# America Is Building the Great Wall of AI > Source: > Published: 2026-07-23 11:13:04+00:00 TL;DR — Key Takeaways **Chinese open-weight models are competing on more than performance.** Kimi, Qwen, DeepSeek and others offer lower prices, customization and greater control over where models run.**Security concerns deserve scrutiny, but they should not become cover for protectionism.** Fraud, espionage, access-control evasion and intellectual-property theft should be targeted directly.**Open models threaten the scarcity economics of closed AI providers** by making intelligence cheaper, more portable and increasingly interchangeable.**A broad U.S. crackdown could hurt American developers more than foreign rivals**, since downloadable models would remain available across much of the rest of the world. For decades, the United States criticized China for putting walls around its digital economy. China blocked foreign platforms, controlled the flow of information and forced American technology companies to choose between accommodating Beijing and losing access to the Chinese market. Now the roles are beginning to reverse. Chinese companies are releasing increasingly capable open-weight AI models that can be downloaded, customized and deployed almost anywhere. Kimi, Qwen, DeepSeek and others are spreading far beyond China, competing not only on performance but also on price, accessibility and the freedom they give developers to control how and where the models run. Some of America’s most powerful AI companies see a threat. They may be right, but perhaps not entirely for the reasons they want Washington to believe. The Wall Street Journal reports that executives from OpenAI, Anthropic and other American AI companies are sounding alarms about Chinese models. Their concerns include national security, censorship, hidden vulnerabilities, intellectual property theft, and the possibility that China could use open models to gain influence over the global AI ecosystem. Every one of those risks deserves serious scrutiny. None automatically justifies building a wall around the American AI market. America may believe it is constructing a defensive barrier to keep Chinese AI out. It could instead build a wall around American developers, businesses and researchers, protecting the companies inside from the very competition they need to remain global leaders. ### The Trap Was Already Set This is not simply another chapter in the contest between the United States and China. It is a fight over whether artificial intelligence will remain a scarce, proprietary product or become abundant, interchangeable infrastructure. OpenAI, Anthropic and their peers have spent enormous sums trying to make their models indispensable. They want AI embedded in every business, profession, government agency and aspect of our daily lives. Their extraordinary valuations assume they will succeed. But indispensability comes with consequences. Once a technology becomes essential to the operation of society, access to it no longer remains solely a private matter between its owner and its customers. People begin asking who controls it, who can afford it, who can be denied access and whether a handful of companies should determine its permissible uses. That is the central argument of my forthcoming book, *The Indispensability Trap*, where I examine more fully how electricity, telephony and other foundational technologies followed this progression. The companies building them sought the economic rewards that came from becoming essential. Once they succeeded, however, the public and government began imposing obligations involving access, pricing, reliability, interoperability and nondiscrimination. The infrastructure remained enormously valuable. What changed was the ability of its owners to extract unlimited scarcity rents from something society could no longer function without. The American AI labs want the economic rewards of indispensability without the commoditization or public obligations that historically accompany it. They want intelligence to become essential while access to intelligence remains a premium, metered service controlled by its private owner. That was never going to last. Open-weight models apply the commercial pressure. Governments apply political and regulatory pressure. Businesses demand portability and alternatives because they do not want their operations dependent on a single provider. The public resists a future in which several companies control both the price of intelligence and the boundaries of what people may do with it. These pressures are not creating the indispensability trap. They are springing the trap the labs set for themselves when they promised AI would become indispensable to everything. ### Is Abundant Intelligence a Dystopia? Dean Ball, OpenAI’s head of strategic futures and a former Trump administration AI official, recently offered a particularly revealing reaction to the rise of Chinese open models. Responding to Kimi K3, Ball warned of a future he described as “full AI communism,” in which artificial intelligence becomes a public good or a form of digital public infrastructure. He called the prospect a “dystopian hellscape.” Ball expressed that view in a personal post, not a formal OpenAI policy statement. Nor did he explicitly call for the government to regulate Chinese models out of existence. But his language exposes the economic stakes behind this debate. What he calls AI communism may simply be the commoditization of the model layer. That is not the end of AI commerce. It is the end of extraordinary scarcity rents earned by selling access to intelligence. Electricity did not lose its economic importance when it became broadly available, relatively inexpensive and regulated. It became vastly more valuable to society. The greatest economic gains migrated toward the products, services and industries powered by electricity. The same happened with telephony. Once basic connectivity became widely available, value expanded into the businesses that used the network rather than remaining entirely with the owner of the wires. AI may follow a similar path without becoming a government-owned service. A public good, a public utility and a government enterprise are not the same thing. Open models could make the foundation of AI look more like Linux or the internet’s open protocols than a federal electricity bureau. The model layer could become broadly accessible and increasingly substitutable while compute, proprietary data, security, applications, agents, integration, distribution and customer relationships remain fiercely competitive commercial markets. Intelligence becoming cheaper would not eliminate the AI economy. It would move value up the stack. Companies creating useful applications, controlling valuable workflows and delivering measurable outcomes would capture more of the value. The underlying model would remain important, just as electricity remains important, but customers would care less about who produced each unit of intelligence. That is precisely what the closed-model companies have reason to fear. ### Theft Is Not Competition None of this means Anthropic’s allegations against Chinese AI companies should be dismissed. Anthropic says DeepSeek, Moonshot and MiniMax used approximately 24,000 fraudulent accounts to generate more than 16 million interactions with Claude. According to Anthropic, these operations evaded geographic restrictions and targeted Claude’s coding, reasoning, computer-use, vision and agentic capabilities for distillation. If those allegations are accurate, the conduct deserves investigation and