Behind the AI Hype: Takeover by Militarized Private Equity
The Pentagon, private equity and AI are reshaping US power
When it comes to China, America has a plan
AI competition with China is driving the United States’ new fusion of military power, industrial planning and private capital. Countries are lining up behind the US strategy.
by Evgeny Morozov
This is another great article by Evgeny Morozov from pp12-13 of the June 2026 edition of Le Monde Diplomatique English. It exposes more of what the USA Tech Bros and MAGA Trumpicans are really doing behind the façade of saving America for AI.
For a sub to the monthly LMD English, go to mondediplo.com.
Evgeny Morozov is founder and editor of The Syllabus ( go to https://www.the-syllabus.com for sub), a nonprofit knowledge curation service. Every week, subscribers get links to lots of new progressive material to read, watch and listen to.
The easiest way to misunderstand the Trump administration is to mistake the noise for the plot.The insults, tariff tantrums, cryptocurrency rackets, cabinet melodrama and television-ready cruelty draw attention to the circus tent.
The plot, meanwhile, has a price tag: $1.5tn for the freshly renamed Department of War, the largest defence budget in US history in nominal terms – and in real terms approaching the annual peak the country managed while fighting Hitler. It is being financed, with the elegant arithmetic of a man billing his own children, by trimming the parts of government that feed, school and treat them.
What that budget is supposed to buy is a structural transformation of the American state, its economy and its place in the global order. The post-cold war republic that called itself a free market is letting the mask slip. The machinery had been there all along – DARPA (Defense Advanced Research Projects Agency), In-Q-Tel (the CIA’s information technology investment fund), the national labs – what political economist Fred Block calls America’s ‘hidden developmental state’ .
Washington now picks industries, sets prices, takes equity in private firms and conditions foreign aid on political loyalty without bothering to call it anything else. Cold war – and even post-cold war – vocabulary doesn’t fit any more: ‘military Keynesianism’ meant increasing defence spending to boost aggregate demand; ‘military neoliberalism’, the same thing routed through private contractors and deregulated supply chains. What’s happening now is allocative, not fiscal – deciding which firms should exist, at which prices, with which inputs, in which sectors. And the war that’s coming is economic, not physical: it will be fought on balance sheets, not battlefields.
Behind all this, obviously, is a fear of China, and the technology that has dragged every other priority into step behind it: artificial intelligence. None of the underlying problems – minerals, chokepoints, chips, cables, reactors – are new. Pentagon planners and RAND Corporation analysts have muttered about them for decades, but every administration has nodded politely and changed the subject. What AI has done is fuse them into a single story in which falling behind looks like losing a war.
Which country can buy what
The story began in 2016-17, when a small group of China hawks – including Peter Thiel, who was almost alone in Silicon Valley in supporting the first Trump administration – argued that the US had to block Beijing’s advance in mobile technology, 5G and chips. The Huawei ban in 2019 was the prototype. None of that was abandoned by Biden; his administration refined it, creating sophisticated bureaucratic tiers that decided which countries could buy which US technologies, and extended the export-control regime beyond chips to chipmaking tools, EDA (electronic design automation) software, advanced memory, biotech, quantum information infrastructure and much else. Across three administrations of nominally opposing parties, there has been nearly perfect agreement on this question. AI brought matters to a head – it has become the commodity every country believes it must have access to, and the US controls the chokepoint.
What Trump’s second term has added to this bipartisan inheritance is consolidation. The planning is now done by the Pentagon, which sits on a massive pile of cash and has the discretion to spend it largely as it sees fit. Its gargantuan budget is, above all, a hiring problem – it needs people fluent in nine-figure decisions, comfortable with deploying capital rather than troops. The task of locating such people is no longer left to a personnel office in a windowless basement but entrusted to highly persuasive headhunters who bill by the hour.
A presentation prepared by one such firm (and eventually leaked) read less like public service than a private-equity fever dream. The Pentagon was looking to form a 30-person team to help oversee the use of up to $200bn of public money over three years, with salaries reaching $300,000 inside the building and $600,000 for those employed via government-aligned nonprofits. It also offered ‘unmatched access’ to senior government officials and generous ‘exit opportunities’ such as launching one’s own fund, drawing on ‘fund-raising channels that include royal families’ (2).
