Who Will Control AI The New Global Energy Race

Have you noticed how quickly the argument about artificial intelligence stopped being a tech debate and started sounding like an energy debate? That is not an accident. Every major leap in human power has followed the same pattern. First people argue about the tool. Then they argue about who owns the fuel. Then they discover, a little late, that the fuel was the real prize all along.

The Quiet Contest Over Mental Energy

I keep coming back to a simple thought. Muscle built the first empires. Coal and oil built the industrial ones. Electricity stitched cities together. Nuclear raised the ceiling again. Now a different kind of power is coming online, and it does not look like a turbine or a refinery. It looks like racks of servers, cooling plants, specialized chips, and models that can multiply human judgment at a scale no office full of analysts could match.

Call it mental energy if you want a plain phrase. Computers already amplified calculation. Artificial intelligence goes further. It can draft, sort, simulate, translate, design, watch, and recommend. Used well, it can stretch what a skilled person can do in a day. Used poorly, it can magnify surveillance, coercion, and strategic pressure. That dual use is why the leadership question is not academic. It is the whole story.

Some people in the United States want to slow the buildout. Some of their worries are honest. Towns ask what a data center will do to local power bills, water use, and traffic. Workers ask whether automation will eat the jobs that used to feel safe. Parents ask what happens when machines write faster than students can think. Those questions deserve answers, not eye rolling.

But there is another current running under the same conversation. If America ties itself in knots while a rival state treats AI as a national project, leadership will not pause out of courtesy. It will move. And the government most eager to see that happen is not mysterious. Beijing has made the ambition obvious for years.

Why Energy History Still Matters

Human development can be told as a long story about borrowed strength. First we used our own bodies. Then we borrowed the strength of animals. Then we borrowed wind and water. Steam arrived. The internal combustion engine arrived. Grids arrived. Fission arrived. Renewables arrived as another chapter, not the last page.

Each shift did two things at once. It raised output, and it rearranged power among nations that could produce, refine, transport, and protect the new source. Oil was never only a commodity. It was leverage. The same will be true of compute. The difference is speed. A pipeline takes years. A model can be copied, distilled, and deployed much faster once the underlying stack exists.

Writing and the printing press expanded memory and argument. They were mental tools. They did not, however, sit beside a person and generate options in real time. That is the jump people keep underestimating. A calculator did not change statecraft. A system that can scan open source data, draft policy options, optimize logistics, and target influence campaigns is another animal.

The country that treats AI as a convenience will be managed by the country that treats AI as infrastructure.

In my experience, the public argument still sounds like a gadget review. Is the chatbot clever? Is the image generator uncanny? Those are sideshows. The main event is capacity: chips, power, talent, data, and the legal room to build. Lose those, and the applications will still exist. They will just be designed elsewhere, under someone else’s rules.

Local Pushback Meets A Global Clock

Data centers are ugly neighbors if you live next to the fence line. They hum. They draw power. They can strain water systems in dry regions. They change the look of a landscape that used to be fields or quiet industrial parks. I get why a county board flinches. Nobody wants to be told that national strategy requires a warehouse the size of several football fields behind their house.

Still, the honest local ledger is not only cost. It is also payroll, tax base, construction work, grid upgrades, and a reason for engineers to stay in places that have been losing young people. Some state leaders have figured this out and started pitching responsible sites instead of hoping the demand evaporates. Demand will not evaporate. It will relocate.

That is the part the slowdown camp underplays. A paused project in one American county does not freeze the technology. It shifts the map. Capital is impatient. Model trainers are impatient. If the permits take too long here, the next cluster stands up somewhere with cheaper power, faster approvals, and fewer hearings.

• Communities worry about electricity prices and land use, which is fair.
• Investors worry about timelines, which is also fair.
• Rivals worry about neither. They worry about catching the frontier and locking in standards.

Perhaps the most interesting aspect is how quickly a local zoning fight can be dressed up as a moral crusade. Some of that is sincere environmental concern. Some of it is ordinary nimby politics. And some of it, frankly, is useful to anyone who would prefer that American capacity stay fragmented while another system plans in five year blocks.

The Rare Earth Warning We Already Lived Through

We have seen this movie. The West allowed a near monopoly to form in rare earth mining, processing, and a thick slice of the manufacturing that depends on those materials. The arrangement looked efficient on a spreadsheet. It looked less charming when supply became a political instrument.

