Tech giants are building colossal AI infrastructure around the world. Take Softbank’s planned 10-gigawatt AI Mega Campus, for instance—a massive complex powered by a mix of natural gas and nuclear Small Modular Reactors (SMRs). To put that scale into perspective, this single data center will demand more than double Scotland’s peak electricity capacity of roughly 4 gigawatts. SMRs deserve an article of their own, but the idea of private tech firms owning and running nuclear plants in an industry teetering on a speculative bubble should terrify us all.
Naturally, all of this is very expensive. It’s estimated that this year spending on AI will be $765 billion and will increase to $1.6 trillion by 2031. So how are they funding this exorbitant spending? To bankroll these staggering costs, tech companies are increasingly turning to creative financing tricks like special purpose vehicles and unregulated loan notes
Loan notes are quite simple things – they’re essentially an IOU. The lender hands the money over to the AI company with the agreement they will pay the whole amount back on a certain date with interest. ‘Regulated’ loan notes are often used for big projects such as property development or business expansion and you can invest in them through brokers in your town. Unregulated loan notes fall out with the scope of bodies like the Financial Conduct Authority, so they lack even the most basic of safeguards that banking regulations impose. As these are agreements between two private entities we have no way of knowing how many and how big these loans are, although they are sometimes included as annotations on balance sheets. Hidden debt that does not appear on balance sheets at U.S. tech giants Alphabet, Microsoft, Amazon, Meta and Oracle swelled eightfold in roughly four years to an estimated $1.65 trillion, exceeding the roughly $1.35 trillion in debt reflected on their balance sheets. Meta’s off balance sheet debt is particularly high at about $420 billion, nearly triple its recorded debt.
So how do these companies keep the debt off their own books? Simple – they ask someone else to do it! The money is raised for tech companies by private investment firms that create special purpose vehicles (SPVs) which are companies set up for one purpose, for AI that purpose is building the data centres. The arrangement is usually that the operator provides the land, buildings and power facilities, which the tech company leases over the long term. All that the AI companies have to declare is the rent that it pays to the SPV, not the billions of dollars of ULNs that built the data centre. Meta has teamed up with Blue Owl Capital to build a data centre in Louisiana and to get the deal over the line they have guaranteed to cover all losses should the data centre become unnecessary and the lease is terminated. Meta estimates this could cost $50 billion. If the AI companies had to put in their accounts the billions of dollars that they have borrowed their margins would look awful, scaring off investors, and holding them back in the race to develop. Tech companies still aren’t making money with costs estimated at being between five and ten times revenue. Tech companies expect future earnings to exceed these debts. They might make profit in the future – Microsoft, Alphabet and Amazon’s publicly disclosed backlogs for cloud services and other businesses totaled roughly $1.45 trillion as of the end of March 2026. Amazon Web Services CEO Matt Garman said his company’s investments are “not speculative.”
When the loan notes are due to be repaid it appears that the companies are going out to fund raise again with a new round of loan notes. When the previous investors are paid off with new investors’ money we usually refer to this as a Ponzi scheme. Marx would call ULNs ‘fictitious capital’, the promise of future value being traded as a commodity today. Real capital consists of tangible assets used to exploit labour and generate value – factories, machines and the likes. Fictitious capital creates the illusion that money can simply generate more money by its own nature, completely hiding the fact that all real economic value derives from unpaid surplus labour.
History doesn’t repeat itself but it does often rhyme. Both Enron and Lehman Brothers used SPVs extensively to hide debt from investors. When the markets started to correct themselves both were caught out and are now case studies in bad practice used in business schools. We all know that when the bubble bursts and the cash dries up it will be the working class that are punished for the gambling of the rich.