The Trillion-Dollar Circle: How Big Tech Is Financing Itself Into a Corner

Artificial intelligence is no longer a promise for the future, it is already inevitably transforming business, work methods, and investment opportunities. One of the most pertinent questions now is not whether AI matters, but who is paying for the race to develop it, and what will happen when the bill comes due.

For all the talk about AI transforming the economy, one figure captures the scale of the bet: $410 billion. That is roughly how much Amazon, Alphabet, Meta, and Microsoft alone spent on capital investment in 2025, largely on AI infrastructure (chips, data centers, and computing power). Putting this into perspective, approximately an amount exceeding 100 billion of Portugal's GDP in the same year was put at stake. That number is set to approach the $650 billion mark in 2026, but the crucial problem still isn't necessarily the scale. The structure is.

Upon closer examination of how all of this is being funded, a rather strange picture begins to emerge. Nvidia is investing $100 billion in OpenAI, OpenAI has invested in AMD, and Microsoft is not only one of OpenAI's largest shareholders but also a major customer of CoreWeave, a cloud computing company in which Nvidia maintains a significant presence. At the same time, Oracle is building data centers primarily for OpenAI, even while anticipating losses in the process.

These same companies often present themselves as simultaneously acting as clients, suppliers, and investors of one another, with capital circulating within the system in ways that can inflate revenues and valuations at each stage. When Nvidia reports record profits, part of that performance may come from companies it helped finance, and the same happens when OpenAI's valuation rises, which will directly translate into the balance sheets of companies that hold stakes in it, including some of the very companies it depends on for chips, financing, and infrastructure. This is precisely what makes this structure so sensitive, it's not just about scale, but a mechanism sustained by mutual dependence and rising expectations. The IMF has already warned that this type of financial interconnection can become a systemic risk, making a correction much more difficult to contain.

Many opinions have been comparing this context to the dot-com bubble, which is understandable, but it's still a flawed analogy. When the dot-com bubble burst, companies essentially collapsed individually, examples include Pets.com and Webvan, and although it was painful, the financial system absorbed them cleanly. What is happening now has far more similarities to the 2008 crisis than to the 2000 crisis: deep interconnection, hidden leverage, and assets valued on assumptions that may not hold.

On top of this, there is another big problem with all this financing: debt. The Bank of England's December 2025 Financial Stability Report estimated that approximately half of the $5 trillion expected to be spent on AI infrastructure over the next five years will be financed externally, mostly through debt. But the assets that are built with this debt, data centers filled with chips that become competitively obsolete within months, may lose most of their value long before the loans are repaid. OpenAI alone has already committed to $1.4 trillion in infrastructure spending over the next eight years, roughly $20 billion in annual revenue, with the gap to be covered by further debt and equity raises. Its suppliers have already taken on $96 billion in debt to fund their side of those commitments, and when one company's distress can travel instantly through that many interconnected balance sheets, the problem stops being that company's alone and becomes somewhat more complicated.

The uncomfortable truth is that the gains that supposedly justify all this simply haven't materialized yet. Goldman Sachs calculations indicate that, despite historical levels of spending, AI contributed a negligible 0.2% to US GDP in 2025, and an NBER study from February 2026 found that 90% of companies report no measurable impact on productivity. The IMF and Goldman Sachs project significant gains in the future, but these are projections, and debt, unlike future returns, matures on a fixed schedule.

The consequences of that mismatch are not abstract. The Bank of England has warned explicitly that losses on AI lending would spill over into broader credit markets, raising borrowing costs for households and businesses that have nothing to do with artificial intelligence. Pension funds and insurers holding the bonds that finance this buildout would take direct hits, and, since AI stocks now represent roughly 30% of the S&P 500, a sharp correction would drag down the savings of anyone with even modest market exposure.

In 2000, the dot-com crash hurt investors who had chosen to speculate, but this time, through debt markets and index funds, the losses would reach people who never made that choice, and the Trillion-Dollar Circle is still spinning, though without steady productivity gains, it may be difficult to bring it to a stable end.