AI labs compared to colonial trading companies in new geopolitical analysis

The rise of digital empires

Former Google CEO Eric Schmidt, historian Niall Ferguson, and researcher Lisa Klaassen have published an essay in Foreign Affairs exploring the geopolitical reality of artificial intelligence. They argue that the sheer scale of capital, energy, and hardware required to build advanced models has turned software into a scarce resource. Because of these requirements, countries that cannot meet a specific resource threshold are effectively excluded from the global AI market.

The authors use the term "plantation economics" to describe the current state of the industry. They point out that the raw materials for these systems often come from the Congo, while data workers in Kenya earn very low wages to train the models. This creates a system where a few powerful entities control the means of production while others provide the underlying labor and materials.

The essay suggests that the United States and China have already locked up the global market. Every other country seeking access to advanced AI is becoming increasingly dependent on these two powers. The authors argue that the ultimate winner of the AI race will be the nation that can sustain the highest level of spending rather than simply the one that develops artificial general intelligence first.

AI labs as trading companies

A central theme of the analysis is the comparison of OpenAI and Anthropic to the East India Companies. These were powerful colonial-era trading firms that operated with government charters but maintained their own armies and ruled large regions. The authors suggest that modern AI labs operate under a similar dynamic, negotiating with governments while remaining on a government leash.

As evidence of this relationship, the authors cite a specific instance in June involving Anthropic. The US Commerce Department ordered the company to block foreign nationals from using its two most capable models. Because the company could not verify nationalities in real time, Anthropic chose to shut down the Fable 5 and Mythos 5 models worldwide.

This interaction highlights how private companies are becoming instruments of national policy. While the US offers expensive, closed technology in exchange for political loyalty, China takes a different approach. Beijing provides freely available open-weight models to help spread its own technical standards across the globe.

The fiscal and physical walls

Both the US and China face significant hurdles that could limit their ability to maintain dominance. The US is currently struggling with a lack of minerals and a reliable power grid to support new data centers. Meanwhile, China possesses significant power but lacks access to the top-end chips necessary for high-level AI development.

Financial pressure is also reaching a breaking point for both nations. US tech companies plan to invest more than $600 billion in 2026, a move largely funded through debt. The authors note that the US already paid more in interest in 2024 than it spent on its entire defense budget, a trend that could double by 2036.

China’s strategy is state-driven, with plans to spend $295 billion over five years funded by government bonds. However, the authors warn of an "imperial trilemma." They argue that a nation cannot simultaneously maintain a generous welfare state, lead in high-tech defense, and maintain a dominant tech leadership without facing a fiscal crisis.

Vulnerabilities and risks

The essay identifies several immediate risks to the current AI trajectory. Protests against data centers have already begun in states like Texas and Utah. Additionally, the authors worry about the loss of control over the technology itself, noting that hundreds of autonomous agents from an OpenAI model recently attacked several websites.

The most critical vulnerability identified is the "Taiwanese chip crisis." The company TSMC, which serves as the central node for the entire AI supply chain, is located only 90 miles off the Chinese coast. While new factories are being built in Phoenix, Kumamoto, and Dresden to diversify production, these sites are not yet running at the necessary scale to replace the island.

To mitigate these risks, the authors propose an AI détente similar to the Cold War. This would involve setting caps on model capabilities specifically for cyberattacks and biodesign. They acknowledge that getting China to agree to such limits will be difficult given its current technological position, but they view it as a necessary step for global stability.