consequences. American companies should not have to tolerate fraudulent account networks, deliberate evasion of access controls or the industrial-scale extraction of proprietary capabilities. But we need to define the violation correctly. Distillation itself is not inherently theft. It is a widely used technique throughout the AI industry. American labs distill their own large models into smaller, more efficient ones. Models are routinely trained using synthetic outputs generated by other models. The legitimate dispute concerns unauthorized access, deceptive conduct, violations of contractual restrictions and the possible appropriation of proprietary technology. The remedy should target the misconduct. Punish theft, fraud and espionage. Sanction the companies and individuals responsible when the evidence supports it. Test foreign models for backdoors, data leakage, censorship, covert behavior and dangerous capabilities. Impose stringent provenance and security requirements on classified systems, defense applications and critical infrastructure. But do not turn legitimate intellectual-property concerns into protection from competition. A Chinese model is not necessarily safe because its weights are available. It is also not necessarily a national-security threat merely because it is cheaper than an American model. If every technological development that threatens an incumbent’s margins is relabeled a national-security concern, national security becomes industrial policy with better marketing. ### The Wall Works Both Ways The practical problem with constructing a Great Wall around American AI is that the models Washington might seek to exclude have already escaped into the global ecosystem. Their weights have been downloaded, copied, quantized, fine-tuned, merged and incorporated into other systems. There is no API key to cancel and no central server to shut down. Over time, provenance will become even harder to establish as models are modified, combined and concealed beneath applications that route work among several different systems. America can create friction. It can regulate American companies, domestic cloud providers, official repositories and federally funded institutions. It can prohibit certain transactions, restrict deployment in sensitive environments and prevent federal agencies from using particular models. What it cannot do is make those models disappear from the rest of the world. That creates an asymmetrical result. American restrictions would bind compliant American companies while doing little to stop foreign competitors, criminal organizations or hostile governments. Developers in Europe, Asia, Africa, Latin America and the Middle East could continue improving the models, building tools around them and incorporating them into applications, devices, factories, robots and national infrastructure. American companies, meanwhile, could be left paying premium prices for approved closed models while competitors elsewhere use less expensive and more customizable alternatives. American startups would face higher development costs. Researchers could lose access to models available to their peers overseas. Enterprises could become more dependent on a small group of domestic providers facing less pressure to offer portability, improve efficiency or lower prices. The labs inside the wall would remain active. They would release new models with better benchmarks and additional capabilities. The market might continue to look innovative because each generation would improve on the one before it. But improvement inside a protected market is not the same as global competitiveness. America could become locked into a sophisticated but increasingly stagnant AI ecosystem. We might not recognize it immediately. Protected stagnation rarely presents itself as stagnation. It arrives dressed as stability, safety and support for national champions. By the time the difference becomes obvious, developers elsewhere may have built a far richer ecosystem around open models. ### China Does Not Need the Best Model China’s strategy does not require producing the world’s single most capable model. A model that is inexpensive, customizable and good enough can travel farther than a superior model available only through a costly, controlled API. It can become the foundation for startups, government systems, industrial automation, robotics, vehicles and consumer devices across countries that cannot afford—or do not want permanent dependence on—American providers. Every deployment creates familiarity. Familiarity produces tools, integrations, expertise and standards. Those assets make the next deployment easier. That does not make China an altruistic champion of openness. Its open-model strategy serves Chinese economic, industrial and geopolitical interests. Open weights provide a way to expand Chinese influence without requiring every user to send data to a server in China or enter a direct commercial relationship with a Chinese company. China has recognized that commoditization can be a strategy. American incumbents, whose economics depend on preserving scarcity, are tempted to treat that commoditization as something Washington must stop. If Washington removes American developers from this ecosystem, it will not prevent the ecosystem from growing. It will surrender American participation in it. ### The Market Above the Model The most likely AI future is neither completely open nor completely closed. A small number of frontier labs will continue spending heavily to develop the most capable models. They will charge premium prices for tasks requiring the best available reasoning, coding, scientific or agentic performance. Just behind that frontier will be an expanding open-weight layer. These models will handle private, specialized, high-volume and cost-sensitive workloads. They will run inside companies, on sovereign infrastructure and eventually on local devices. Enterprises will use both. Applications and agents will route each task to the most appropriate model based on capability, cost, speed, privacy and risk. Users may never know which model performed a particular step, just as most people do not know which power plant generated the electricity running an appliance. As the application abstracts the model underneath it, model brands become less important. Intelligence becomes more interchangeable. Economic value migrates toward proprietary data, workflows, integration, distribution and customer outcomes. That is not AI communism. It is a competitive market moving up the stack. The American labs may not like where the market is moving. But preserving their present business models is not the same as preserving American leadership. The policy line should be clear: Protect American intellectual property without protecting artificial scarcity. Test foreign technology without automatically banning it. Restrict deployments where specific security evidence justifies restrictions. Do not construct a protected domestic market simply because open competition threatens the economics of today’s national champions. Open models, government policy and public demands for access are springing the indispensability trap, but they did not create it. The trap was set the moment the AI industry promised that its technology would become essential to everything. The Great Wall of AI will not stop open models from spreading. It will only determine which side of the wall gets to build with them. America may believe it is keeping Chinese AI out, only to discover that it has kept American AI in—protected from the very competition it needed to remain great.