The numbers are genuinely staggering. Tech giants – the so-called hyperscalers like Microsoft, Google, and Meta – are locked in an arms race for computational power. According to a recent analysis by McKinsey, the world will need to spend an eye-watering $6.7 trillion on data centres to power the AI revolution. To put that in perspective, that’s more than the entire GDP of the United Kingdom and France combined. AI dealmaking hit a cool $1 trillion in 2025, and it’s only accelerating. This isn’t just about building a few more server racks; it’s about constructing a global infrastructure on a scale never seen before.
The Trillion-Dollar Question: Where’s the Money Coming From?
You might think these tech behemoths, sitting on mountains of cash, would fund this expansion themselves. But that’s where the story takes a fascinating and slightly unnerving turn. Even they are balking at the price tag. Take Oracle, a cornerstone of the enterprise tech world. According to JPMorgan, its debt-to-equity ratio has ballooned to a shocking 500% as it scrambles to keep up. The hyperscalers, which now account for over 26% of the entire S&P 500, are increasingly looking for… let’s call it ‘creative financing’.
This is where the real complexity begins, and where the risk starts to seep into the foundations of our financial system. We’re seeing the rise of massive, off-balance-sheet deals structured to fund these AI mega-projects. As detailed in an excellent piece by Racket News, these aren’t your typical business loans.
– Meta is reportedly working on a project codenamed ‘Beignet’, a $30 billion data centre fund. But here’s the kicker: Meta is only putting in $3 billion of its own equity. The other $27 billion is being raised as debt.
– Similarly, Elon Musk’s xAI is financing its ‘Colossus’ supercomputer with a $20 billion deal, comprising $7.5 billion in equity and $12.5 billion in debt notes.
So, who is lending them tens of billions of pounds? The money is flowing from the vast, opaque world of private credit. And where do private credit funds like Blue Owl and Apollo get their capital? You guessed it: from institutional investors. This means pension funds and insurance companies are funnelling our retirement savings and policy premiums into these high-stakes insurance tech investments.
A House of Cards Built on Silicon?
This intricate financial plumbing creates a systemic risk that looks eerily familiar to anyone who remembers 2008. Back then, the risk was hidden in complex mortgage-backed securities. Today, it’s being packaged into AI asset-backed securities, with $11 billion worth created in 2025 alone.
Think of it like this: Instead of a loan being backed by a house in a suburban neighbourhood, it’s backed by a shiny new data centre in a remote corner of Arizona. But what is that data centre actually worth? Its value is almost entirely dependent on the thousands of Nvidia GPUs humming away inside. This presents two glaring problems.
1. Rapid Obsolescence: The pace of AI development means today’s cutting-edge chip is tomorrow’s paperweight. If a new technology makes the current generation of GPUs obsolete, the value of the asset backing that multi-billion-pound loan could evaporate overnight.
2. Concentration Risk: The entire ecosystem is propped up by a handful of companies. The hyperscalers need the AI models from firms like OpenAI and Anthropic to justify their infrastructure spend. But are those models a sound long-term bet? HSBC Global Investment Research projects that OpenAI still won’t be profitable by 2030, even as it needs a further $207 billion in compute power.
This raises serious tech bubble concerns. We are building a trillion-dollar industry on the promise of future profitability that may never materialise, and the financial instruments used to fund it are untested and incredibly risky.
Where is the Accountability?
All of this points to a monumental failure of algorithmic accountability. Not in the AI models themselves, but in the financial algorithms used by investors. How can a pension fund accurately price the risk of a 10-year loan to a Special Purpose Vehicle whose only asset is a data centre filled with chips that might be worthless in three years, all to power an AI company that has never turned a profit?
The simple answer is: it can’t. The complexity serves to obscure the risk, not mitigate it. We are pouring the safest money in our economy – the money set aside for our retirements – into some of the most speculative ventures in technological history.
This is precisely where financial regulation needs to step in. The current rules were not designed for a world where off-balance-sheet vehicles backed by pension funds are used to finance a technological arms race. Regulators seem to be several steps behind, mesmerised by the same AI hype as everyone else. We need a regulatory framework that can pierce the corporate veil of these complex deals and force a transparent accounting of the real AI finance risk exposure.
The party is still going strong, and the returns from private credit look fantastic on paper. But it’s built on a chain of assumptions: that AI demand will be infinite, that the current technology will remain dominant, and that the AI model-makers will eventually become wildly profitable. If any one of those assumptions proves false, the chain breaks.
When the music finally stops, who will be left without a chair? It won’t be the tech giants, who have cleverly shifted the risk off their own books. It will be the pension funds, the insurance companies, and ultimately, the millions of ordinary people whose financial security was gambled away on the promise of a thinking machine.
What are your thoughts on this? Are regulators doing enough to protect our savings from the fallout of a potential AI bubble? Share your perspective in the comments below.
