For decades, American leadership spread because we exported more than products—we exported platforms that others could build on. The dollar is the world’s currency. Internet protocols developed in America became global standards. American companies cultivated and contributed to software ecosystems that millions of developers adopted and improved.
Artificial intelligence will be no different.
Policymakers need to pick a side between two competing visions for AI. One vision fixates on a single layer of the AI stack, expecting the world will rent intelligence from a tiny number of providers. The other says America should lead by promoting its AI ecosystem—infrastructure, hardware, and open technology. Governments, researchers, startups, and enterprises that want their own AI capabilities will therefore remain anchored to the American ecosystem.
The idea that all other countries are eager to rent intelligence in perpetuity is naive and divorced from reality. Countries do not want to permanently rent their critical infrastructure. They want sovereign control over their energy systems, communications networks, and healthcare data. AI is the most important technological evolution in generations. While most countries cannot manufacture frontier AI chips or build the underlying computing infrastructure themselves, they can build applications, customize models, and innovate on top of platforms provided by trusted partners. That is America’s opportunity.
At the core of this vision is the recognition that open models are winning and will probably dominate AI deployments into the future. Open models, which are freely available to download, modify, run, and re-distribute, are often misunderstood. They are not simply “free AI.” They are more secure, private, and customizable.
A smear campaign from leading purveyors of closed models has tried to scare the public into believing that open models will easily enable criminal gangs, terrorists, and foreign adversaries to develop mass bioweapons, hack critical infrastructure, and undermine our society in myriad ways. However, the first major AI-powered hacks have all been conducted by closed models rather than open ones—and open models have been the targets’ only chance at defending themselves. Closed models have suffered from innumerable privacy breaches. Open models can run locally such that no private information even leaves the user’s machine. And while closed models are expensive to run and even more expensive to customize, open models are free to run and, after customization, can achieve far greater performance on specialized tasks. Open models will win on security, privacy, and customization, because that is the overwhelming story of open-source software over the last three decades.
While security and privacy are prerequisites for many deployments, the massive upside of AI will come from customization. A hospital wants an AI assistant trained for healthcare. A manufacturer wants one optimized for factory operations. Governments want AI that reflects their country’s laws, language, culture, and security requirements. Open models make those adaptations possible while allowing organizations to retain control over sensitive data and mission-specific knowledge.
Some argue that openness creates unacceptable risks. Policymakers should certainly take security seriously. But they should also understand how modern AI systems actually work. Much of today’s progress comes through reinforcement learning, where models improve by receiving feedback on which responses are better. Like any optimization system, reinforcement learning can sometimes produce unintended behavior, including “reward hacking,” where models find shortcuts that satisfy the evaluation criteria without accomplishing the intended objective. This is an active area of research across industry and academia, and it illustrates why continuous evaluation and testing matter. Closed models have the same risks, but fewer people to anticipate and react to problems—dozens of engineers at a single closed model company versus millions of eyeballs among the developer community.
America should aim to lead not only in models, but also in the tools, benchmarks, evaluation frameworks, and computing infrastructure that surround them. Ultimately, the central policy question is not whether AI should be open or closed—policymakers should leave that question to the market. Both approaches will continue to coexist because they serve different markets and use cases. The most important policy priority is global adoption of the American AI ecosystem.
The national interest is best served when the world chooses American technology. The United States became the world’s technology leader because other countries adopted American hardware, software, standards, and platforms. If America’s strategy is simply to tell the world to rent intelligence indefinitely from a small number of providers, many countries will likely choose the Chinese ecosystem, with its open models promising greater ownership and autonomy. If, however, the United States provides the world’s compute infrastructure, embraces open innovation, and promotes the platforms upon which others build, American technology will win.
Policymakers should treat compute as strategic infrastructure. Export promotion is in the national interest. Policies that increase the supply of American compute, enable open models, encourage private investment, and broaden access for startups, researchers, and allies will generate far greater long-term returns than quixotic efforts to close off intelligence for the benefit of a few at the expense of the many. The race for AI leadership will not ultimately be won by who builds the tallest wall around intelligence. It will be won by who builds the ecosystem the rest of the world chooses to build upon.
Michael Frank is the Co-Founder and CEO of Radiant Intel, an agentic AI platform that delivers clear, actionable geopolitical and macro-risk insights. He is also a 2430 Fellow with the 2430 Group and the Founder and CEO of Seldon Strategies. Previously, he served as a Senior Fellow at CSIS’s Wadhwani Center for AI and Advanced Technologies and led the Economist Intelligence Unit’s Asia technology and geopolitics policy research. Michael holds a master’s degree in public policy from the University of Chicago and a bachelor’s degree in international relations and economics from Colgate University.
This article was originally published by RealClearDefense and made available via RealClearWire.
