Opinion · By Trey Schmitt
At a glance
This essay argues that AI competition depends on physical infrastructure and state policy. Its future scenarios and probabilities are the author’s forecasts, not established outcomes.
Related reading: Bitcoin Farm to Data Center: Costs & Risks. Editorial standards and corrections.
I. The Frame I Am Working From
I do not believe the next world order is going to be decided by some clean, floating, frictionless version of artificial intelligence where the machine thinks and the world simply obeys, because that version of the future is too polished, too bloodless, too removed from the wires, grids, mines, ports, water systems, shipping lanes, server farms, permits, subsidies, sanctions, and desperate national ambitions that actually decide who gets power and who gets left renting the illusion of it.
What I see is much harder than that.
I see AI becoming the excuse, the weapon, and the accelerant for a new kind of mercantilism, not the old mercantilism with wooden ships and crowns and flags planted in foreign soil, but a modern version built through export controls, chip restrictions, sovereign AI programs, critical-mineral leverage, public subsidies, domestic-content rules, data localization, national cloud systems, energy policy, water permitting, and the slow but unmistakable realization that the countries that control compute will control more than software.
They will control speed.
They will control access.
They will control intelligence.
They will control industrial planning.
They will control military advantage.
They will control the transaction layer of modern life.
That is why I do not treat AI as an app economy. I treat it as infrastructure. I treat it as a power-grid issue, a semiconductor issue, a water issue, a mining issue, a data-sovereignty issue, a finance issue, and finally a sovereignty issue, because once intelligence becomes industrial, whoever owns the infrastructure underneath intelligence owns the future built on top of it.
So the basic chain is this:
AI demand rises → compute becomes strategic → chips become controlled → minerals become leverage → data centers expand → grids strain → water becomes political → governments subsidize domestic AI infrastructure → the world divides into infrastructure blocs. That is the thesis. Not model versus model. Not chatbot versus chatbot. Not one company winning the internet. The real fight is infrastructure bloc versus infrastructure bloc.
II. Compute Becomes the New Strategic Commodity
The first dynamic I believe will shape global events is that compute becomes the new strategic commodity, the thing nations pretend is commercial until they realize it is too important to leave exposed, the way oil, steel, shipping lanes, nuclear materials, and semiconductors all became too strategically important to be treated as ordinary goods once the stakes became obvious.
This is why my forecast that U.S. AI chip controls on China remain active through 2027, with an 84% probability, is one of the anchors of the whole framework. I do not see those controls going away because they are no longer just trade policy. They are containment policy. They are industrial policy. They are military policy. They are a way of saying: you may participate in the global economy, but not at the layer where frontier intelligence gets trained.
That is modern mercantilism in plain clothes.
The old empire controlled ports.
The modern state controls chips.
The old empire taxed trade routes.
The modern state licenses compute.
The old empire protected shipyards.
The modern state subsidizes fabs, data centers, cloud infrastructure, and national AI stacks.
And once the United States restricts advanced chips, China has no choice but to accelerate its own chip ecosystem, deepen control over upstream minerals, develop substitutes where it can, and use its own leverage where the West is vulnerable. Then the U.S., Europe, Japan, South Korea, India, and others respond by funding domestic capacity, friend-shoring supply chains, and treating AI infrastructure like a strategic industrial base instead of a normal market.
That is why my forecast that at least 3 G20 economies create or expand sovereign AI compute programs by 2029 matters. Once compute becomes power, countries will not want to rent all of it from somebody else. They will want national AI clouds, public AI supercomputers, domestic model programs, government-backed compute access, and procurement systems that favor trusted infrastructure.
The language will sound clean.
Innovation.
Security.
Competitiveness.
Digital sovereignty.
National resilience.
But underneath the clean language is something older and more brutal:
No serious state wants to depend on a rival for the machinery of intelligence.
III. Minerals Become the Pressure Point Beneath the Machine
The second dynamic is that minerals stop being treated as boring inputs and start being treated as leverage, because the AI economy may look digital from the user’s screen, but underneath that screen is a physical chain of copper, rare earths, lithium, gallium, germanium, graphite, nickel, cobalt, antimony, silicon, steel, aluminum, power electronics, transformers, chips, cooling systems, and industrial equipment that has to be mined, refined, processed, transported, financed, permitted, and protected before anyone ever types a prompt.
That is why my forecast that China maintains critical-mineral controls through 2027, with an 80% probability, fits directly into the thesis. China understands that the future is not won only at the final product stage. It is won in the middle of the chain, where raw materials become usable industrial power. Whoever controls processing controls pressure. Whoever controls export licensing controls timing. Whoever controls timing controls leverage.
