Silence is the first vote in a true consensus.
I have spent twenty-four years watching power crystallize around infrastructure. In 2017, while auditing The DAO’s transaction logs in Tallinn, I saw how a single vulnerability in a smart contract could drain millions—not because the code was broken, but because the governance layer was absent. Today, I find myself staring at a different kind of vulnerability: a statement from the U.S. Treasury Secretary, Scott Bessent, declaring that America will control 80% of the world’s AI compute. As a DAO governance architect who has designed quadratic voting for MakerDAO and drafted ethical frameworks for L2 rollups, I see this as a crystal-clear signal that the same centralization disease that plagued early smart contracts is now metastasizing into the physical fabric of computing itself.
The statement was brief, almost surgical: “The United States will maintain dominance over 80% of global compute to ensure AI advantage over China.” No technical data, no definition of “control,” no mention of the energy or supply chain costs. But for those of us who read blockchain governance signals, the subtext is loud. This is not a prediction; it is a strategic commitment to monopolize the raw material of the digital age. It echoes the same tone I heard in 2020 when a whale-dominated DAO tried to centralize voting power under the guise of “efficiency.” Back then, I facilitated twelve town halls to introduce quadratic voting. Today, I wonder: what kind of governance framework can prevent a nation-state from becoming the ultimate whale of compute?
Let me ground this in personal experience. In 2024, after the Spot Bitcoin ETF approval, I was invited to a closed-door panel in Geneva with institutional investors. I presented a deck titled “Beyond Speculation: Blockchain as a Trust Layer,” arguing that capital inflows must adhere to decentralized standards. One asset manager asked me bluntly: “If the U.S. government controls the majority of Bitcoin mining hash power, does that make Bitcoin centralized?” I answered yes, and the room fell silent. That silence is now louder than ever. Bessent’s claim is essentially the same question applied to AI compute—but at a scale that dwarfs any blockchain. We are not talking about 51% of a network; we are talking about 80% of all neural network training capacity.
The core insight here is not about AI; it is about how we define and audit “control.” In my work auditing the reentrancy vulnerabilities of The DAO, I discovered that the exploit was possible because the smart contract treated ETH transfers as atomic actions, ignoring the recursive callbacks. The code assumed a single-threaded reality. Similarly, Bessent’s statement assumes a single-pole geopolitical reality where the U.S. can seamlessly own the supply chain from chip fab to datacenter to model deployment. But as any blockchain engineer knows, assumptions about atomicity break when you introduce sharding, rollups, or heterogeneous validators. The physical reality of global compute will push back: energy grids are not unified, chip fabs are in Taiwan, talent is global, and open-source algorithms can run on any hardware. The U.S. may control 80% of the headline compute, but the remaining 20%—if deployed in smart, resilient networks—could become the Ethereum of AI: smaller but more adaptable, permissionless, and antifragile.
But here is the contrarian angle that keeps me awake at night. Maybe the pursuit of 80% control is not a bug; it is a feature of how decentralized systems evolve. When I retreat to Hiiumaa island during the bear market of 2022, I wrote “The Hollow Promise of Yield” and realized that much of DeFi’s innovation was just financial engineering. Similarly, the blockchain community’s obsession with “decentralization as an absolute” may blind us to the reality that some level of centralization is inevitable for efficiency—especially in compute-intensive tasks like AI training. The real question is not how to prevent centralization, but how to govern it. What if the U.S. built a publicly auditable, permissionless compute infrastructure—a kind of national cloud that anyone could use, with ZK-proofs ensuring privacy and verifiability? That would be a radically different form of control: stewardship rather than ownership.
From my work on MakerDAO’s governance redesign, I learned that token-weighted voting is mathematically fair but psychologically excluding. Small holders need to feel heard, not just counted. Similarly, if 80% of AI compute is controlled by one state, the other 20% must be designed for inclusive participation, not just residual scraps. During the pilot of my decentralized identity protocol for AI agents in Tallinn, I saw how ZK-rollups could allow autonomous agents to prove their origin without revealing proprietary data. That same technology could let non-U.S. entities prove they are not using banned hardware, thus gaining access to a global compute pool without sacrificing privacy. The infrastructure exists; the governance does not.
Let me be clear about the risks. In my MakerDAO town halls, I watched whales propose governance attacks dressed as “liquidity incentives.” Similarly, Bessent’s declaration risks sparking a compute arms race where capital is poured into redundant datacenters, inflated energy contracts, and locked-in hardware that becomes obsolete in three years. The real cost is not the $100 billion spend; it is the opportunity cost of not building open, portable compute standards. During the 2024 Geneva panel, I negotiated with three asset managers to adopt a “Green-DAO” reporting standard for their crypto holdings. One of them admitted that the biggest risk in AI infrastructure is not technical but political: a policy flip after the next election could leave investors holding billions in stranded assets. That is the same risk every DAO faces when a single whale controls the treasury.
My experience auditing the reentrancy vulnerability taught me that code is not law—governance is. The The DAO hack did not happen because of a bug; it happened because there was no mechanism to pause or patch the contract without social consensus. Today, Bessent’s declaration may trigger a wave of U.S. datacenter construction, but without a governance layer—a framework for distributing compute access, pricing it fairly, and auditing its use—that infrastructure will reproduce the same power imbalances that blockchain was supposed to solve. I call it the “iron law of infrastructure”: every concentrated resource eventually produces concentrated power, unless you design for explicit checks and balances from day one.
So where does that leave us? I propose a thought experiment inspired by my work on the MakerDAO quadratic voting model. Imagine a global compute commons: a network of datacenters, each operated by a different entity (government, cooperative, DAO), all running the same open-source software stack, with a tokenized credit system for allocating training time. The U.S. could hold 80% of the physical compute, but if that compute is governed by a democratic, transparent protocol—with safeguards against majority capture—then the “control” becomes stewardship, not tyranny. This is not utopian; it is the exact same design pattern that separates a DAO from a dictatorship. The technical components exist: ZK-proofs for identity, rollups for scaling, and on-chain voting for governance. What is missing is the political will to build it.
During my six weeks on Hiiumaa, I wrote that “winter teaches what spring forgets.” The crypto winter of 2022 taught me that hype masks fragility. Now, the AI summer of 2024 masks a deeper fragility: the concentration of compute in a few hands. If we as a blockchain community cannot apply our own governance principles to this new infrastructure, we will have failed our core mission. I remember the words of a small token holder in one of my MakerDAO town halls: “Don’t just give me a vote; give me a voice.” That is the difference between control and governance. Bessent’s 80% claim is a rallying cry for control. Our response must be a blueprint for voice.
I am not naive. I know that nation-states do not easily cede control over strategic assets. But I also know that every successful decentralized system I have helped design—from MakerDAO’s quadratic voting to the AI agent identity protocol in Tallinn—was built by starting with the ethical framework first, then layering technology on top. The silence before a consensus is where the real work happens. Let us listen to that silence before we build the next generation of compute. Otherwise, we will have engineered a server farm that centralizes what the internet was supposed to liberate.
The takeaway is not a policy recommendation; it is a call to ask deeper questions. Who audits the auditor of compute? How do we measure the decentralization of a global resource that is inherently physical? And most importantly, can we design a governance system for 80% compute control that still leaves room for the other 20% to innovate, dissent, and grow? I do not have the answers. But I know that the first vote in any true consensus is silence—the willingness to stop talking and start listening to the risks, the hopes, and the quiet engineering work that will determine whether this technology liberates or enslaves.
