Every conversation about AI seems to revolve around one question: will intelligence become abundant?
I think we are asking the wrong question. The bigger risk isn't that intelligence becomes a commodity. It is that frontier intelligence becomes concentrated in the hands of a handful of companies and countries.
Many people already think of AGI as a monster, one that could eventually outsmart governments, corporations, and perhaps even humanity itself. If that is your mental model, history offers a useful lesson. If you want to control a monster, don't build a bigger weapon. Own the food it needs to survive.
Civilisations have rarely controlled powerful actors by competing with them head-on. More often, they controlled the strategic resources those actors depended upon. AI should be viewed through the same lens.
Intelligence has a metabolism, and it is physical
A frontier model cannot survive on code alone. It needs electricity. It needs semiconductor chips, and those chips need fabrication plants, which in turn need land, water, critical minerals and billions of dollars of capital. AI needs massive data centres and global internet connectivity. It needs high-quality data, not just to train foundation models but to keep improving, specialising and adapting to the real world. And it needs engineering talent capable of building and maintaining the entire stack.
Intelligence may become digital. Its metabolism will always remain physical, and increasingly informational.
Every technological revolution has had a survival stack like this. The steam engine depended on coal. The internet depended on fiber-optic networks. AI depends on energy, chips, land, connectivity, data, capital, and talent. That survival stack may ultimately matter more than intelligence itself.
Why owning distribution isn't enough
Today, every country wants to build the next frontier AI model. Some will succeed. Most won't. That is simply how markets work. But if only a handful of organisations end up owning frontier intelligence, society's bargaining power shouldn't disappear along with them. It should come instead from owning the infrastructure that intelligence depends on.
This is where I differ from an increasingly popular school of thought. In India, several voices, including Nandan Nilekani, have argued that the country's greatest opportunity may not lie in building frontier models at all, but in controlling their diffusion through India's digital public infrastructure and its enormous consumer market.
There is real truth in this. Distribution matters. Markets matter. Being the first place where a new technology scales creates influence. But I believe that kind of advantage is inherently temporary. Markets are mobile. If one country becomes difficult to operate in, businesses eventually find another. The world is simply too large, and there is always another emerging economy eager to attract investment. Today it may be India. Tomorrow it could be Indonesia, Nigeria, or another market with hundreds of millions of future users. Demand eventually finds alternatives.
Strategic infrastructure doesn't. A semiconductor ecosystem cannot be relocated overnight. Neither can a 10 GW power grid, a network of hyperscale data centres, decades of engineering talent, industrial land or advanced manufacturing capability. These assets take years, often decades, to build. Markets provide negotiating power for a while. Infrastructure provides negotiating power for generations, and that is a far more durable source of strategic influence.
Two lessons from history
History has taught this lesson before, in two very different ways.
The Industrial Revolution wasn't powered by brilliant ideas alone. It depended on coal, steel, iron, ports, railways, and factories. Countries that controlled these resources accumulated extraordinary wealth and geopolitical influence. Many resource-rich societies, by contrast, became suppliers rather than partners. Their resources were extracted while the real value creation happened elsewhere, often because they lacked the institutional, financial, or military capacity to convert resource ownership into bargaining power. The lesson wasn't simply that resources matter. It was that understanding the strategic value of a bottleneck before everyone else determines who shapes the next era.
The IT revolution tells the opposite story. That time, the scarce resource wasn't coal or steel. It was people: millions of engineers, software developers, mathematicians, and English-speaking technical graduates. India happened to invest in precisely that resource. When global demand for software exploded, India wasn't merely a customer; it became an indispensable supplier to the global economy. That single strategic advantage transformed the country's economic trajectory for decades.
Every technological revolution creates a different bottleneck. Those who own the bottleneck gain leverage.
AI's bottlenecks are unglamorous, and that's the point
AI is creating its own set of bottlenecks: energy, semiconductor manufacturing, data centres, industrial land, power transmission, internet connectivity, critical minerals, data, and engineering talent. None of these will dominate headlines the way a new frontier model does. But they could become the real sources of geopolitical power in an AI-driven world.
This is not an argument against building frontier models. Countries should absolutely compete, and innovation thrives through competition. Some will produce world-class AI companies; others will not, and that is healthy. The mistake would be allowing the same small group of players to own both the intelligence and everything that intelligence needs to survive. Once a handful of players control the monster and its food supply, everyone else loses meaningful negotiating power.
What governments should actually be asking
So how should the world prepare? Not by assuming every country will build the next frontier model, because that is unrealistic. Instead, governments should ask a different question: Which parts of AI's survival stack can we own, strengthen, or diversify?
That means investing in:
- Abundant and reliable energy
- Semiconductor manufacturing and packaging
- Data centres and compute infrastructure
- Industrial land suitable for AI infrastructure
- High-speed internet connectivity
- Secure, high-quality data ecosystems
- Critical mineral supply chains
- Engineering education and technical talent
The objective should not be equal ownership of intelligence. It should be distributed ownership of the infrastructure intelligence depends upon.
A missed opportunity, and a lesson from food security
This point may sound obvious, but I don't think we have fully put it in perspective. India already produces solar electricity roughly ten times cheaper than many parts of the world, yet it has never branded itself as a low-cost destination for data centre power. That is partly understandable, since we are still working to meet our own people's demand. But selling electricity to data centres at a premium could help finance the construction of more power capacity within the country.
We made a similar choice with food, and I think there is a lesson in it. Our focus was always food security for the nation rather than export, and we did meet that goal. But I sometimes wonder what would have happened if we had also focused on export-quality produce. Would our farmers have been more prosperous without compromising our food security? I suspect there was a middle ground we never fully explored, and I think we are at risk of making the same choice again with energy.
Distributed power, not perfectly equal power
Civilisations rarely remain stable because power is perfectly equal. They remain stable because power is distributed. Finance depends on energy. Manufacturing depends on trade. Technology depends on infrastructure. No single actor controls every layer, and that mutual dependence creates negotiation instead of domination. The AI era should aspire to the same balance.
If intelligence eventually becomes the monster many people fear, then perhaps we are focusing on the wrong battle. Instead of spending all our energy trying to build a bigger monster, we should make sure the monster never owns all of its own food.
Because in every era of history, those who controlled the bottlenecks ultimately shaped the future. The AI age will be no different.
