Last June, French President Emmanuel Macron said something remarkable at the G7 summit: "The United States can cut off AI overnight." Indian Prime Minister Narendra Modi voiced the same concern from the same stage. Heads of state openly warning about a fellow ally's technology infrastructure at an official G7 forum — that doesn't happen often. And as those words were being spoken in the conference hall, Anthropic's Claude service went dark without warning for several hours.
That outage was a technical glitch. Within days, it was forgotten. But the scenario Macron and Modi described hasn't gone anywhere. What they feared wasn't a technical failure — it was a political decision. A U.S. executive order blocking AI platform access for specific countries. An American company restricting services under geopolitical pressure. If you're wondering whether any of this reaches solo entrepreneurs and small businesses, the answer is yes. The scale is different; the structure is identical.
How Fast U.S. AI Became the World's Infrastructure
U.S. companies now hold over 70% of the global generative AI market. OpenAI, Anthropic, Google DeepMind, Meta AI, and Microsoft together provide most of the world's AI infrastructure. Cloud computing took a decade to capture enterprise infrastructure. Generative AI took two years — from 2023 to today — to become a core tool for knowledge work.
South Korea is no exception. As of 2024, more than 30% of the country's small businesses and solo entrepreneurs have adopted AI tools, and the overwhelming majority rely on foreign-owned platforms. Marketing copy, report summaries, customer service responses, code, images — a significant share of the work that Korean solopreneurs handle through AI travels through U.S. servers. Viewed purely through the lens of convenience, that's a smart choice. But G7 leaders said it can become a vulnerability at any moment.
Why did they raise it? Because precedent already exists. Since October 2022, the U.S. has progressively restricted exports of high-performance semiconductors to China. The same logic played out in cloud services: when sanctions hit Russia, AWS, Google Cloud, and Microsoft Azure halted or severely curtailed service in that country. Those decisions looked voluntary. They weren't made in a vacuum — they happened inside a framework of diplomatic pressure and sanctions. The infrastructure for applying that same logic to AI services is already in place. Macron wasn't warning about a hypothetical. He was pointing to a precedent.
What the G7 declaration emphasized this year was "AI security" — not military security, but the idea that access to AI services can become a bargaining chip in diplomatic and trade negotiations. Just as semiconductor export controls can be imposed by executive order, so can restrictions on AI service access. What Macron flagged was how far that logic can extend: sanctions are targeted at states, but the precedent of using technology infrastructure as leverage is accumulating — and it's no longer confined to adversaries.
The Case for Staying Dependent on U.S. AI
Not everyone agrees with this warning.
The practical counterargument is straightforward: if you cut out U.S. AI platforms, there's no alternative of comparable capability ready to fill the gap. France's Mistral, Germany's Aleph Alpha, South Korea's HyperCLOVA X, and LG's EXAONE all exist — but each still has real performance gaps against GPT-4o or Claude 3.7 Sonnet for general-purpose work. Using the best available tool is a rational choice, and dependence isn't inherently bad if what you're depending on is genuinely better.
There's also a business-logic counterargument: companies like Anthropic and OpenAI have no rational incentive to arbitrarily cut off global customers. Their revenue depends on a global subscriber base. Pulling service hurts their own bottom line first. From this view, G7 leaders were overreacting.
Both counterarguments have merit. But they're answering the wrong question. The risk isn't that these companies will voluntarily shut down service. The risk is a forced shutdown — triggered by diplomatic tension or a trade dispute. As the Russia case showed, U.S. cloud companies didn't cut service of their own accord. They cut it because the policy environment changed. A company's goodwill is only as durable as the policy context it operates in. And when policy changes, there's no advance notice.
OpenAI's terms of service include a clause allowing service termination without prior notice to users. Anthropic's terms say the same. When a blackout happens, the only option is to wait. What Macron wanted to eliminate — at the national level — was exactly that waiting.
What Your Business Should Audit Right Now
It's hard for a solo business owner to feel a head-of-state warning in their day-to-day work. But translate the structure of that warning down to the individual level, and it becomes a business continuity question.
The starting point is figuring out how many of your core workflows depend entirely on a single foreign AI platform. This isn't an argument to switch tools. It's asking whether you know which tasks would grind completely to a halt if one platform disappeared. If you don't know, that gap itself is the risk.
It's also worth getting hands-on with at least one domestic or open-source alternative. South Korea's Naver HyperCLOVA X can compete with foreign models on Korean-language document processing. LG's EXAONE has strengths in specialized document summarization; KT's Mi:dm is designed for the Korean enterprise environment. Open-source models like Llama 3 and Mistral carry no external service-shutdown risk by structure — they're not commercial services. Performance may not be the top priority for a backup, but the difference between having a tool you've used before and having never touched it is enormous when you actually need it.
It's also worth reading the terms of service and data processing clauses of your AI platforms at least once. U.S. public companies are required to include a "risk factors" section in their annual filings. It covers service interruption risk, regulatory changes, and geopolitical risks. Investors read that section before committing capital. If you're entrusting your operations to that company's service, the same logic applies. OpenAI's recent filings explicitly state that "geopolitical tensions may affect our ability to provide services." That's the company acknowledging the risk in a public document.
I'm not raising this to induce anxiety. I'm saying: before you embed yourself deeply in any one tool, read the terms of that dependence at least once. It's the same logic as checking the exchange rate before a foreign currency transaction, or reading the early termination clause before signing a lease. Treating an AI platform as mission-critical infrastructure is a contract. Contracts have conditions.
It's also worth routing at least some work to AI services built on domestic infrastructure. Kakao's Kanana, SK Telecom's A., and Wrtn are platforms insulated from diplomatic variables. They may fall short of foreign platforms in some capabilities, but when a geopolitical event occurs, having at least one option still running matters. Full independence isn't realistic. But the distance between having one more option and having none becomes much larger in a crisis than it ever appears in normal times.
The Tools You Rely On Can Disappear at Any Time
What G7 leaders were afraid of was who controls the switch. What Macron and Modi asked for wasn't a rejection of American AI — it was shared access rights, or credible independent alternatives. Translated to the individual business level, the question becomes: have you ever imagined, even once, what happens to your work when the tools you depend on are gone?
You don't need to know where the power plant is to use electricity. But you do need to know that the power plant can double its rates or cut supply. AI platforms work the same way. Use them freely — but the difference between someone who has thought through what happens when they disappear and someone who hasn't is invisible when nothing is wrong. It shows up the next time there's a blackout.