And this is exactly where a lot of countries have historically lost the future.
They had the minerals.
They had the land.
They had the ports.
They had the workers.
They had the commodity.
But they did not have the processing, the refining, the financing, the industrial policy, the local ownership, the grid, the water system, the rail corridor, or the bargaining power to keep enough of the value before it left.
So when I forecast that at least 5 countries will target critical-mineral processing, not just mining, by 2029, I am really forecasting the spread of value-chain nationalism. Countries are waking up to the fact that it is not enough to dig up lithium if someone else makes the battery materials, not enough to export copper if someone else manufactures the grid equipment, not enough to mine rare earths if someone else controls refining, magnet production, and advanced manufacturing.
The countries that only extract will sell ingredients.
The countries that process will sell leverage.
That is the difference between having resources and having power.
IV. AI Turns the Grid Into the Battlefield
The third dynamic is electricity, and this may be the one that turns the AI debate from a technology story into a public-infrastructure fight.
AI sells intelligence, but it consumes power.
It needs data centers.
Data centers need electricity.
Electricity needs generation.
Generation needs fuel, renewables, nuclear, hydro, geothermal, storage, or gas.
Power has to move through transmission lines.
Transmission lines need permits.
Substations need transformers.
Transformers need copper and steel.
Grids need workers.
Workers need housing.
And the public eventually asks why their bills are rising so that somebody else’s data center can train somebody else’s model for somebody else’s profit.
That is why my forecast that at least 5 U.S. states will cite AI or data centers as major drivers of grid planning by 2028, with an 86% probability, is one of the strongest forecasts in the set. It is already happening in principle because AI demand is not evenly spread across the country; it clusters where land, tax incentives, fiber, power, and permitting line up, and when those clusters form, utilities have to respond.
That response will not be theoretical.
It will show up in integrated resource plans.
Transmission expansion.
New generation.
Rate cases.
Substation upgrades.
Local opposition.
Water debates.
Land-use fights.
Environmental review.
Nuclear restarts.
Gas backup.
Renewable procurement.
Battery storage.
And eventually the question that always matters in infrastructure:
Who pays, and who benefits?
That is also why I forecast that at least 3 OECD countries will approve or announce firm-power projects tied partly to AI or data-center demand by 2029. The AI sector wants clean energy rhetorically, but it needs reliable energy operationally, and there is a difference between energy that looks good in a press release and power that keeps a data center running every second of every day.
This is where the digital future gets dragged back into the physical world.
AI may be trained in data centers, but the political fight will happen at the grid.
V. Water Becomes the Local Veto
The fourth dynamic is water, and water is where the grand national strategy comes down to earth and has to face the people who live near the project.
A federal government may want AI dominance.
A governor may want investment.
A mayor may want jobs.
A utility may want a large industrial customer.
A tech company may want cheap land and power.
But if a community believes a data center is taking scarce water, stressing the grid, raising rates, and producing fewer jobs than promised, then the project becomes vulnerable no matter how futuristic it sounds.
That is why my forecast that at least 5 jurisdictions will regulate or enforce data-center water-use rules by 2028, with a 73% probability, matters beyond environmental policy. Water is the boundary condition of growth. It decides where cities can expand, where mines can operate, where farms can survive, where factories can open, and now where AI infrastructure can scale.
This is one of the cleanest intersections between my infrastructure thesis and AI mercantilism.
The national-security state says: build compute.
The market says: build data centers.
The grid says: find power.
The land says: find permits.
The community says: what about our water?
That question is not going away.
It will matter in the U.S. West, Mexico, Chile, Spain, the Middle East, North Africa, India, Australia, and any place where data-center growth collides with drought, agriculture, urban demand, or public distrust.
The AI companies that understand this will invest in water reuse, closed-loop systems, better cooling, site discipline, and local benefit agreements.
The ones that do not will become symbols of extraction.
And once a technology becomes a symbol of extraction, the politics changes.
VI. Subsidies Turn AI Into Bloc Competition
The fifth dynamic is subsidy, because once AI infrastructure becomes strategic, governments will not leave it to the market. They will subsidize it, protect it, regulate it, procure it, and try to localize enough of it to avoid dependence on rivals.
That is why my forecast that at least 5 countries outside the U.S. and China will subsidize AI infrastructure by 2029, with an 84% probability, is so central. This is not about one country copying another country’s tech policy. This is about the global spread of infrastructure mercantilism.
