Last June, French President Emmanuel Macron said something startling at the G7 summit: "The United States could cut off AI overnight." Indian Prime Minister Narendra Modi, sitting in the same room, voiced the same concern in public. It's rare for a head of state to level a warning like this — about an ally's own technology infrastructure — from an official G7 stage. And around the very moment that warning was being spoken, Anthropic's Claude went down for several hours without any notice.
That outage, caused by a technical glitch, was forgotten within days. But the scenario Macron and Modi described didn't go away. What they feared wasn't a technical failure — it was a political decision: a single executive order cutting off a country's access to an AI platform, or a US company restricting its service under geopolitical pressure. Does this reach all the way down to a one-person business? Yes. Only the scale is different; the structure is the same.
How Fast US AI Platforms Became Global Infrastructure
US companies now control more than 70% of the global generative AI market. OpenAI, Anthropic, Google DeepMind, Meta AI, and Microsoft supply most of the world's AI infrastructure between them. Cloud computing took a decade to take over enterprise infrastructure; generative AI became the core tool of knowledge work in just two years after 2023.
Korea is no exception. As of 2024, AI adoption among Korean small businesses and solo entrepreneurs has passed 30%, and the overwhelming majority of that runs on foreign platforms. Writing marketing copy, summarizing reports, handling customer inquiries, writing code, generating images — a large share of the AI work Korean solo entrepreneurs do routes through US servers. Purely as a matter of convenience, that's a good choice. But G7 leaders just said it could become a vulnerability at any time.
G7 leaders raised this because there's already precedent. Since October 2022, the US has progressively restricted exports of high-performance semiconductors to China. The same pattern played out with cloud services. When sanctions hit Russia, AWS, Google Cloud, and Microsoft Azure suspended or severely curtailed service there. Those decisions looked voluntary, but they were made inside a context of diplomatic pressure and sanctions regimes. The structure that would let AI services be restricted by the same logic already exists. What Macron warned about wasn't hypothetical — it was precedent.
In this G7 declaration, the leaders' shared emphasis was on "AI security." Security here doesn't mean military security — it means that access to AI services can become a bargaining chip in diplomatic and trade negotiations. Just like semiconductor export controls, access to AI services can be restricted by a single US executive order. What Macron was pointing to was how far that logic could extend. Sanctions are typically aimed at whole nations, but the precedent building up is that signals of technology-infrastructure control could just as easily apply to democratic allies.
The Case for Depending on US AI Anyway
Not everyone agrees with this warning.
The practical objection comes first: if you rule out US AI platforms, there's no equally capable alternative ready to use today. France's Mistral, Germany's Aleph Alpha, Korea's HyperCLOVA X, LG's EXAONE — these models exist, but there are still real gaps where they can't reliably replace GPT-4o or Claude 3.7 Sonnet at general-purpose tasks. The argument is that using the best tool available is simply rational — dependence isn't the problem, using the better product is.
There's also a business-logic objection: companies like Anthropic and OpenAI have no incentive to arbitrarily cut off service to global customers. Their revenue depends on a global subscriber base. Cutting service would hit their own bottom line first. By this view, the G7 warning is an overreaction.
Both objections have merit. But they answer a different question entirely. The risk isn't that a company voluntarily cuts service — it's that service gets cut off by force, in the middle of diplomatic tension or trade disputes between governments. As the Russia case shows, US cloud companies didn't choose to pull out on their own. Corporate goodwill can change the moment the policy environment changes. And policy environments don't change with advance notice.
OpenAI's terms of service include a clause allowing it to suspend service to users without advance notice. Anthropic's terms say the same. When a blackout hits, all a user can do is wait. What Macron wanted was to eliminate that waiting at the national level.
The Checklist Your Business Needs Right Now
It's hard for a solo entrepreneur to feel a head-of-state warning directly. But bring the same logic down to the individual level, and it becomes a business-continuity problem.
The starting point is checking how much of your core workflow depends on a single foreign AI platform and nothing else. This isn't a call to switch tools. It's a question of whether you know which tool, if it vanished, would bring your work to a complete stop. If you don't know, that itself is the risk.
It's worth considering keeping at least one domestic or open-source tool in regular use alongside your main one. Naver's HyperCLOVA X and Clova Studio hold their own against foreign models on Korean-language document tasks. LG's EXAONE is strong at summarizing specialized documents, and KT's Mi:dm offers a service tailored to Korean business environments. Open-source models like Llama 3 or Mistral aren't commercial services, so they carry no structural risk of an external shutdown. Performance might not be the top priority here, but there's a world of difference between having a backup you already know how to use and not even recognizing its name.
It's also worth reading, at least once, the terms of service and data-handling clauses of the AI platform you use. US publicly traded companies are required to include a "risk factors" section in their annual disclosures, spelling out the possibility of service disruption, regulatory shifts, and geopolitical risk. Investors read that section before they decide to invest. By the same logic, anyone entrusting their work to that company's service has reason to read it too. OpenAI's recent disclosures explicitly state that "geopolitical tensions could affect our ability to provide our services." That's a risk the company itself has acknowledged publicly.
I'm not raising this to stoke anxiety. I'm saying that before you become deeply dependent on a specific tool, it's worth reading the terms of that dependence at least once. It's the same logic as checking exchange-rate risk before a foreign-currency trade, or checking the early-termination clause before signing a lease. Making an AI platform your operational backbone is also a contract, and contracts have conditions.
You might also consider routing at least some of your work to AI services built on Korean data centers. Kakao's Kanana, SK Telecom's A. (Adot), and Naver's Wrtn all run on infrastructure independent of diplomatic variables. They may fall short of foreign platforms on features, but what matters is that if a diplomatic incident hits, you still have at least one option left. Full independence isn't realistic. But the distance between having one extra option and having none shows up much larger in a crisis than it does on an ordinary day.
The Tool You Use Could Disappear at Any Time
What G7 leaders feared was who holds the switch. Macron and Modi weren't demanding that the world reject American AI — they were asking to share control of that access, or to secure independent alternatives. Translated for a solo entrepreneur, that becomes a single question: have you ever once imagined what happens if the tool you use right now disappears?
You don't need to know where your power plant is to use electricity. But you do need to know it could double its rates or cut supply altogether. AI platforms are no different. Use them for the convenience they offer — but the gap between someone who has once imagined life without it and someone who hasn't stays invisible on an ordinary day. It only shows up when the next blackout hits.
