Investigative Report: An Army for the Machine: Trump Declares Those Warning About AI His Enemies and Makes the Planet Pay for His Megalomania

The most powerful man in the world has called up a force to protect something that thinks faster than its creators, and in the same breath declared those who want to slow it down his enemies. Donald Trump announced on his Truth Social platform that he was creating an “Artificial Intelligence Force” to protect the technology from its “destruction.” Its future was under threat from the “Radical Left,” he wrote. He would fight any attempt to hinder or restrain the growth of this industry. And just like that, the tool becomes the one being protected, and anyone who questions it becomes the enemy.

To justify it, he rattled off a list of alleged deceptions that the “Radical Left” had used to mislead the country. He mentioned “men in women’s sports” and the dispute over Russia and Ukraine. He included global warming, along with the impeachment proceedings during his first term. Now, he claimed, Democrats were intent on the “destruction of AI.” He would not stand by and watch, the president declared. The attack had started with data centers and failed once people realized how useful they were. “Higher wages, lower taxes, safer streets, that was the result,” Trump said. Under no circumstances would he hinder or restrain the industry’s growth. On the contrary, it would be “protected, supported, and supervised as it grows.” Anything bad could easily be uncovered through the criminal and civil justice system. He was creating the AI Force just as he had created the Space Force during his first term, which he said had been a tremendous success.

Read also our article: Trump’s answer to the AI threat: his “high IQ” should be enough - Rahm Emanuel wonders whether the president is even awake

Soon, Trump announced, he would appoint an “AI czar” with an “extraordinarily high IQ.” In Washington, that is what presidential appointees tasked with coordinating a particular area are called. A supervisor, then, for a field nobody has mastered yet. The language gives away the ambition. An industry gets an army and a czar, borrowed from the vocabulary of war and empire. What began as a business is declared a matter of national importance, complete with a commander in chief and an enemy. Once a technology is built up this way, it no longer asks for oversight. It demands obedience.

The announcement came in the middle of a heated debate over the future of the technology and the dangers of its breakneck pace. One of the biggest points of contention is the construction of new data centers, which some want to rein in. And now the industry itself is calling for a slowdown. In mid-September, the three largest American AI companies agreed to “slow down” the pace of development: Anthropic, OpenAI, and Google. Their concern is that the ability to control these systems may no longer keep up with their growing capabilities.

The push came on Saturday from Dario Amodei, the CEO of Anthropic, the company behind the Claude assistant. He described China as an authoritarian state that could misuse the technology, while also demanding that the United States negotiate international limits with Beijing. The warning from this company is nothing new. Back in June, co-founder Jack Clark and Marina Favaro, who heads the Anthropic Institute, warned about systems that could improve themselves recursively and grow beyond human control in the process. The warning is coming from the people building the machines, and that makes it hard to ignore. People rarely question their own work because it is the easy thing to do. The industry’s simple fear is that these systems are growing more capable faster than the means of keeping them under control. In Washington, that very doubt is being turned into an offense, with criminal and civil law set to be deployed against those trying to hit the brakes. A government that threatens the watchman and protects the fire has turned the roles upside down.

The method has been tried before. Anything that does not fit the picture gets dismissed as a trick by the opposition, global warming just like the war in Ukraine, the old investigations just like the doubts about AI now. Every inconvenient truth gets tossed into the same drawer, labeled “Deception.” That is how a kind of politics grows that would rather go after the messenger than examine the message.

Beijing pushed back sharply. On Monday, a spokesperson for the Chinese Foreign Ministry accused executives at American AI companies of spreading “threat scenarios” and thereby fueling confrontation and unfair competition. State-affiliated media portrayed the calls for a slowdown as part of a broader effort to harm China’s AI industry. A Chinese state newspaper spoke of a “Cold War playbook” and a “silent Cold War” over artificial intelligence aimed at containing China, calling the American approach “hypocritical and shortsighted.” The real goal, it argued, was to slow China’s development through technological barriers and control over the rules of the future. That would preserve American dominance and leave China shut out of the global process of shaping AI regulations.

