We are now witnessing two historic and contradictory scenes unfolding simultaneously as the paradigm of the global economy shifts.
On one side, an explosion of capital is taking place. Last week, SpaceX listed on the Nasdaq, raising $75 billion to record the largest initial public offering (IPO) in history. Around the same time, Anthropic and OpenAI completed their listing preparations one after another, making 2026 the year of what is called the "AI and Space $1 Trillion Listing Trifecta."
On the other side, a powerful signal of control has been switched on. By order of the Commerce Department, all foreign access to Anthropic's top-tier AI models, Fable 5 and Mythos 5, was immediately blocked. The capital-driven rush to "sell AI to the entire world" has collided head-on with the national security brake of "the most powerful AI cannot be handed over to foreign countries."
These two scenes, which at first glance appear contradictory, in fact complete a single grand picture. It is a declaration that the United States has begun treating "inference" as a core national strategic asset and a new instrument of trade control.
A New Export of the Digital Age: 'Intellectual Services'
The life of an AI model is largely divided into one-time "training," which requires massive capital investment, and "inference," which actually generates answers and code while creating recurring revenue with every API call.
Whenever companies and individuals around the world call ChatGPT, Claude, and Gemini on US clouds and pay in dollars, an invisible form of trade takes place. This is not the sale of raw materials such as oil, but an entirely new form of intellectual service export, a "super-service export" of "renting out thinking services in real time." The United States is now building a massive super-surplus structure that offsets its manufacturing trade deficit with the intangible product called inference.
Massive Capital Investment and a Fierce AI Big Four Race
The fuel driving this enormous export engine is the massive capital pouring in through the largest IPOs in history. These funds flow directly into expanding inference infrastructure, including GPU clusters, data center power, and custom chips. As capital is invested and inference throughput increases, unit costs fall, which in turn dramatically boosts the price competitiveness of US AI models in the global market.
Add to this Nvidia's hardware and software (CUDA) ecosystem, which boasts overwhelming efficiency, and the intense "Darwinian competition" of the Big Four system comprising OpenAI, Anthropic, Google, and xAI, and the speed of evolution of the US AI ecosystem is beyond anyone's reach.
Selective Inference Trade and Korea's Crossroads
However, the recent Anthropic shutdown incident has starkly exposed the greatest internal contradiction of the "inference superpower." It is the collision between the economic ambition to export inference to the entire world, and the security logic that the highest level of inference capability could become a security threat and must therefore be controlled. The era of "selective inference trade," allowing access selectively to allies while blocking adversaries, has begun.
In the face of this enormous systemic shift in the United States, Korea is facing both clear benefits and a dilemma at the same time.
In the short term, the benefit is clear. As global inference infrastructure expands explosively, demand for HBM from Samsung Electronics and SK hynix, which hold the core of the Nvidia supply chain, will become even more solid. Furthermore, the strengths of Korean manufacturing can shine in the areas of data center power, cooling, and robotics.
In the medium to long term, however, Korea will fall into a painful dilemma. The higher the dependence of Korean companies and the government on US APIs, the greater the risk that Korea's industrial landscape itself becomes subordinate to the range of US export controls. This is because, as in the Fable incident, the United States can flip the "switch of thinking" off at any time under its own security logic.
The United States is now reinventing itself as "a country that exports thinking." Capital pours in the fuel, innovation lowers the unit costs, and regulation seals it as a strategic weapon. Now we must ask and answer. By the coming year 2030, will Korea remain merely a "thinking importer" that supplies HBM and components, or will it become a nation that defends its own "sovereign AI" with a brain of its own? Korea's AI survival strategy must begin again from precisely this question.
