AI Is Centralizing Power — Can Blockchain Actually Fix That?

Here's a tension that doesn't get discussed enough: AI, the technology we celebrate for democratizing access to knowledge, is simultaneously one of the most centralizing forces in modern tech. A handful of companies control the data, the compute, and the models. A new editorial published in ACM SIGMIS Database argues that blockchain isn't just a counterweight to that centralization — it's a necessary complement. The authors call the intersection "Decentralized Intelligence," and their case is worth examining.

The Centralization Problem Nobody Wants to Talk About

The editorial, authored by researchers from Marywood University, University of Scranton, University of Nevada Reno, and Cal State Dominguez Hills (arXiv:2603.11299), opens with a blunt framing. One of the researchers quotes Ali Yahya from a16z: "AI is communist. Crypto is libertarian." It's deliberately provocative, but the underlying observation is grounded in structural reality.

AI centralizes because its economics demand it. Training GPT-4 reportedly cost over $100 million. State-of-the-art models require access to massive GPU clusters, petabytes of data, and infrastructure that only a few companies can afford. This creates three compounding problems:

1. Data monopolization. The best models need the most data, and the companies with the most data build the best models. Smaller players can't compete for access. Selective data inclusion introduces systemic biases that go unchecked.
2. Resource monopolization. The compute cost alone creates a "winner-takes-all" dynamic. If you can't spend nine figures on training runs, you're not at the frontier.
3. Concentration of power. When a few companies control the most capable AI systems, they also shape the policies, set the terms of access, and influence regulation in their favor. Users become dependent on a handful of providers for AI-driven services.

None of this is speculative. It's the current state of the industry. OpenAI, Google, Meta, Anthropic, and Microsoft collectively control the models that most of the world uses. The editorial frames AI as a "sustaining innovation" — one that reinforces incumbent advantage rather than disrupting it.

Blockchain as Structural Counterweight

The authors position blockchain as a fundamentally opposing force — not because of crypto hype, but because of architectural properties. Blockchain distributes control across a network instead of concentrating it. It's designed to enable cooperation without intermediaries.

The editorial maps four specific tensions:

That last row is underappreciated. As generative AI floods the internet with synthetic content, blockchain-based provenance tracking (NFTs, content attestation chains) becomes one of the few reliable mechanisms for establishing that a piece of content was created by a human. The editorial argues this isn't a niche use case — it's foundational for maintaining trust in media.

Complements, Not Competitors

The more interesting argument isn't that blockchain opposes AI. It's that the two technologies fix each other's weaknesses.

What blockchain does for AI:

• Decentralized training and inference. Distributing model training across a network reduces single-point-of-failure risks and breaks data monopolies.
• Auditable AI. Blockchain's immutable ledger enables verifiable audits of training data inputs and model outputs, creating accountability that doesn't exist today.
• Zero-Knowledge Machine Learning (ZKML). This is perhaps the most technically compelling development. ZKML uses blockchain to verify that an AI computation was performed correctly without revealing the underlying data. You get integrity verification and data privacy simultaneously.
• Proof-of-humanity. Blockchain-based mechanisms to confirm content was generated by a human, not an AI — a growing necessity as synthetic media improves.

What AI does for blockchain:

• Smart contract automation. AI can generate, audit, and detect vulnerabilities in smart contract code, reducing the risk of exploits.
• Content curation on decentralized platforms. Without a central authority to moderate, AI-driven filtering becomes essential for managing spam and misinformation.
• Security monitoring. AI-powered transaction monitoring and MEV (miner extractable value) defense protects users from exploitation on-chain. AI can also flag deepfakes and misinformation on blockchain platforms.

Decentralized Intelligence: The Research Agenda

The editorial coins the term "Decentralized Intelligence" (DI) for this convergence — an interdisciplinary research area focused on building intelligent systems that function without centralized control.

The authors trace the intellectual lineage back further than you'd expect. Distributed computing in the 1960s. Multi-agent systems in the 1980s. Swarm intelligence and peer-to-peer networks in the 2000s. Google's Federated Learning in 2017, which allowed machine learning across devices without centralized data storage. DI isn't a new idea — it's a reframing of decades of work, now made urgent by the concentration of AI power in a small number of companies.

The editorial proposes six concrete research initiatives:

1. Government-funded open AI systems — public alternatives to proprietary models
2. Research consortia — universities and industry sharing resources for decentralized AI research
3. Regulatory frameworks — rules designed for decentralized AI's unique challenges (privacy, accountability, liability)
4. Decentralized data cooperatives — individuals pooling data and collectively controlling its use for AI training
5. Standardization bodies — interoperability standards to prevent vendor lock-in
6. Open-source development platforms — a decentralized GitHub for AI models and applications

Where This Gets Real (and Where It Doesn't)

The editorial is an honest framing of a real structural problem. AI centralization isn't a theoretical risk — it's observable today. And the blockchain complementarity argument is technically sound: ZKML, federated learning, and decentralized compute are real technologies with active development.

But there are important caveats the editorial acknowledges only briefly:

"The mere use of a blockchain does not guarantee decentralization; instead, it depends heavily on its design, governance, and implementation."

This is the crux. Many "decentralized" systems end up recentralizing around whoever controls the protocol, the validators, or the governance tokens. Bitcoin mining is dominated by a handful of pools. Ethereum's validator set is concentrated among large staking providers. Putting "blockchain" on AI doesn't automatically distribute power — it shifts where the bottlenecks live.

The data cooperative idea is compelling in theory, but faces massive coordination problems. Getting millions of people to pool data and agree on governance is a social challenge, not a technical one. And government-funded open AI systems already exist (think BLOOM, or France's Mistral backing) — the track record is mixed.

None of this invalidates the core argument. It just means that "Decentralized Intelligence" is a direction, not a destination. The hard work is in the governance design and the economic incentives, not just the cryptography.

Source: Jin, Z., Li, X., Joshi, K.D., & Deng, X. "Counterweights and Complementarities: The Convergence of AI and Blockchain Powering a Decentralized Future." ACM SIGMIS Database, Vol. 56, No. 2, April 2025. arXiv:2603.11299