This enticing prose was created for the Economic Defense Unit (EDU), created within the Pentagon in April 2026 (3) by deputy defence secretary Stephen Feinberg, billionaire co-founder of Cerberus Capital Management and one of Donald Trump’s largest donors. Cerberus built its $65bn empire on the ruins of former leading US manufacturers – from car giant Chrysler to Remington, among the country’s oldest makers of firearms.
The EDU will work with the Office of Strategic Capital, a Biden-era loan facility now reorganised under Feinberg as an investment desk pushing public money into minerals, undersea cables, drones and satellites. How the OSC actually moves money can be seen from its first big contract. Donald Trump Jr, is a partner at 1789 Capital, which took an equity stake in Vulcan Elements three months before Vulcan received a $620m Pentagon loan – the largest in the agency’s history, fast-tracked under a March 2025 executive order that reduced approval requirements for critical mineral deals. This new capital is as strategic as it gets: Vulcan makes the rare-earth magnets the Pentagon needs for everything from F-35s to AI data-centre cooling, and on which China has a near-monopoly.
The Vulcan deal is less anomaly than prototype, and the people replicating it have stopped pretending otherwise. Enter the EDU’s big boss George Kollitides II, a former Cerberus executive last seen steering Remington through its post-buyout convulsions. Now the Pentagon’s principal advisor on economic competition, he said the quiet part out loud at the Milken Institute conference this May.
A ‘transaction-driven approach’
China, he explained, had been ‘successfully waging economic warfare’ for 30 years; the answer was a ‘transaction-driven approach’. The unit’s doctrine, as he described it, has two pillars. One he called ‘fix it’: the reshoring, midstream processing and metallisation of critical minerals. The other is ‘break it’: counter-industrial-base measures, denying adversaries the capacities the US is belatedly trying to reclaim. And the EDU has the right team to fix and break alike. Its staff, Kollitides explained, hail from ‘private equity backgrounds or leverage finance backgrounds’.
A diagnosis underpins this doctrine, and it is here that Kollitides briefly, almost accidentally, became interesting. The post-1991 American economy, he explained, had been the product of ‘flawed thinking’: capital flowing abroad chasing lower regulation and cheaper labour, the ‘peace dividend’ promised by George HW Bush turning into a peace trap, industrial muscle bled dry to finance a 30-year consumer boom. It’s the kind of indictment one expects from Steve Bannon, not a private equity man speaking in a Beverly Hills hotel ballroom. The Pentagon’s economic warfare directorate is now staffed by people like Kollitides who spent three decades profiting from the decline of US industrial capacity before returning as national saviours.
And they are the story. The post-cold war US has always known how to plan when it has to. The novelty is who is doing the planning now: the alumni of Cerberus, Apollo, Cantor Fitzgerald – arsonists applying to be firefighters, who learned to allocate other people’s money inside private partnerships, and who now spend taxpayers’ money with the same instincts and a similar fee structure.
Their machinery has three gears: capital, price and energy.
For the first, under Feinberg the Office of Strategic Capital has moved beyond lending toward outright ownership – a step which, lawyers have noted, exceeds its powers and which the Republican-dominated Congress has declined to enquire about. The Pentagon now holds equity in MP Materials and Trilogy Metals, and warrants (securities that entitle the holder to buy stock) in Vulcan Elements and ReElement Technologies, alongside the government’s headline stakes in Intel and Lithium Americas. Trump’s One Big Beautiful Bill quietly handed the OSC $1.5bn in fresh credit subsidy, leveraged into roughly $200bn of loan authority over four years – half earmarked for critical minerals, half for 31 ‘critical technology’ sectors. The old US catechism held that the state must never pick winners. The new rule is to pick them before Beijing does – and make sure they know whom to thank.
