In 1971, US Treasury Secretary John Connally famously told European finance ministers that the dollar was “our currency, but your problem.” The Nixon administration had just suspended the greenback’s convertibility into gold, and the rest of the world had no choice but to absorb the shock. That is what happens when the global economy runs on a currency that most countries do not control and cannot influence.The same logic is apparent in the global adoption of new generative AI tools. European companies increasingly run on AI systems whose price, availability, and rules are set in the United States or China. On one side are leading Chinese models, most of which are open-weight, improving fast, and startlingly cheap. DeepSeek’s V4 Flash, for example, costs roughly $0.14 per million input tokens, compared with $3 for a comparable US model. No wonder the share of tokens used by US companies on Chinese models (through OpenRouter) has skyrocketed from under 10% in early 2025 to 45% as of July.
On the other side are American frontier labs whose models are more capable at the top end, but pricier and shaped by their own government’s whims. In June, the Trump administration temporarily blocked foreign nationals’ access to Anthropic’s newly launched Claude Mythos 5 and Fable 5 models. Vendor availability is thus becoming a salient issue for businesses, governments, and geopolitical strategists, making the debate over open-source/weight AI as much a strategic matter as a commercial one.
The choice facing Europeans seems stark: embrace the cheaper, open option and risk exposure to a state whose data governance remains opaque, or show loyalty to the pricier, closed option and pay a rising toll to vendors who will conclude that we have nowhere else to go. Either way, there is a high and compounding cost in terms of sovereignty. Unfortunately, the situation is not new for Europe, whose dependence on American cloud and software services costs the economy an estimated €264 billion ($304 billion) per year. As the global technology, media, and telecom sectors’ market capitalization surged from $7 trillion in 2000 to $34 trillion by 2024, Europe’s share collapsed from 30% to 7%, representing an $8 trillion missed opportunity. And now, generative AI is layering a new strategic dependency on top of an already lopsided cloud stack. One of us (Bouygues) runs a French manufacturer of industrial equipment whose order book reflects the current state of play: the demand from Chinese clients is breathtaking, whereas the demand from historical European partners has nearly stalled.
Viewed in this broader context, the debate over whether to go with a potentially state-directed model or one built by a private firm threatens to distract Europeans from their most pressing problem, which is that they have no seat at either table. The most urgent priority is to establish technological sovereignty, which requires domestic compute (data centers), data protection, market competition, clear liability rules, common standards, pro-European procurement, and ample energy.Europe’s default policy instrument has long been regulation, aimed less at foreign states than at foreign firms. Through what the legal scholar Anu Bradford calls the “Brussels Effect,” multinationals may voluntarily adopt EU rules across all their operations, simply because doing so is cheaper than crafting separate versions of their products or services for each market. And once that happens, other governments may shape their regulations to align with the standard Europe already set.
But while this approach worked well for privacy protections, it is far less promising with respect to AI. Analyses from the Brookings Institution and the Center for European Policy Analysis find very few jurisdictions copying the EU AI Act, reflecting the fact that the AI race runs on compute, capital, and talent—factors that European rulemaking has little bearing on.
To close its own capabilities gap, Europe needs sustained investment—a homegrown AI Marshall Plan—and strategic partnerships with other critical suppliers, not legalistic texts. Perhaps the biggest dependency is hardware. Taiwan produces over 90% of the world’s leading-edge chips, and Samsung and SK Hynix in South Korea dominate the market for memory chips, which gives them real pricing power and diplomatic leverage. But it also means that every country betting on American or Chinese models is also betting on continued stability in the Taiwan Strait.
Of course, Europe is not the only bystander. Most emerging markets and developing economies face the same dilemma, and they have even less capital and diplomatic weight to negotiate with either the US or China. Their best option is to hedge rather than pick a side. Through diversified sourcing, regional compute alliances, and South-South coordination, they can make it less likely that a single shock will cascade through their economies.
Hedging is not the same as neutrality. No country today can avoid the implications of the US-China contest, because all will inherit whichever technological ecosystem wins. The outcome will be decided not through war, but through everyone optimizing, quarter by quarter, for the cheapest model. The risk for Europe and the Global South is that they will end up with no leverage—a case of “our currency, but your problem” all over again. The greatest costs will land on those who failed to build an alternative while they still could.
Fortunately, Europe still has real assets and leverage, owing to its large internal market, relatively clean grid, rules-based tradition, and possession of the only regulatory toolkit designed to treat AI governance as a first-order issue. Among Europe’s most powerful tools is competition policy. A handful of firms setting the world’s AI prices is exactly the situation that antitrust was created to address. What Europe lacks is the will to pair that toolkit with industrial-scale investment, and the imagination to partner with Taiwan, South Korea, and other emerging economies dominating the hardware layer of the AI stack.
AI need not follow the same script as the Nixon shock. But that is what will happen if the world keeps acting as if the only choice is between Chinese state-directed models and a handful of unaccountable US firms.