Prices jumped around. Adoption of products that needed those inputs got bumpier. Environmental costs did not vanish. They moved to a jurisdiction with weaker standards and then came back as strategic risk. That is the pattern. Offload the messy stage. Celebrate the cheap finished good. Wake up when the supplier notices the dependency.

AI sits on a similar stack. Specialized semiconductors. Critical minerals. Power. Cooling. Talent pipelines. Cloud regions. If one state dominates enough of that stack, it does not need to win every software bake off. It only needs to make other countries hesitate.

I’ve found that people accept this logic instantly when the topic is oil or shipping lanes. They get fuzzy when the topic is models and data halls. The fuzziness is dangerous. Compute is becoming a strategic commodity. Treat it like a lifestyle app and you will negotiate later from a weaker chair.

Oversight Records Are Not A Side Issue

Trust is not a vibe. It is a track record. China’s scientific and industrial oversight has already shown cracks with global consequences. Laboratory safety debates around earlier outbreaks were not invented last week. Delayed information has a cost when a pathogen is involved. An earlier SARS escape in the mid 2000s is part of the public record. None of that proves a specific claim about any one lab in isolation. It does prove that opaque systems handle accidents badly.

Now raise the stakes from a virus to a general purpose technology that can be pointed at speech, finance, logistics, weapons targeting support, and population control. If the same political structure that prizes narrative control also prizes AI leadership, outsiders should not pretend the two goals will stay neatly separated.

Environmental standards tell a related story. Looser rules can look like a production advantage until the externalities show up in air, water, and worker health. Export the dirty stage, import the cheap component, then act surprised when leverage appears in the relationship. AI will not be cleaner by slogan. Training and inference burn electricity. Cooling burns water or power, depending on design. The question is whether those costs are managed in the open or hidden behind state secrecy.

How A One Party State Would Use The Tool

We do not have to guess in the abstract. The same government already uses digital tools to watch its own people at a density most open societies would reject. Cameras, scoring systems, keyword flags, and now model assisted sorting are not science fiction there. They are administration.

External behavior matches the internal habit. Border pressure with India has been recurring. Artificial islands and confrontations in contested waters have been a feature, not a glitch. The Hong Kong arrangement was rewritten in practice. Taiwan lives under a drumbeat of warnings. Japan has heard its share of sharp language. Investment programs across poorer countries look generous until the debt and political alignment come due.

There are also growing claims about money finding its way into protest scenes abroad. Some of those claims will be overstated. Some will not. The point is simpler. A state that already mixes commerce, intelligence, and political pressure will not suddenly treat a more powerful analytic engine as a neutral utility.

1. At home, AI becomes a sharper instrument of social management.
2. Abroad, it becomes a cheaper way to map elites, supply chains, and public mood.
3. In gray zones, it becomes a way to harass, flood, and confuse without firing a shot.

I am not saying every Chinese engineer is a villain. That would be lazy. Plenty of researchers want to publish, ship products, and live ordinary lives. The issue is the political machine that can conscript the stack when it wants to. In an open system, companies, courts, journalists, and voters still fight over limits. In a closed system, the limit is whatever the party needs this quarter.

Internal Control Is The Preview

Foreign policy headlines grab attention. Domestic practice is the better forecast. Laws framed as ethnic unity sit on top of a longer record of harsh control in Xinjiang and Tibet. You can argue about labels. You cannot argue that the state has been shy about using technology against groups it views as problems.

If a government already treats speech, movement, worship, and kinship as administrative variables, a stronger model is not a philosophical puzzle. It is an upgrade. Better facial matching. Better translation of intercepted chatter. Better prediction of who might organize. Better generation of official narratives. That is the unromantic version of leadership.

Open societies will also misuse tools. People are people. Companies chase engagement. Agencies overcollect. The difference is friction. Courts, legislatures, rival media, and messy elections create drag. Drag is annoying when you want to build fast. Drag is also why a surveillance stack is harder to complete in public view.

Speed without accountability is not efficiency. It is a bet that the operator will stay benevolent after the tool becomes too useful to surrender.

Robotics And The Wider Industrial Race

AI is not a standalone product line. It is sinking into factories, warehouses, vehicles, and humanoid prototypes. Chinese firms have moved aggressively in several of those lanes. That does not mean they have won every benchmark. It does mean the contest is already physical, not just conversational.

