Empires are built on control—of resources, infrastructure, and narrative. In the 21st century, those empires are algorithmic. Karen Hao’s *Empire of AI* dissects how a handful of firms have turned artificial intelligence from a tool into a system of concentrated power. That concentration is not just an ethical problem—it’s a structural vulnerability.

A few companies—OpenAI, Google DeepMind, Anthropic, Meta—control the compute power, data, and research talent needed to build frontier models. Their platforms now mediate communication, commerce, and even governance. Like historical empires, their dominance extends beyond economics; it shapes culture and policy.

AI models are trained on global data but governed by private interests. This creates a form of information mercantilism: nations and businesses depend on systems they neither own nor fully understand. The result is dependency masquerading as progress.

When so much cognitive infrastructure relies on so few entities, risk centralizes. A technical failure, policy change, or geopolitical disruption in one of these firms could cascade globally. The corporate world learned this with supply chains during the pandemic; we’ve yet to internalize it for AI.

In physical security, overcentralization is weakness. A single perimeter breach can collapse a system. The same principle applies to AI. The global network of APIs, model hosts, and compute hubs forms a soft perimeter that no single entity defends. Attack that perimeter—whether through code poisoning, insider threat, or regulation—and entire industries wobble.

Defenders claim centralization ensures safety and consistency. There’s merit to that. Regulation and trust are easier when oversight is concentrated. But centralization eventually erodes accountability. Empires always argue they’re benevolent until they can’t be.

Security and risk leaders must think geopolitically:

1. Diversify model vendors and avoid single-provider lock-in.

2. Establish independent fallback models for continuity of operations.

3. Demand contractual transparency clauses for training data, retraining triggers, and security audits.

4. Treat AI dependencies like critical infrastructure—with redundancy and oversight.

This concentration of AI power also skews global influence. Nations without frontier-model capacity risk digital dependency. Corporations without internal AI literacy risk strategic capture. Hao’s warning is subtle but clear: whoever controls the cognitive infrastructure defines the limits of imagination itself.

History teaches that empires fall not from external invasion but from internal overreach. The AI empire’s overreach is computational—growing faster than it can secure itself. The next great cyber incident may not come from a foreign adversary, but from the implosion of trust in one of these centralized systems.

Empires expand until they can no longer defend their borders. The same pattern repeats in technology. Our challenge isn’t to destroy the empire of AI, but to decentralize it—ethically, operationally, and strategically—before dependency becomes destiny.