AI Governance: Anti-Monopoly Principles and the AI Stack

Understanding Anti-Monopoly Law

• 💡 Anti-monopoly law is a broader concept than antitrust, encompassing various regulatory interventions beyond litigation.
• ⚖️ Historically, it addressed concentrated industries through antitrust suits or direct regulation, especially for natural monopolies.
• 📚 For further study, the anti-monopoly tradition and its application to modern markets are rich research areas.

Natural Monopolies and Their Characteristics

• 🎯 A natural monopoly exists when a market is most efficiently served by a single provider, often due to high fixed costs and low marginal costs.
• 📞 The classic example is the telephone market, where building duplicate networks is inefficient, leading to higher average costs per customer with competition.
• 📊 Key factors indicating a natural monopoly include a homogeneous good, high economies of scale, network effects, and economies of scope.

Common Carriage and Non-Discrimination

• ⚠️ Natural monopolists, while efficient, can abuse their power by charging high prices or skimping on quality.
• 🚫 Common carriage principles require service providers to serve all customers on equal terms, preventing discrimination against users or devices connecting to the network.
• 🌐 This tradition extends to modern debates like net neutrality, concerning how platforms should treat different types of traffic or users.

The AI Stack and Concentration

• 🏗️ The AI stack can be viewed as a supply chain with four key layers: hardware (chips), cloud computing infrastructure, models (LLMs), and applications.
• 📉 Each layer, particularly hardware (Nvidia, TSMC) and cloud (AWS, Azure, Google Cloud), exhibits severe concentration and high barriers to entry.
• 🧩 Foundation models also show strong natural monopoly characteristics, with significant investment required for development.

Anti-Monopoly Applications to AI

• 🤖 The concentration in the lower layers of the AI stack raises concerns about monopoly leveraging and control over application-level innovation.
• 📈 Even with competition at the application layer, control over essential inputs like chips, cloud, or models can stifle disruptive innovation.
• 💡 Potential regulatory solutions include subsidies for new entrants, non-discrimination rules for model owners and cloud providers, and rate/quality regulation for natural monopolies.
• 📚 Understanding the technology and the AI supply chain is crucial for effective AI governance and addressing competition concerns.