As regards price, in December 2024 China put an embargo on sales to the US of certain refined rare earths, including gallium, for which it controls 98% of the market. Though the embargo was lifted a year later, Washington’s answer, announced in February 2026, was Project Vault, a $12bn fund combining public and private capital to create a stockpile of the 60 minerals on the US Geological Survey’s critical list, with off-takers committing in advance to fixed purchase prices. Alongside this is the Forum on Resource Geostrategic Engagement (FORGE), a preferential trading bloc of allied countries built around coordinated price floors, subsidies and long-term offtake commitments. The free market, after a few patriotic drinks, has discovered administered pricing.
And of course AI doesn’t run on fresh air, even if Silicon Valley presents it as a clean ascent into the ethereal realms of abstraction: models, tokens, clouds, genius. In material terms it’s a demand shock – for electricity, water, gas turbines, uranium, the chips that go inside the racks and the substations that feed them. The Department of Energy, where Peter Thiel is said to have personally vetted candidates for a top nuclear job, while his Founders Fund-incubated startups have since won the largest contracts, has been instructed to restore atomic power to the strategic importance it had in the early cold war.
The Department of Energy’s Genesis Mission, announced in November 2025 with language drawn from the Manhattan Project, consolidates 17 national laboratories – including Los Alamos, Oak Ridge, Sandia and Argonne, with 40,000 scientists between them – into a single AI-for-science platform whose stated priorities run from nuclear fission to semiconductors and critical materials.
Demand is years ahead of supply, yet the government’s response is not market clearing through prices but administrative fast-tracking by executive order, treating environmental review as a vestigial organ. Without the AI panic, the same financiers would still be in Park Avenue offices writing op-eds about American decline. They now have offices in the Pentagon, and the op-eds have become procurement orders.
Carrot and stick
Seen from outside, this system appears not as industrial policy but as coercion with a smile tight enough to crack porcelain. The US does not ask allies to admire its values. It only asks them to align supply chains, chip exports, mineral contracts, data rules and energy infrastructure with Washington’s China strategy. The carrot, for those who comply, is US-made GPUs, cloud capacity and cheap financing. The stick is having their names struck from US licensing lists. So far the project has met almost no resistance.
Spearheading this policy is the Pax Silica initiative, launched on 12 December 2025, which aims to secure the AI supply chain and ‘reduce coercive dependencies’ (7). The seven founding signatories – the US, Australia, Japan, South Korea, UK, Singapore and Israel – have since been joined by Finland, Greece, India, Norway, the Philippines, Qatar, Sweden and the United Arab Emirates. Each new signatory has come bearing its own bargain. India arrived with $210bn in commitments from its oligarchs to build domestic AI infrastructure on US tech platforms (India is a BRICS member, but that contradiction has gone unaddressed). The UAE brought G42, scrubbed of inconvenient Chinese entanglements and refitted with Microsoft oversight under pressure from the Biden administration. Malaysia, never a signatory, supplied the cautionary tale: in May 2025 its deputy communications minister announced a Huawei-based national AI strategy; 24 hours and one tweet from Trump’s AI czar later, her remarks had been retracted. The EU is now in talks to join the alliance (Bloomberg, 12 May 2026).
The Netherlands attended the founding summit but, with the polite intransigeance of a country whose national champion ASML supplies lithography machines to every advanced semiconductor manufacturer on earth, declined to sign. The US may have found a way to make it fold. The MATCH (Multilateral Alignment of Technology Controls on Hardware) Act (8), introduced in early April with bipartisan sponsorship, would give allied countries with chip-equipment industries 150 days to harmonise their export controls with Washington’s, or face unilateral retaliatory measures. ASML, whose CEO long described the controls as more ‘economically motivated’ than security-driven, is in the act’s crosshairs; the Dutch government has formally objected.
What remains for the global South?
For the global South, the language is less upholstered. USAID, Washington’s foreign aid agency, was gutted in the first half of 2025 and the instruments that replaced it – the Development Finance Corporation (DFC) and Export-Import Bank – have been openly redrawn as tools of economic warfare. The DFC was reauthorised in late 2025 at $205bn, more than triple its previous cap, and handed a fresh equity investment mandate. Its chief executive Benjamin Black comes from private equity and has no development experience. At a Council on Foreign Relations forum in April, he described the agency’s purpose with the unembarrassed plainness of a deal memo: the DFC, he said, would ‘build holistic economic ecosystems, anchored to US capital markets, insulated against adversary control and structured to generate strong returns for the American taxpayer’.
