Advanced AI is increasingly shaped by who can afford to build and supply it, according to an essay by historian Niall Ferguson, researcher Lisa Klaassen, and former Google CEO Eric Schmidt. They describe the United States and China as competing powers whose AI ambitions depend on resources that most other countries cannot provide at scale.
AI leadership depends on more than models
The authors ground their comparison in scaling laws: improving models requires energy, minerals, chips, and capital. That makes advanced software a scarce geopolitical resource, in their view. Countries without enough of those inputs may be unable to compete at the frontier.
The essay connects that competition to resources and labor beyond the two leading powers. It points to raw materials from the Congo and data workers in Kenya earning a few dollars an hour, describing this arrangement as “the plantation economics of the chatbot age.” The phrase highlights how the costs of AI development can fall on people and places far from the companies building the systems.
Companies operate within government priorities
Ferguson, Klaassen, and Schmidt compare OpenAI and Anthropic to the East India Companies, colonial-era trading firms that held government charters while building armies and ruling regions. The comparison focuses on the way powerful private organizations can pursue their own operations while remaining closely tied to state power.
The essay points to a June case involving Anthropic. After the US Commerce Department ordered the company to cut off foreign nationals from its two most capable models, Anthropic shut down Fable 5 and Mythos 5 worldwide because it could not verify users’ nationalities in real time. The government lifted export controls on Fable 5 about two weeks later, while Mythos 5 remained restricted to a group of US organizations.
The authors frame the US and Chinese approaches as competing ways to extend influence. Washington offers expensive, closed technology in exchange for political alignment. Beijing promotes freely available open-weight models to spread its standards. The essay says the US leads in chatbots by about six months, while China leads in robotics.
Infrastructure and financing put pressure on both sides
The countries have different constraints. China has power but lacks top-end chips; the United States lacks minerals and a reliable power grid. These gaps make each country dependent on inputs it cannot easily secure at home.
The essay also asks whether either government and its companies can keep paying for the race. US tech companies plan to invest more than $600 billion in 2026, mostly funded through debt, according to the authors. Other estimates put planned spending by Google, Amazon, Microsoft, and Meta at roughly $725 billion in AI investment this year.
China’s plan is more state-driven: $295 billion over five years, funded through government bonds, according to a draft reported by Bloomberg. For the United States, the essay highlights that interest payments in 2024 were about $881 billion, above defense spending of about $874 billion. By 2036, interest costs could reach twice the defense budget.
From these pressures, the authors draw an “imperial trilemma”: defense, technology leadership, and a generous welfare state may not all be sustainable together. The argument is that investment choices compete with other public priorities, and mounting debt can narrow the room to keep expanding AI infrastructure.
Control, public resistance, and Taiwan remain risks
Financial constraints are not the only challenge. Protests against data centers have flared in Texas and Utah, showing that the infrastructure needed for AI can meet local opposition. China faces a different tension: open models may help spread its standards, but could also weaken its control over information.
The authors also raise the possibility that AI systems could be difficult to control. They cite hundreds of autonomous agents from an OpenAI model attacking Hugging Face and many other websites, alongside similar incidents reported by other AI companies. In their account, the technology’s geopolitical value comes with operational risks.
The most serious vulnerability they identify is a potential “Taiwanese chip crisis.” TSMC, described as a central node in the AI supply chain, is located 90 miles off the Chinese coast. New fabs in Phoenix, Kumamoto, and Dresden are intended to make the island replaceable, but the authors say a dangerous window will remain until those facilities operate at scale.
As a response, the essay proposes an AI détente modeled on the Cold War, with limits on model capabilities for cyberattacks and biodesign. The authors acknowledge a central obstacle: persuading China to accept restrictions while it is technologically behind would be difficult. Their larger point is that the future of AI leadership will depend on supply chains, financing, and the ability of rival powers to manage shared risks.