China’s Minister of State Security, Chen Yixin, drew a direct line in a magazine published by China’s internet regulator between the push to slow development and American export restrictions on chips and AI software. Some countries, he wrote, were using their technological lead to restrict others’ access to advanced technology under the banner of “national security.” Artificial intelligence had become the main arena of technological competition, a new battleground in the strategic rivalry between major powers. He warned that artificial intelligence could increasingly threaten the political stability and power of the Communist Party. Hostile forces were using generative AI to spread political rumors and false information; at the same time, he said, American AI systems such as Claude and ChatGPT-5.5-Cyber posed a threat to China’s critical information infrastructure. The very government that has spent decades building an enormous censorship and surveillance apparatus now finds itself facing a technology that could slip beyond its control. Meanwhile, former employees of American AI companies are warning about the reckless development of increasingly powerful systems, while Washington and Beijing are expected to negotiate international rules. Chen’s remarks make clear that China now sees artificial intelligence as a potential threat to its own political rule. Both American administrations, Joe Biden’s and Trump’s, had held back China’s development through chip restrictions. Beijing had responded with processors of its own.

Advocates of tighter controls hope the issue will come up when Trump and Xi Jinping meet in Washington on September 24. Beijing’s skepticism and Trump’s own rejection of any slowdown, which he voiced on Monday, make a strict international agreement seem unlikely.

Two powers are dividing up the future of this technology between themselves, and both talk about security when what they mean is dominance. Beijing sees the call to hit the brakes as a weapon, and American export restrictions as proof of it. Caught between them are more than 150 countries without computing power of their own, countries that get no say in building it or making the rules but will still end up bearing the costs. A race for control over the thinking of machines, fought over the heads of the rest of the world.

A way out would require trust, and that is precisely what is missing. Amodei calls for an understanding with Beijing while describing it as authoritarian in the same breath. Trump rejects any attempt to hit the brakes, and China sees a trap behind every proposal. The two powers that could draw a line distrust each other the most, leaving the most consequential technology of this century without a shared hand on the controls. While the giants circle each other, people down in the states are stepping on the brakes. On July 14, New York Governor Kathy Hochul signed an executive order banning the construction of new AI data centers exceeding 50 megawatts for one year, making New York the first state in the country to do so. Rising utility bills, the depletion of natural resources, and uncertainty for residents were looming; she said it was her duty to act.

The order denies permits for particularly large facilities exceeding 50 megawatts and requires local governments to develop standards for future construction that take energy and water consumption, as well as environmental impacts, into account. Similar bans are being considered in more than 10 states. In Maine, the legislature even passed one, but Democratic Governor Janet Mills vetoed it. New York’s legislature also approved a bill, but Hochul did not sign it and instead turned to an executive order. At the local level, Monterey Park in California got things started; among cities, Seattle became the largest to take the same path so far.

Example: The planned Stratos data center in Box Elder County, Utah, is being developed by O’Leary Digital Utah Development Co., a company owned by Canadian businessman Kevin O’Leary. Physicist Rob Davies of Utah State University calculated that, at full buildout, the facility and its power generation could release an amount of heat each day equivalent to the energy of 23 atomic bombs. The project was originally planned to cover around 40,000 acres, but the developer has since announced plans to reduce that to around 20,000 acres. For the planned Stratos project at full buildout, Davies calculated a total thermal output of around 16 gigawatts, taking into account both the data center and its power generation. Over 24 hours, that adds up to an amount of heat he compared to the energy released by roughly 23 atomic bombs.

Our investigation in Utah shows just how deeply the expansion of these machines is already reaching into the way local governments operate. On June 10, Box Elder County imposed a 180-day freeze on new data center applications so it could examine the consequences for infrastructure and resources. One project alone was spared. The county itself makes clear that Stratos is not affected by the freeze because the land is no longer under its control.

While Box Elder County in Utah put new data centers on ice for 180 days, the gigantic Stratos project was exempt from the permitting freeze. The official documents shown here reveal why: The county had already transferred authority over land use in the project area to the state agency MIDA for 50 years. The agreement names O’Leary Digital Utah Development Co. as the project developer and provides for power generation facilities alongside the data centers. The county hits the brakes, but Stratos falls under a different set of authorities.

Back on May 4, the county had handed authority over land use in the project area to the state’s Military Installation Development Authority, or MIDA, for 50 years. Agreement No. 26-15 gives the agency primary authority over the area. The county’s usual building and land-use rules no longer apply there. The project plan lists around 40,000 acres of private land and another 1,200 acres of military and state land. The agreement with O’Leary Digital Utah Development Company is still marked as a draft to this day. It has not been signed. The county emphasizes that its approval of the project area does not amount to a permit for full construction; further environmental and building permits are still required. In May, Utah’s environmental agency stated that it had not yet received an application for an air quality permit.

