But China can align state power, industrial policy, data access, energy, surveillance, and capital in ways a democratic country cannot replicate, nor should it want to. If the contest is only about who can centralize the most computing data, energy, and capital, Chinese state-owned enterprises will eventually narrow the gap with U.S. competitors.

Fortunately, centralization is not the only path to AI leadership. Rather, the U.S. can compete with a distributed architecture, one in which AI models are controlled by the individual. A distributed architecture plays into America’s strengths — open markets, democratic institutions, and entrepreneurial speed — rather than concentrating capability in a handful of centralized hubs.

In practice, this approach puts AI close to the decision maker — embedded in trusted devices, local networks, and mission systems — while the cloud serves as a support layer rather than a single point of dependence. Distributed systems enable individuals and teams to adapt, improve, and use AI they trust without waiting for direction from the center. That is a model of innovation and operational flexibility that authoritarian competitors will struggle to match.

The military offers a clear example. A modern architecture designed around American and allied warfighting doctrine should be edge-first: AI operating locally across units, platforms, and devices while remaining interoperable across the force. Put simply, AI should work where decisions are actually being made.

That could mean a maintainer troubleshooting aircraft on a flight line in the middle of the night, a small team operating in a communications-denied littoral outpost, a shipboard watch team managing operations under emissions control, or a medic moving through a blackout zone after an attack. In those environments, forces cannot depend on constant access to the cloud. They need trusted capabilities that continue to work in degraded, disconnected, and high-risk conditions — exactly when the mission matters most.

The technology behind distributed AI is advancing quickly. Intelligence is becoming smaller, cheaper, and more portable. Capabilities that once required large data centers can increasingly run on laptops, workstations, tablets, and edge devices. But the bigger shift is toward specialization. A pilot, maintainer, doctor, medic, or front-line worker does not need a single all-purpose model trying to answer every question. They need AI that understands their mission, their data, their permissions, and the operational context in which they are working.

That architecture depends on many small, specialized models, each tailored to specific missions, roles, and workflows. A centralized system can be highly effective in stable environments with reliable connectivity. It becomes far less useful when communications are degraded, time is compressed, and decisions cannot wait. In those conditions, a distributed network of locally controlled intelligence is the only way AI can deliver reliable operational value.

THE AI ARMS RACE HAS A TRACKING PROBLEM

China’s approach prioritizes centralized control. America’s advantage should come from accountable empowerment. We need systems that are distributed by design, interoperable across allies, and resilient enough to function when communications fail or conditions deteriorate.

After Beijing, the U.S. should keep talking about guardrails. But the guardrails alone are not a strategy. The real strategic task that lies ahead is to build an AI architecture that reflects how free societies compete and win: resilient, decentralized, and trusted by the people who depend on it.

Dr. Jason Rathje is the President of Public Sector at webAI; former founding Director of the Department of Defense Office of Strategic Capital and co-founder of the U.S. Air Force AFVentures