Under Trump II, loans come with rights over outputs. In December 2025 the DFC signed a $553m loan to the operators of the Lobito rail corridor. The object is to upgrade the Angolan section of this strategic line, which transports Congolese copper and cobalt, and eventually Zambian copper, westward to the Atlantic, towards US battery and chip supply chains – and out of China’s chokehold on the region’s mineral exports.
When conditional loans aren’t enough, the US threatens to close the medicine cabinet. The America First Global Health Strategy, rolled out in September 2025, bundles HIV and tuberculosis funding to dozens of countries – most of them in Africa – with mandatory transfer of national health data and, in the Zambian draft, 25 years of pathogen genomic sequencing data to US agencies and pharmaceutical contractors. A leaked State Department memo proposed offering Zambia $1bn in health aid over five years – roughly half its funding level until 2025 – on condition that Lusaka sign a separate compact granting the US access to its copper, cobalt and lithium (9). Kenya’s High Court has suspended its arrangement pending a constitutional challenge; and when Zimbabwe walked out of a $367m deal, Washington promptly announced it would wind down its health assistance there.
The greatest trap for the global South, however, lies in the false promise of ‘AI sovereignty’ it has been offered as consolation. Buying into the American stack means importing the data centres, the chips, the models and the software, locating them domestically, and pretending the result is national capacity. It is not. The chips can be turned off. The models are black boxes protected under trade secrets law. As for the data centres, without a link to a broader industrial development strategy they become extractive appendages whose chief output is more leverage for Washington.
The classical developmentalist playbook from the 1970s and 80s, in which a country began by manufacturing parts and worked its way up the ladder to cars and then electric vehicles, has no obvious AI equivalent. Without one, ‘sovereign AI’ is a trap, not a solution. The country that understands this best is China, whose data centres in Inner Mongolia serve the domestic economy first and export industry second. China’s National Data Administration began, this spring, to measure economic activity in tokens, the basic units of text processed by AI models – that is to say, according to criteria closely linked to energy utilisation, which puts capacity planning and allocation in a new light. China’s daily consumption has now topped 140 trillion tokens, up more than a thousandfold since 2024.
‘The British Empire modelled it’
At the Milken conference, Kollitides offered the most candid description to date of the change under way in Washington: ‘The British Empire really modelled it,’ he said, with ‘privatised companies like the East India Company, which were really public-private, government-driven organisations largely built around commerce and economics, and allowed them to leverage [the British] navy to apply all sorts of economic control around the world.’
That such an analogy could be made in 2026, without anyone raising an eyebrow or rising to mention the Bengal famine, or Plassey (the 1757 battle often regarded as the beginning of British colonial rule in India), or the centuries during which ‘public-private, government-driven organisations’ starved a subcontinent to feed Britain, is an admission in itself. The chosen model is not the New Deal, nor even the cold-war arsenal state. It is a chartered monopoly whose name remains shorthand for licensed plunder across half the world.
What Kollitides appears not to have noticed is that he is already inside the company. The Pentagon writes the charter, the giants of the cloud run the depots, and a small constellation of private equity houses provides the working capital. Palantir reads what is happening. Nvidia supplies the substrate. The US Treasury, somewhere down the corridor, fills in the paperwork. Holding the assemblage together, as the spice trade once did, is the AI build-out itself, which has fused profit and statecraft so tightly that anyone trying to distinguish where one ends and the other begins is treated as a quaint relic of the era of rules and procedures.
The East India Company took its dividends in cotton, opium and tea. This one takes them, principally, in tokens generated, prompts served, models fine-tuned on someone else’s data. Behind the inference sits the older ledger – copper and cobalt out of Congo, genomic sequences signed away in Zambian clinics in exchange for tuberculosis drugs that may or may not arrive, aquifers drained to cool data centres. From the Beverly Hilton, none of this is visible. From East India House on London’s Leadenhall Street, in 1770, the Bengal famine was not visible either. Blame it on the architecture.