A humanoid that can work a shift is not a novelty if the software brain improves every quarter. Combine cheap hardware, dense supplier networks, and state credit, and you get iteration speed. Combine that with export markets hungry for automation, and you get standards that other countries end up living inside.

The West still has deep advantages: research universities, capital markets, immigrant talent, and a culture that lets oddballs try strange ideas. Those advantages shrink if domestic politics treats the buildout as a moral contaminant. You cannot lecture the world about responsible AI while refusing to host the plants that make responsible capacity possible.

Look at that stack and the rare earth analogy stops being a metaphor. It becomes a packing list.

The American Temptation To Constrain First

Open societies have a habit. When a technology feels large and strange, the first instinct is to write rules before the thing even settles. Sometimes that instinct saves us. Seat belts. Drug trials. Bank capital. Sometimes it just hands the field to a player who does not share the same hesitation.

I have no patience for the fairy tale that AI should be left in a garage with zero guardrails. Fraud, child exploitation, bio risk, and automated scams are real. So is the chance that sloppy models will make bad medical or legal suggestions sound confident. Guardrails can be smart.

The naive version is different. The naive version says pause the American side until every social anxiety is soothed, then assume the rest of the world will wait in the hallway. That is not how contests work. A pause here is a subsidy there.

Some politicians talk as if throttling domestic development is a form of virtue. It can sound compassionate in a hearing room. It sounds less impressive if the result is a world in which the most capable systems answer to a party that already jails critics and maps dissidents.

There is a grown up sequence. Build enough capacity that democratic states are not beggars. Write rules that target concrete harms. Keep research in the open where possible. Keep the most dangerous capabilities behind serious access controls. Compete for talent. Fix permitting so clean power and transmission can actually show up where the loads are. That is harder than a slogan. It is also the only path that does not outsource the century.

What Leadership Would Actually Require

Leadership is not a press release about values. It is a pile of unglamorous work. Grid interconnection queues that do not last a decade. Water policy that distinguishes a well run campus from a sloppy one. Trade tools that keep the most sensitive chips from walking into military end use without pretending every student visa is a spy novel. Immigration rules that still attract the people who can train and govern these systems.

• Power first, because models do not run on statements.
• Permitting that is strict on pollution and fast on process.
• Allies that share compute and standards instead of fragmenting into boutique regimes.
• Public honesty about job churn, so workers are not treated as an afterthought.

West Virginia style pitches, and others like them, are easy to mock from a coastal office. They may still be right on the industrial point. Places with spare land, hungry labor markets, and political appetite can absorb load that richer suburbs will fight. If those sites are built with decent environmental practice, the national interest and the local interest can overlap. If they are built sloppily, opponents will have a better case next time.

I’ve sat with enough community meetings in my head to know the tone. People do not hate the future. They hate being the last to hear about it. Show the tax numbers. Show the water plan. Show the decommissioning bond. Show who gets hired first. Skip that work and every project becomes a morality play.

Culture Is Part Of The Strategic Picture

There is a softer layer that hard power people like to skip. A society that cannot defend its own cultural confidence will struggle to defend a technology stack. That is not a culture war punchline. It is a practical observation. If elites treat their own inheritance as an embarrassment while a rival treats cohesion as policy, the rival will read that as fatigue.

Beijing does not hide the preference for internal conformity. New legal language around ethnic unity is of a piece with older campaigns. The West, by contrast, often performs disagreement as a public sport. That sport has benefits. It also has a cost if it becomes inability to name a national interest out loud.

AI will sit inside schools, newsrooms, churches, parties, and barracks. The models will absorb the priorities of the people who fund, regulate, and fine tune them. If those people are embarrassed by their own political tradition, the systems will still have a tradition. It just will not be theirs.

The Oversight Paradox

Everybody agrees there will be oversight. The United States already overlays finance, aviation, medicine, and communications with thick rule books. AI will not float outside that habit. The live question is dosage.

Too little, and you get scams, leaks, and reckless deployment. Too much, and you get a compliance maze that only the largest incumbents can survive, while state backed foreign labs keep shipping. The sweet spot is boring: audit trails, evaluation suites, liability for specific harms, and export controls that track real military usefulness instead of every open source repo.

A workable formula is blunt:
  Build capacity in the open.
  Constrain proven harms in law.
  Compete as if the other side means it.
  Do not confuse a pause with peace.

