How will the AI bubble burst?

In the last year, tech giants like Meta, Microsoft and Oracle have been taking on tens of billions in debt to finance their artificial intelligence (AI) projects and data centres. The key supplier of these firms is Nvidia, the most valuable company on the planet, which designs and manufactures the graphics processing units (GPUs) and chips needed to power AI models.

Crucially, Nvidia’s chips are not the product – the AI services they enable are. But in a market where the cost of the hardware – not to mention the water and energy bills – massively outstrips the returns, how can this model stay afloat? With valuations becoming increasingly detached from traditional financial indicators like revenue, some commentators argue it’s not a question of if but how the AI bubble will burst.

In this instalment of Finextra’s Explainer series, we examine the frothy, interconnected market of AI and how, if at all, it might finally burst.

Oracle’s earnings: The canary in the coal mine?

One of the most significant needles hanging over the AI bubble is the financial performance of Oracle, a major American multinational technology company that provides the market with database software and cloud computing services.

Only three months ago, Oracle looked to be one of the biggest winners of the AI boom. In September 2025, its co-founder and chief technology officer (CTO), Larry Ellison, briefly became the richest person in the world – overtaking Elon Musk – when the company announced a leap in projected revenues, and the stocks spiked. In the months following, however, Oracle’s performance has tumbled. Its share price now sits 40% below September’s all-time high.

This kind of volatility is raising eyebrows on Wall Street, which is increasingly seeing Oracle as the potential weak link that could cause the entire AI market to implode. Last week, Oracle’s quarterly earnings were unveiled, $16.1 billion, which fell just short of the market’s expectations. This triggered an 11% slump in its share price, as investors became nervous about the Silicon Valley data giant’s long-term prospects in AI.  

One of the reasons for this anxiety is the fact that much of Oracle’s expected future revenues will come from buyers like OpenAI, the American AI research and deployment company behind the revolutionary large language model (LLM), ChatGPT. Indeed, Oracle has been borrowing billions to fund the expansion of its data centre capacity. To honour that debt, Oracle is betting on the likes of OpenAI paying for all the requisite data centre usage.

So, the big question on investors’ lips is: “Can Oracle rely on revenues from OpenAI?” If the bubble is to be contained, Oracle must answer with a convincing “yes”, by proving in the coming months that it can pay its bills. Unfortunately, OpenAI is considered one of the financially weakest AI providers on the market, having been – as yet – unable to make its flagship LLM product deliver a profit.

Private credit: The next financial crisis?

Underpinning speculation around how the AI bubble will burst is something far more concerning. Much of the borrowing that swirls about these American tech giants is sourced from the private credit markets – the potential epicentre of the next financial crisis. Indeed, with banks tapping their breaks after the 2008 crisis due to regulation, the private credit market has come to the fore, offering investors higher yields than public bonds and private equity more flexible financing.

The concern about the private credit market is that transparency is lacking and losses can more easily remain hidden; that loans are illiquid, and many borrowers are highly leveraged; that underwriting standards, due to competition, are more optimistic on cash-flow assumptions and extend fewer lender protections; and that, from a regulatory perspective, it lies beyond bank capital rules, stress tests, and so on.

All these factors bode poorly for a space that is slated to become a $5 trillion market by 2029.

The warning signs

Policymakers, investors, and risk managers must stay vigilant. One of the key warning signs that the market could be moving from a position of stress to a position of systemic peril is not defaults per se, but rather a delayed recognition of losses – with lenders modifying their loan terms so that borrowers don’t technically default. This would conceal credit deterioration, keep zombie companies on life support, and kick bloating losses into the future.

What markets want to hear, and what we’ve heard a lot of late, is the “it’s different this time” narrative. But the awkward truth is that confident promises from lenders – that defaults are staying low, that borrowers are resilient, or that covenants aren’t so important – often precede a crash.