In a recent Fox News interview, President Trump urged state and local governments to welcome AI data centers, framing them as "large factories" that generate jobs, tax revenue, and capital inflow. He acknowledged that "most Americans oppose having a data center in their community," yet the narrative remains one of boosterism—jobs, taxes, growth.
We audit the code, but who audits the conscience of the infrastructure that powers AI?
This is not a question of if we should build, but of how we build. The AI data center boom is not just a tech story; it is a local governance story, a story of land, power, water, and community consent. And for those of us who have spent years inside the blockchain ecosystem, watching the centralization of compute power, the parallels are uncomfortable.
Context: The Infrastructure Race
AI data centers are no longer just IT facilities. They are industrial-scale power consumers, often requiring 100 megawatts or more—equivalent to a small factory town. The shift from "server room" to "AI factory" means that the constraints are no longer just bandwidth and cooling, but grid capacity, substation upgrades, long-term power purchase agreements, and the willingness of local communities to host a 24/7, high-noise, high-water-use facility in their backyard.
Trump's call for states to compete for these projects signals a new phase: AI infrastructure is becoming a political asset. Governors and mayors see tax revenue and construction jobs; tech giants see scarce land and cheap power; residents see noise, traffic, and environmental risk. The tension is real.
Based on my experience auditing smart contract governance models during the DAO boom, I learned that the most dangerous centralization is not in code but in the unspoken assumptions about who benefits and who bears the cost. The same principle applies here.
Core: The Centralization of Compute Power
Let’s talk about what an AI data center actually contains. It is not a few racks of servers. It is tens of thousands of GPUs—mostly NVIDIA H100s or B200s—housed in high-density cabinets, cooled by liquid or immersion systems, connected by high-speed interconnects, and powered by dedicated substations. The capital cost runs into billions. The operational cost is dominated by electricity, which can account for 40-60% of lifetime expenses.
This concentration of hardware and energy creates a new kind of monopoly: the monopoly of compute. Only a handful of companies—Microsoft, Amazon, Google, Meta, and a few specialized operators like CoreWeave—can afford to build at this scale. The rest of the AI ecosystem, from startups to researchers, must rent compute from these giants.
In the blockchain world, we have long warned about the centralization of mining power. After the Bitcoin halving, we saw hash rate concentrate in three pools. Now, the same pattern is emerging in AI. The difference is that AI compute is even more capital-intensive and less distributed. The “AI factory” is the ultimate scaling of centralization.
But here is the nuance: not all centralization is bad. There are real economies of scale in power negotiation, cooling efficiency, and network interconnect. A single large data center can achieve a PUE (Power Usage Effectiveness) of 1.1, while a distributed setup might struggle to reach 1.5. Yet the trade-off is resilience and sovereignty. If one data center goes down—due to grid failure, a natural disaster, or a cyberattack—a significant portion of AI capability is lost.
Furthermore, the infrastructure itself is a double-edged sword for local economies. The construction phase creates thousands of jobs, but once built, a data center typically employs only a few dozen engineers and technicians. The long-term tax base is property tax on a facility that can be depreciated quickly. The promise of “good jobs” may be overstated. The real beneficiaries are often the construction firms, the electrical equipment suppliers, and the power company—not the local barista or the small business owner.
Contrarian: The Hidden Costs of the AI Factory
Here is the counter-intuitive truth: the biggest risk to AI data center deployment is not technical—it is social. Trump admitted that “most Americans oppose” data centers. This is NIMBYism on steroids. And it is not just noise or visual blight. The real concerns are water consumption (a single data center can use millions of gallons of water per day for cooling), grid strain (which can raise electricity prices for residents), and the long-term liability of decommissioning a facility filled with toxic materials.
But there is a deeper blind spot. The narrative of “jobs and taxes” is a political framing that obscures the fact that AI data centers are a bet on a specific technology trajectory. We are building infrastructure for a future that may not materialize as expected. If AI demand plateaus, these facilities could become stranded assets. If energy prices spike, the economics collapse. If communities revolt, permits are delayed indefinitely.
In my work as an open source evangelist, I have seen how the promise of decentralization can be co-opted by centralizing forces. The same is happening here. The AI factory is the opposite of the distributed, resilient, community-owned infrastructure that blockchain advocates like myself dream of. Yet we cannot simply reject it. We need compute power to train models that can solve real problems. The question is: can we build it in a way that shares the benefits and mitigates the harms?
Build not for the peak, but for the plain.
Some blockchain projects are attempting to create decentralized compute networks—Golem, Akash, Render, and others. They allow users to rent idle GPU cycles from individuals or small data centers. The idea is democratic: anyone can contribute, anyone can use. But the reality is that these networks currently lack the scale, reliability, and performance to compete with a hyperscale data center. The latency is higher, the security is weaker, and the coordination costs are significant.
That does not mean they are irrelevant. In fact, the tension between centralized and decentralized compute is the same tension we see in every layer of the crypto ecosystem. The solution is not to pick one side, but to design hybrid systems that combine the efficiency of centralization with the resilience of distribution. For example, a federated model where local data centers participate in a global compute marketplace, governed by smart contracts, could be a middle ground.
Takeaway: The Conscience of Infrastructure
The AI data center boom is not just a tech story. It is a test of our ability to balance growth with equity, efficiency with resilience, and profit with community. Every state and local government that competes for these projects must ask not just "how much tax revenue will we get?" but "who will bear the cost?" The same applies to the blockchain industry. We have spent years building systems that are permissionless and trustless. But we have not yet built the infrastructure to support them at scale.
If we fail to learn from the AI data center story, we will repeat the same mistakes—centralizing power, ignoring community concerns, and betting on a single technological trajectory. The future of compute is not just about hardware. It is about governance, consent, and the long-term stewardship of shared resources.
We audit the code, but we must also audit the conscience of the infrastructure we build. The question is not whether we will have AI factories, but whether we will have the wisdom to build them in a way that serves the many, not the few.