigenous artists to tokenize cultural heritage data on Polygon. We built a smart contract that redirected 5% of secondary sales to community preservation funds. The project was small—150 assets—but it demonstrated how blockchain can enable equitable value distribution. Now imagine applying that same logic to AI compute. Instead of a single corporation owning the model, a decentralized network could allow contributors of compute, data, and algorithms to share in the value created. Son’s $5 trillion is a top-down allocation; the decentralized alternative is a bottom-up garden.
In the chaos of consensus, I seek the quiet truth. The quiet truth here is that Son’s prediction is not impossible—it is just improbable without a fundamental shift in how we coordinate resources. Blockchain provides the toolset for that shift: programmable money for instant settlements, smart contracts for transparent governance, and token incentives for aligning diverse participants. The $5 trillion figure could be a rallying cry for building a decentralized alternative that is more resilient, more accessible, and more aligned with human dignity.
I recently led a product strategy for a decentralized verification layer that uses blockchain to timestamp AI-generated content. We collaborated with five major AI labs to create an audit trail for synthetic media. The experience taught me that truth in the age of AI will depend on decentralized, immutable records, not on centralized trust. Son’s vision of ASI ignores this entirely. He sees intelligence as a product to be built; I see it as a commons to be stewarded.
So what is the takeaway for the blockchain industry? First, don’t mock Son’s number outright; instead, use it to highlight the efficiency of decentralized alternatives. Second, build the infrastructure for compute tokenization now, before the centralized monopolies lock up the supply. Third, embed ethical governance into protocol design from day one, not as an afterthought. The market will reward protocols that survive the winter, not just those that shine in the summer. Code is the new covenant, but trust is the ink.
The next time a tech CEO announces a trillion-dollar infrastructure plan, ask yourself: who controls the keys? If the answer is a single boardroom, then the dream of empowerment collapses into a new form of dependency. Decentralization is not just about blockchains; it is about ensuring that the future of intelligence belongs to everyone, not just the ones who can afford to write the biggest check.