Back on May 4, the county had handed authority over land use in the project area to the state’s Military Installation Development Authority, or MIDA, for 50 years. Agreement No. 26-15 gives the agency primary authority over the area. The county’s usual building and land-use rules no longer apply there. The project plan lists around 40,000 acres of private land and another 1,200 acres of military and state land. The agreement with O’Leary Digital Utah Development Company is still marked as a draft to this day. It has not been signed. The county emphasizes that its approval of the project area does not amount to a permit for full construction; further environmental and building permits are still required. In May, Utah’s environmental agency stated that it had not yet received an application for an air quality permit.

The public support is there. According to a June poll, 46 percent of New Yorkers support a one-year moratorium, while 21 percent oppose it. Nationwide, around three-quarters of Americans oppose having data centers in their neighborhoods, fearing environmental damage and rising costs. What is at stake was measured by the United Nations University in June 2026. Data centers, the backbone of AI worldwide, could consume up to 945 terawatt-hours of electricity annually by 2030. That is three times what Pakistan and Bangladesh consume together in a year, even with Nigeria added, countries with a combined population of more than 650 million people. By the end of the decade, the technology’s water consumption could equal the annual household needs of 1.3 billion people. Its land requirements would exceed 14,500 square kilometers, more than twice the size of the Jakarta metropolitan area.

Electricity is only the tip of it. Every unit consumed comes with a water footprint, from cooling the facilities and generating the electricity. Add to that the land footprint left behind by energy production and the long supply chains. The technology’s ecological footprint is growing at a pace that could overwhelm the planet’s natural resources. The usual figures on global warming capture little of that.

Read also our article: Not Smart, Just Huge” - The Resistance Against AI

A machine whose thirst could equal that of 1.3 billion people is no longer some distant threat. It is already standing in the landscape, humming behind fences and drawing water from rivers that are already drying up in droughts. In regions already struggling with scarcity, that is where the impact will be felt first. The thirst of the computers and the thirst of the people reach for the same jug.

The study lays bare just how badly the measurements are skewed. Most of the attention goes to greenhouse gases, especially those produced while training large models. The far greater appetite lies in everyday use: According to the study, 80 to 90 percent of energy demand comes from daily use. One widely used service alone processes around 2.5 billion requests a day and draws hundreds of gigawatt-hours of electricity a year to do it. Generating a single AI image uses more than 1,000 times as much energy as simply classifying a text, and a video uses far more still. Efficiency alone will not save anything. When costs fall and performance improves, consumption rises, the so-called rebound effect.

Nothing seems easier than typing a request into a text box. Its real weight is felt somewhere else, in rivers and power grids, on piles of scrap. Billions of such interactions every day add up to a burden that no bill ever shows the user. The convenience is real. Its price is simply far enough away to overlook. Even what looks green can be deceptive. What looks clean in one calculation drives up the demand for water and land in another, especially where resources are already scarce. Switching to certain renewable energy sources cuts emissions while consuming water and land at the same time.

The burdens are unevenly distributed. The benefits are felt worldwide, while the costs pile up in a handful of places. In some countries, data centers already draw a significant share of the electricity supply; in others, they consume vast amounts of water in the middle of droughts. By 2030, up to 2.5 million tons of electronic waste could be generated every year, much of it ending up in poorer countries with limited capacity to dispose of it safely. Mining the critical minerals needed for the equipment also leaves destruction and social upheaval in the regions where they are extracted. And computing power is concentrated: More than 90 percent is located in just two countries, the United States and China, while more than 150 nations have no significant infrastructure of their own. Without their own computing power, many of them also lack a voice in setting the rules.

And so a new divide opens up across the globe. Some enjoy the convenience, while others pay with water and land without sharing in the profits. A technology that promises to calculate everything has barely begun to reckon with the consequences of its own existence. All these numbers carry the same deadline. By 2030, by the end of the decade, those are the markers set by the study, and that is not some distant future. Over that same period, the builders themselves warn, the systems could improve faster than rules can be written. Two clocks are running, one for the climate and one for control, and both are moving faster than politics.

The researchers at the United Nations University are not arguing against the technology. They demand that it grow within the planet’s limits, through a responsible approach based, among other things, on openness and resource-efficient design from the very beginning. That includes fairness and accountability throughout the entire life cycle. Global cooperation is part of it, too, along with careful use of resources. Governments should plan for electricity and water demand from the outset, and land use must be part of that planning. Companies should keep their consumption low, and users should choose less resource-intensive applications. The future of artificial intelligence does not depend on technology alone. The political decisions being made today carry just as much weight.

An army for the machine, a czar for its rise. And anyone who questions it is under suspicion. The fire is guarded, while the hand holding the fire extinguisher is treated as the enemy. But the bill, paid in electricity and water, lands on the people who never lit the fire.