The weaponization of artificial intelligence (AI) as an economic tool is shaking up the international order without firing a shot.

Concerns are rising that even if foreign countries can use the latest U.S. AI models, they may have to pay higher fees than U.S. citizens, or face limits on usage and speed.

In the race to build AI ecosystems, countries that fail to establish at least a basic supply chain could find themselves relegated to the status of "tenant farmers."

The reality is that, with the exception of the United States and China, it is difficult for any single country to secure the infrastructure and capital required to independently lead a large-scale AI ecosystem.

For this reason, calls are growing for countries to at least position themselves as an essential part of the existing AI supply chain.

The argument is that to strategically deter adversaries and increase bargaining power, countries must secure their own critical leverage.

The U.S. government's control over AI has gone a step further than its existing semiconductor export restrictions targeting China.

When the U.S. blocked foreign nationals from accessing Anthropic's advanced AI models, Claude Mythos 5 and Claude Fable 5, it did not make exceptions for allies.

This marks the first instance of export controls being applied to cloud-deployed AI models.

Although the U.S. government recently lifted the export controls after 18 days, countries relying on the infrastructure of U.S. AI companies have been given a harsh reality check.

The Financial Times (FT) recently reported that foreign clients were also excluded from the initial access list for OpenAI's next-generation model, GPT 5.6, which was released recently.

In terms of AI infrastructure, what is effectively an "export tax" has also emerged.

In January, President Donald Trump signed a proclamation imposing a 25% tariff on AI semiconductors, such as the H200, that are manufactured abroad, imported into the U.S., and then re-exported.

This measure is designed to tighten the noose and prevent AI semiconductors from falling into China's hands.

Consequently, even more radical scenarios are being discussed among experts.

Some raise the possibility that an AI superpower could control the usage of AI model tokens for foreign nations, thereby driving up costs.

This is the exact opposite of the current situation, where the European Union (EU) is attempting to levy a digital services tax (DST) on foreign AI companies.

If the U.S. firmly secures dominance over AI models and other infrastructure, the EU will not be in a position to tax U.S. companies; rather, the U.S. could impose export taxes on them.

The moment U.S. AI models become irreplaceable, the U.S. will effectively rise to a position of absolute dominance.

Even allies may not be exempt from exploitation.

Early signs of this have already appeared in the GPU ecosystem monopolized by Nvidia.

Not only foreign companies but also governments are forced to walk on eggshells around Nvidia just to secure GPU allocations.

The physical limitations of AI infrastructure are also becoming visible, which is concerning.

According to a report by global real estate services firm CBRE, the vacancy rate for AI data centers in some parts of North America has fallen below 1%.

With supply extremely tight, token consumption is growing exponentially.

Earlier this year, major U.S. companies tended to encourage AI usage, but as token costs surged, they appeared to be imposing restrictions once again.

Now, AI cloud companies have begun restricting the capacity allocated to their clients altogether.

Google recently failed to allocate all the infrastructure capacity requested by Meta.

Notably, companies are also beginning to use tokens as the unit for billing.

According to a report by The Information on June 29, Anthropic changed the way it bills Amazon, moving from computing hours to the number of tokens used.

Furthermore, tokens are more than just a simple billing unit for calculating AI model usage fees.

They are closer to the final unit of consumption in actual services and workplaces, encompassing computing resources, data centers, and electricity.

In short, securing tokens means gaining access to AI infrastructure.

Han Un-hee, CEO of TRS Insight, said, "Token access is more than just a matter of cost; it is a strategic condition that determines the stability and quality of access to the 'frontier AI full-stack.' If this structure expands, token access could potentially become a bargaining chip in national-level trade and security negotiations."

This means that access to high-performance models, priority processing rights, usage quotas, and the deployment of dedicated cloud and data centers within a specific country could become key negotiating terms in contracts between U.S. AI firms and foreign companies or governments.

Han predicted, "Some countries might seek to secure high-performance model access, high usage limits, low latency, and support for building and operating data centers within their borders from U.S. frontier AI firms, in exchange for easing digital services taxes or data localization requirements on U.S. cloud and AI services. There is a strong possibility that token access will emerge as a new negotiating variable that determines the competitiveness of nations and corporations."

The global concentration of economic power in the AI ecosystem is expected to further accelerate this trend.

In particular, the shift toward AI and the resulting polarization of wealth are extending beyond economic issues to affect the political sphere.

This means social sensitivity is rising over how to share the wealth generated by AI.

In the United States, politicians have already begun to take action.

The Trump administration has proposed a plan for the government to directly acquire stakes in U.S. AI companies.

MAGA, President Trump's core support base, is backing the distribution of profits generated by AI.

On the Democratic side, some are even arguing that the government should collect shares of AI companies as taxes to establish a sovereign wealth fund for distribution.

With forecasts suggesting that AI will replace human labor and reduce tax revenues, the likelihood of U.S. politicians intervening in AI profits is growing by the day.

South Korea's economic structure, where the share of Samsung Electronics and SK Hynix is growing daily, also clearly illustrates the concentration of wealth driven by AI.

Recently, performance bonuses at Samsung Electronics led to feelings of relative deprivation, escalating into a source of social conflict.

This indicates that excess profits and tax revenues related to the AI sector have also become major political and social agendas in South Korean society.

In response to the trend of weaponizing AI infrastructure, South Korea is also scrambling to devise countermeasures.

Weight is increasingly being given to calls for a multi-dimensional response that goes beyond the existing discourse on "sovereign AI," such as LLM development.

The forum jointly held on June 25 by the Presidential Advisory Council on Science and Technology and the National Economic Advisory Council under the theme of "AX Challenges and Responses: National Strategy for Innovation, Growth, and Inclusion" is an extension of such efforts.

In particular, attention is focused on whether the plans to strengthen the competitiveness of physical AI and advanced strategic industries, announced at the "National Presentation on the Three Mega Projects for South Korea's Great Leap Forward" recently hosted by President Lee Jae-myung, will manifest as a multi-dimensional response.

Kim Sung-sik, Vice Chairman of the National Economic Advisory Council, emphasized at the forum, "We must soberly face the AI lock-in strategy being built by global Big Tech."

This is a warning that technological dependence can lead to economic exploitation.

Kim stressed, "We must leverage our strong manufacturing capabilities, high-quality data, and device competitiveness to leap forward as co-designers of AI infrastructure. To build an AI full-stack strategy, a national-level cooperation platform must be fully mobilized."

Europe is also facing an urgent crisis.

United Kingdom is debating whether to develop its own LLM to reduce its technological dependence on the United States.

The Trump administration's regulatory measures on Claude Mythos and Claude Fable added fuel to this debate.

Some within the UK government are voicing opinions that the country must secure its own independent AI model.

On the other hand, some argue that since it is an area where competing with the U.S. is fundamentally difficult, the UK should secure competitiveness in other AI fields, but even this is being held back by insufficient AI infrastructure.

France, which has a well-established power grid infrastructure, is in a better position than the UK.

France recently beat out other European nations, including the UK, to secure tens of billions of dollars in AI infrastructure investment from SoftBank.

France has also built an AI model called "Mistral."

Under these circumstances, some point out that rather than going head-to-head with the U.S. on AI models, effectively leveraging Europe's existing strengths in the AI ecosystem would increase its bargaining power in the era of "AI weaponization."

Bloomberg recently noted in a column that attention should be paid to the critical roles played by ASML, Infineon, and Zeiss in the global semiconductor supply chain.
 

(Photo: Yonhap News)