Notice what is missing from that formula. Missing is the fantasy that a technology this useful will remain a consumer toy. Missing is the idea that moral language can substitute for megawatts. Missing is the hope that a rival will accept second place because a Senate speech was eloquent.

What Losing The Race Would Feel Like

People imagine losing as a single dramatic moment. A model announcement. A robot demo. A stock drop. That is not how infrastructure losses arrive. They arrive as defaults. Your hospital buys a diagnostic layer trained under another country’s privacy norms. Your manufacturer licenses a factory brain whose update path runs through a foreign cloud. Your teenagers grow up inside recommendation systems tuned to someone else’s political sensitivity list.

Then come the quieter dependencies. Spare parts. Fine tuning services. Safety filters that silently refuse certain queries. Insurance models that price risk with a political accent. By the time the public notices, the switching cost looks enormous.

That is why the rare earth story stings. It was sold as specialization. It became a choke point. AI could become a thicker choke point because it sits closer to decision making itself. Minerals constrain hardware. Models constrain thought at the margin. Not by magic. By convenience. People use what is cheap, fast, and good enough.

A More Adult Public Conversation

We need fewer apocalypse posters and fewer utopia posters. The useful register is industrial. How many gigawatts? How many advanced packaging lines? How many grid upgrades? How many water reuse systems? How many evaluation labs that can actually red team a model before it sits in a government workflow?

Rhetorical questions help only if they lead to lists. Who trains the next cohort of power engineers and chip designers? Who keeps basic research public enough that smaller firms can still enter? Who explains to a town that a campus can be a neighbor rather than a siege engine? Those are unfashionable questions. They are also the ones that decide outcomes.

I do not buy the claim that American leadership is guaranteed by past glory. Past glory built the internet, the chip industry, and a research culture that the world still copies. Past glory does not pour concrete. It does not issue a transformer. It does not sit through a utility hearing.

Nor do I buy the opposite claim that the race is already lost. It is not. Capital still concentrates here. Talent still wants to work here when the welcome is real. Allies still prefer an open stack to a party stack, even when they grumble about American companies. Those are cards. Cards expire if you refuse to play them.

The Stake Is Larger Than A Market Share Chart

Markets will price winners and losers, of course. Chipmakers, utilities, construction firms, and software platforms will move. Investors should watch power availability as closely as they watch parameter counts. That is a market story. It is not the whole story.

The whole story is who sets the default settings of the next public square and the next factory floor. Default settings sound dull until you live inside them. They decide what can be said, what can be sold, what can be seen, and what gets flagged as unsafe. A government that already treats disagreement as a security issue will not design generous defaults.

If AI is mental energy, then control of AI is control of a new kind of grid. Grids are never neutral once they exist.

So the practical ask is narrow and demanding at the same time. Do not romanticize a pause. Do not romanticize a free for all. Build. Measure. Constrain the abuses that can be named. Keep the least accountable major government from becoming the landlord of the century’s most useful machine.

That last sentence will sound dramatic to people who still think this is about chat windows. Let them. The rest of us can look at energy history, look at rare earths, look at the surveillance already in place, and admit the pattern. Tools this large do not float. They dock in a political harbor. The only open question is which harbor gets to charge the rent.

What To Watch Over The Next Few Years

Ignore most product launches. Watch interconnection queues, nuclear restarts, gas peakers paired with storage, and whether environmental reviews can be serious without becoming eternal. Watch packaging plants and the dull machinery that turns a wafer into something that can sit in a rack. Watch allied export coordination that actually holds. Watch whether local governments learn to negotiate community benefit deals instead of performing purity.

Also watch the models that never get a friendly demo. The ones used for targeting ads can be spun toward targeting people. The ones used for warehouse routing can be spun toward logistics under sanction. Dual use is not a theory. It is a calendar.

If those indicators improve in open countries, the panic can stay in proportion. If they stall while closed systems keep iterating on robots, chips, and domestic control software, the argument we are having now will look quaint. Future readers will not ask whether a county board was right to worry about a substation. They will ask why a civilization that invented so much of the stack decided to hesitate at the exact moment the stack became statecraft.

I keep a working bias, and I will say it plainly. I would rather live with American arguments about AI than live under someone else’s finished system. Arguments are loud. Finished systems are quiet. Quiet is not always peace. Sometimes quiet is just the sound of a standard that already won.