The Next Superpower Won't Be a Country

Most conversations about AI and global power focus on which country will build the most powerful model. America? Or China? Perhaps the European Union can move fast enough?

That is the wrong race to watch.

The countries that will define the next century of geopolitical power are not necessarily the ones building the most sophisticated AI. They are the ones that become indispensable to everyone else's AI. There is a significant difference between the two, and most national AI strategies are oriented toward the first, whereas the second is where durable power actually accumulates.

AI Power Comes From Dependency, Not Innovation Alone

To understand what is happening, it helps to think about how resource-based power has worked historically.

Oil did not give countries power because they refined the best petroleum products. It gave them power because the entire global economy depended on continuous access to their resource. You could not run a factory, move goods, heat a building, or project military force without it. Dependency created leverage. Leverage created power.

AI infrastructure works on the same logic, at a greater scale and with greater complexity. The dependency is not on a single resource but on an interlocking set of layers, each of which creates its own form of leverage.

The Real AI Race Has Three Layers

When I work with governments and organizations thinking about AI strategy, I use a framework that separates the competition into three distinct layers. Most public discourse is focused on one of them. The real battle is being fought in the other two.

1. Compute: The Foundation of AI Power

Compute means chips, cloud infrastructure, and data centers. It is the physical foundation of everything AI does. Without access to sufficient compute, no organization and no country can train or run competitive AI systems at scale.

The United States currently controls this layer more than any other nation, primarily through semiconductor design. The Netherlands controls a critical chokepoint through ASML, the only company capable of producing the extreme ultraviolet lithography machines required to manufacture leading-edge chips. Taiwan manufactures the majority of the world's advanced semiconductors.

These are not software advantages. They are physical, geographic, and industrial advantages that take decades to replicate.

2. Data: The Hardest Advantage to Replicate

Proprietary datasets are the second layer. Models improve with scale, but they improve faster with the right data. Countries and organizations sitting on large, high-quality, domain-specific datasets, like medical records, financial transactions, industrial sensor data, and legal documents, hold an advantage that cannot simply be trained away by a competitor with more compute.

Data advantages are also self-reinforcing. The more a system is used, the more data it generates, the better it becomes, and the more it is used.

3. Models: The Most Visible but Least Defensible Layer

Foundation models and domain-specific models are the layer everyone talks about. GPT. Gemini. Claude. Llama. The public competition in AI is almost entirely narrated at this layer.

Models matter. But they are also the most contestable layer. A sufficiently resourced competitor can train a competitive model. They cannot as easily replicate semiconductor manufacturing infrastructure or accumulate decades of proprietary data.

The Biggest Winners May Never Be Consumer Brands

Think about how the smartphone era played out.

The names most people remember are Apple and Samsung. Those were the visible winners, the companies whose products consumers held in their hands. The narrative of the smartphone revolution is largely told through them.

What most people do not remember is the layer underneath.

• TSMC manufactured the chips that made the devices possible.
• Qualcomm supplied the modems and processors that connected them.
• ARM designed the chip architectures that nearly every mobile device runs on.
• Telecom networks built and operated the infrastructure without which the devices would have been useless.
• Cloud providers stored data, ran applications, and processed requests.

Those players did not win the consumer narrative. They captured enormous, durable economic value. Some of them became more strategically important than the device makers themselves. ARM was acquired for $40 billion. TSMC became a matter of explicit national security concern for multiple governments simultaneously.

The AI era is following the same pattern. The companies and countries building the infrastructure layer are accumulating the kind of power that does not show up in product announcements but does show up in trade policy, export controls, and geopolitical leverage.

Four Assets Will Determine AI Leadership

Based on where the infrastructure dependencies are forming, the nations that will hold the most durable AI power are those that control:

• Energy. AI data centers consume enormous amounts of power. Countries with abundant, reliable, low-cost energy have a structural advantage in hosting the compute infrastructure the world needs.
• Compute infrastructure. Semiconductor supply chains, chip design capabilities, and data center capacity. The export controls the United States has imposed on advanced chips are an explicit acknowledgment that compute is a strategic resource.
• Semiconductor supply chains. Manufacturing, materials, equipment. The full stack, which is currently distributed across a surprisingly small number of countries.
• Data and data governance. The ability to collect, store, process, and govern large-scale data. Countries setting the rules for data governance are also setting the terms on which AI can be built and deployed within their borders and increasingly, for companies operating globally.

Many Countries Are Sleepwalking Into AI Dependency

A handful of countries are currently competing for AI dominance. The United States and China are the primary competitors. A small number of others, the United Kingdom, Canada, France, the UAE, South Korea, Japan, are positioning themselves as meaningful players in specific layers.

The rest of the world is largely watching. That is a strategic error.

Countries that do not ask hard questions about their position in the AI infrastructure stack are not standing still. They are drifting toward dependency.

An AI colony is not a country that lacks powerful models. It is a country whose critical systems, such as healthcare infrastructure, financial systems, government services, and communications networks, run on AI built, owned, and controlled elsewhere.

A country that generates valuable data but captures little of the economic value that data produces. A country whose AI policy is effectively set by the terms of service of foreign platforms.

This is not a hypothetical future risk. The infrastructure dependencies are forming now. The data relationships are being established now. The governance frameworks that will shape who controls what are being written now.

A National AI Strategy Starts With Owning One Layer

Winning at AI infrastructure requires identifying which layer of the stack offers a realistic path to relevance and moving deliberately toward it.

• A country with abundant renewable energy can position itself as a compute host.
• A country with deep domain expertise in healthcare or finance can build proprietary data assets that make it indispensable to global AI development in those sectors.
• A country with strong engineering talent can contribute to the model layer or to the tooling and infrastructure that surrounds it.

The countries that will matter in twenty years are not necessarily the ones that win the model race today. They are the ones that figure out which layer they can own, and build toward it with the same seriousness that previous generations brought to energy infrastructure, maritime routes, and industrial capacity.

The window to decide which side of that equation you want to be on is open now. It will not stay open indefinitely.

← Back to all articles