Financial bubbles are often linked to phases of rapid innovation or technological progress that stimulate investment and can give rise to irrational market exuberance. The mechanisms are well understood: groundbreaking developments open up lucrative prospects, attracting huge inflows of capital against a backdrop of overly optimistic profitability expectations. This optimism, amplified by investor herd behaviour, pushes valuations higher to the point where they become disconnected from fundamentals. To maximise gains, increased use of debt and ever-bolder risk-taking also become commonplace. In other words, a financial bubble arises from excessive faith in future share price gains combined with self-reinforcing collective dynamics, making the entire edifice particularly vulnerable to any pivot in expectations.

Today, there is little doubt that artificial intelligence qualifies as a general-purpose technology (GPT) that is set to profoundly reshape our economies over the long term, just like the steam engine and electricity in their time. The big tech firms are investing huge amounts in AI, with a circular financing loop that brings together chip makers (Nvidia, AMD), cloud computing providers (Microsoft, Oracle) and Large Language Model (LLM) developers like OpenAI. These massive circular flows of hundreds of billions of dollars sustain the surge in tech share prices but expose markets to the risk of a severe correction should the promised profitability fail to materialise.

We’ve already seen such waves of financial speculation in the past – from railways and electrification to the internet – attracting vast amounts of capital to finance transformative technologies whose returns were not guaranteed up front. 

According to the current market consensus, an AI bubble burst would be comparable to the dot com bust of the early 2000s, when equity-focused speculation – with minimal overspill into credit and scant exposure among banks – inflicted limited losses on exposed investors and generated a modest economic hit, far short of the systemic fallout of the 2008 financial crisis. The real danger lies in the formation of bubbles fuelled by excessive debt in an environment where risks are systematically underestimated and often obscured by complex interconnections, with poorly understood effects propagating risks towards the heart of the financial system, namely the banking sector.

From this perspective, debt financing of massive investment in AI could flow through channels outside the traditional banking system, relying on complex and sometimes opaque financial structures that reflect the growing sophistication of globalised finance. Tighter regulations since the 2008 crisis have favoured the growth of shadow banking, whereby non-banks (investment and debt funds, asset managers) directly or indirectly finance the economy by pooling capital from investors (pension funds, insurers or private investors) while remaining tightly interconnected with the banking system via credit facilities, derivatives, repos and various risk transfer mechanisms such as debt securitisation. Systemic risk persists, then, but in a more diffuse form, propagated by cross-exposures that make it harder to identify and less traceable. The recent bankruptcies of US firms Tricolor and First Brands highlight these vulnerabilities linked to the opacity of non-bank financing, associated in both cases with fraud or irregularities.

Technological and financial shocks thus appear interwoven, generating self-amplifying forces that are today driving gains but could tomorrow deepen losses. For the time being, the financial vulnerabilities we’ve been discussing are mainly confined to the United States. However, if these isolated cases were to spread, events in the United States – given its central role in global finance – could trigger widespread contagion, with domino effects likely to compromise global liquidity, intensify negative correlations between asset classes and, ultimately, threaten global financial stability.