By Georgios Papadopoulos*
In terms of the ability to produce and control the most advanced AI technologies today, the United States holds a clear lead. This superiority is based on political, economic and investment factors, which have created a self-reinforcing advantage in capital, computing power and talent.
Europe is clearly lagging behind, while China is making progress in targeted areas, without currently threatening American leadership.
American supremacy is reflected in capitalizations and investment capital flows.
– Nvidia has surpassed $5 trillion in market capitalization and controls most of the advanced accelerators (GPUs) used to train large language models.
– SpaceX has a market capitalization of more than one trillion dollars, OpenAI is valued at levels exceeding $1.5 trillion, while Anthropic has reached valuations close to 2 trillion.
– Microsoft and Alphabet maintain market capitalizations close to $4 trillion each.
These valuations reflect the ability of U.S. capital markets to funnel vast sums to technologically pioneering companies. Billion-dollar funding rounds enable the purchase of computing power, the attraction of researchers and the training of models on a scale not available in Europe.
In terms of capabilities, OpenAI and Anthropic's latest models are consistently at the cutting edge in complex reasoning, programming, and factorial use tasks.
The main developer of large language models in Europe is Mistral AI, which gives the user both the ability to perform reasoning tasks and the construction of generative intelligence tools. However, Mistral's AI models do not reach the level of OpenAI and Anthropic's leading models in the demanding tests. This difference is linked to the scale of the infrastructure between these three companies, which results from the investment funds made available to each company.
The American Advantage
The investment advantage of the United States is crucial. Venture capital markets and public markets are channelling capital into AI companies, in volumes far exceeding European ones. This allows not only the development of models but also the control of the infrastructure.
European companies are often forced to seek financing in the United States or buy computing power from American providers. Capital pooling enhances talent pooling, as higher salaries and greater potential for experimentation attract researchers.
In China, funding is largely state-driven. Companies such as Baidu, Alibaba and Huawei receive support for the development of models and accelerators, however the overall scale of private equity funds remains smaller compared to America. Also, access to the most advanced processors, which are marketed primarily by companies of American interests, is limited. The result is progress in specific applications, without equivalent control of absolute technological cutting-edge.
The European capital market
In Europe, economic constraints are more pronounced. Capital markets have failed in many cases to attract the same volume of investment as American ones, unless we are referring to companies such as LVMH and Estee Lauder.
Also, the bank-centric financing system makes it difficult to invest in fast-growing companies (hyperscalers). Mistral has raised significant funds at European levels, but these pale in comparison to the funds raised by OpenAI and Anthropic.
ASML (a Dutch company of interests) maintains a high market capitalization thanks to its position in the production of digital lithography machines, however this does not translate into control of the production of Artificial Intelligence models. In addition, part of this capitalization is due to the sale of the company's shares on both the Dutch stock exchange and the American Nasdaq.
Absence of domestic alternatives
The political dimension exacerbates economic differences. In the United States, the looser regulatory framework facilitates the rapid allocation of capital and experimental development. In contrast, in Europe, the European Artificial Intelligence Regulation incurs additional compliance costs, discouraging private investment in startups. The result is that investment funds are directed where the return and speed are highest.
The consequences for the European public and private sectors are already visible and tend to become more entrenched. Public administrations, health systems, educational institutions and critical infrastructure are increasingly dependent on models and platforms developed and controlled by American companies. In the private sector, European companies integrate AI solutions through contracts with providers that determine the terms of access, prices, usage limits, and results from that use.
The absence of domestic alternatives means that both the state and businesses operate as endless customers, who are even unable to judge the final result qualitatively. This limits the possibility to enforce European standards of transparency, data protection and accountability, as key design decisions are taken outside the Union.
The question that arises is whether Europe will find itself in a state of constant dependence on private companies such as Anthropic, Palantir and SpaceX. Anthropic controls large language models that can be used in administrative functions and technocratic decisions. Palantir provides data analytics platforms that strategize and outline forecasting and alternative future scenarios. SpaceX, in addition to space activities, represents an operational model that provides analysis and information solutions with the utilization of Artificial Intelligence.
The political dimension of dependency
The political dimension of this dependence is crucial: Europe's strategic autonomy is diminished when critical functions rely on technologies whose owners are subject to US jurisdiction and can be subject to export controls, political pressures and deep state lines.
The loss of the ability to produce and control this technology translates into reduced bargaining power, both vis-à-vis companies and the United States. In the long term, this means limiting Europe's ability to determine the conditions under which AI is used on its territory. Even more worrying, however, is the inability to evaluate and compare the deliverables of these models, since the Union will not be able to know whether the models deliver solutions, suggestions and strategic models, based on the well-meaning interest of the EU, or on the underlying interest of their manufacturer and the United States more broadly.
* (MBA), PhD Candidate in Artificial Intelligence and Educational Administration