The European Union wants AI gigafactories because it does not want to be permanently dependent on American cloud companies or Chinese hardware chains.
The Gulf wants AI infrastructure because sovereign wealth is trying to buy post-oil relevance.
India wants AI infrastructure because population scale, digital public infrastructure, and national ambition all point toward sovereign compute.
Japan and South Korea want AI infrastructure because advanced industry and security depend on it.
Smaller countries will want pieces of the stack: data centers, cloud regions, public compute, AI research clusters, language models, or digital-government infrastructure.
This is why my EU forecasts matter too. I think at least 1 EU-backed AI gigafactory becomes operational by 2030, but I also think the EU likely misses its 20% semiconductor production ambition by 2030. That combination is important because it shows the difference between policy direction and execution capacity.
Modern mercantilism can announce the target.
It can subsidize the target.
It can protect the target.
But it still has to build the grid, find the workers, permit the facilities, secure the materials, finance the fabs, manage water, and compete with places that already have scale.
The state is back.
But the state still has to execute.
VII. My Broader View: Infrastructure Blocs Replace Naïve Globalization
When I put these forecasts together, I do not see the end of globalization. I see the end of the innocent version of globalization, the version that believed everything strategic could be outsourced forever as long as the price was good and the shipping lane stayed open.
That world is fading.
The new question is not only:
Where is it cheapest?
The new questions are:
Where is it trusted?
Where is it powered?
Where is it cooled?
Where is it processed?
Where is it legally protected?
Where is the data stored?
Where are the minerals refined?
Where are the chips made?
Where can the grid survive the load?
Where can the public tolerate the project?
Where does the value stay?
This is what I mean by infrastructure blocs.
The U.S. and its allies will try to control advanced chips, trusted cloud, semiconductor equipment, and secure AI supply chains.
China will try to build around those controls while using its strengths in processing, manufacturing, minerals, and state-directed industrial capacity.
Europe will try to build sovereign AI capacity while struggling with energy costs, regulation, and execution timelines.
The Gulf will use capital and energy to buy a seat at the AI infrastructure table.
India will try to build a sovereign AI and digital-public-infrastructure model at population scale.
Resource-rich regions like Latin America, Africa, Australia, Canada, Indonesia, and Central Asia will gain leverage if they move beyond extraction into processing and infrastructure ownership.
But the winners will not be the countries that simply have one piece of the stack.
The winners will be the places that can stack the stack:
power + water + chips + minerals + data centers + law + capital + workforce + public legitimacy.
That is the real map.
VIII. The Cause-and-Effect Map
| Cause | Effect | World Impact |
| AI demand rises | Data centers expand | Power and water become AI bottlenecks |
| Compute becomes strategic | Export controls remain | AI trade becomes national-security trade |
| Chip controls tighten | China accelerates alternatives | Tech blocs harden |
| Mineral controls spread | Processing becomes policy target | Resource-rich regions gain leverage |
| Data centers stress grids | Utilities seek new generation/transmission | AI becomes public-infrastructure politics |
| Water conflicts grow | Data-center permitting tightens | Local communities gain veto power |
| Governments fear dependence | Sovereign AI programs expand | AI infrastructure becomes industrial policy |
| Subsidies expand | AI blocs form | Modern mercantilism becomes global |
IX. The Bottom Line
My bottom line is simple, and I think it is where this whole thesis lands:
AI will not just change what the world can think. It will change what the world has to build.
The countries that win will not only have the best models.
They will have the power to run them.
The chips to train them.
The minerals to manufacture the chips.
The water to cool the data centers.
The grids to carry the load.
The ports to move the equipment.
The laws to control the data.
The capital to finance the buildout.
The workers to maintain the systems.
And the ownership discipline to keep the value from leaving.
That is why modern mercantilism and AI belong in the same conversation. AI creates the new strategic demand. Modern mercantilism creates the policy response. Infrastructure becomes the battlefield underneath both.
The old world fought over oil, ports, rail, steel, and shipping lanes.
The new world will fight over compute, chips, power, water, minerals, data, and the corridors that connect them.
And that is why I do not see AI as some clean miracle hovering over the world.
I see it as a machine that will reveal the truth of every nation beneath it.
Who has power?
Who has water?
Who has minerals?
Who has chips?
Who has data?
Who has law?
Who has capital?
Who has the courage to build?
Who has the discipline to own?
Because the future will not belong to whoever says “AI” the loudest.
It will belong to whoever can build the infrastructure underneath intelligence and keep enough of the future after the machine turns on.