Jalil Alizadeh
Geopolitics is now measured in nanometers. At customs checkpoints, governments are no longer searching only for weapons or military components. Increasingly, they are trying to control something more subtle and more consequential: the speed at which a country can think.
The shipment delayed at the border may not be a glamorous industrial machine, but a chip small enough to fit on a fingertip—yet powerful enough to shape a nation’s rate of innovation, redirect tens of billions of dollars of investment, and alter the balance of power between rival blocs.
This is not merely a technology story. It is a story about growth, power and the changing architecture of the global economy.
For years, the dominant idea was that software would eat the world. That was broadly true. But the new reality is harsher: hardware can now hold the world hostage. Not because code matters less, but because even the best software is of limited use without enough machine capacity behind it. A brilliant model without access to advanced compute is like a perfect formula on a blackboard—elegant, correct and economically useless.
Compute as a Factor of Production
Advanced compute has become more than a tool. It is now a factor of production.
Artificial intelligence matters economically not only because it automates tasks, but because it compresses learning cycles. Firms that can train models faster, test them faster and iterate faster convert capital into useful knowledge more efficiently than their rivals.
Over time, that affects productivity, margins, exports and national competitiveness. In that sense, compute is no longer just an input into the technology sector. It is becoming an input into growth itself.
That helps explain why governments have become so interested in it. States usually intervene aggressively when three conditions exist at once: large commercial upside, serious security implications and a handful of controllable choke points. Advanced compute now has all three.
It is commercially transformative. It is also unmistakably dual-use. The same infrastructure that trains large language models can support cyber operations, intelligence analysis, sensitive simulations and military optimization. In export-control politics, therefore, the issue is no longer just the movement of goods. It is the distribution of capacity—who can access it, where it can be deployed and for what end use.
The State in the Server Room
There is another reason compute has moved from the realm of markets into the realm of statecraft: it is physical.
Compute is not produced in the abstract. It is built on electricity, cooling, land, water, bandwidth and permits. The International Energy Agency estimates that global data-center electricity demand could approach 945 terawatt-hours by 2030 in its baseline scenario—roughly double today’s level. That is the power consumption of a large industrial economy. A technology that is simultaneously growth-enhancing, security-sensitive and power-hungry was never going to remain the private concern of firms for long.
This is the essence of a new kind of industrial policy: computational protectionism. Unlike old-style protectionism, which relied on tariffs for steel or quotas for cars, this new version classifies and engineers access to advanced compute. It does not always seek to shut off the engine. More often, it seeks to regulate the engine’s speed.
That is why the world is beginning to resemble a tiered system of access: some countries enjoy broad and reliable access to compute, some face conditional access, and others face restriction or exclusion. In the 20th century, borders were drawn on maps. In the 21st, they are increasingly drawn around processing power.
Power Lies in Clusters, Not Chips
There is also an engineering reality that much economic commentary misses: power in AI is not created by a single chip. It is created by a cluster.
The real advantage enjoyed by the largest firms and most capable states is not ownership of a few high-end accelerators. It is possession of stable compute clusters: dense racks, high-speed interconnects, fast storage, liquid cooling, reliable grid access and the operational expertise to keep the whole system running at scale. In advanced rack-level systems, a single rack may consume around 120 kilowatts of power. At that point, a data center ceases to be a mere technology project. It becomes industrial infrastructure.
This is why the most politically sensitive bottlenecks are not always the chips themselves. They include chipmaking equipment, cluster-scale deployment, grid connection, water access, land, cooling systems and regulatory approval. These are the places where policy has the highest leverage, because they are the places where computation meets physics.
Innovation Will Be Redirected
The usual question is whether tighter controls will slow innovation. That is too crude. The better question is: where will innovation go instead?
History suggests that controls rarely stop technological development outright. They more often redirect it. Restrictions create incentives for efficiency, substitution and evasion. When access to frontier compute is limited, firms respond by building smaller and more efficient models, redesigning supply chains, pushing harder on advanced packaging and memory, and shifting investment toward places with cheaper electricity and more stable policy.
In other words, controls do not simply reduce innovation. They reallocate it. One path becomes slower; another becomes more valuable.
This matters for investors as much as for governments. In the emerging compute economy, risk is no longer defined only by interest rates, recession probabilities or consumer demand. It is increasingly defined by access: access to chips, to power, to permits, to trusted suppliers and to the legal right to scale. Capital, accordingly, is moving toward what might be called the physics of innovation—the energy, infrastructure and industrial systems that make computation possible.
The Silicon Curtain
That is the central tension of this era. Markets want scale. States want to classify scale.
The “silicon curtain” is therefore more than a metaphor. It is a new allocation regime—one that determines who can turn thought into products, products into profits and profits into power. In the last century, nations competed over who had more: more oil, more steel, more ships, more factories. In this one, the contest is increasingly over who can learn faster.
And in that contest, the loser does not simply fall behind. The loser risks becoming invisible.
