The potential and the froth within the bubble

2025-10-25

Macroeconomic indicators are largely positive, recession has been avoided, and yet hardly anyone at the recent IMF summit in Washington DC was expressing calm confidence over prospects. There was a feeling that the state of affairs is not normal, and that something is about to break.

No one doubts that AI valuations represent an investment bubble. In the sober language of the IMF, its latest Financial Stability report described valuations as ‘stretched’. It also referred to risks with high levels of government debt and private credit. The IMF’s growth-at-risk framework shows that risks to global financial stability are ‘elevated’.

A positive take on the AI bubble is that such over-investment is a perfectly healthy feature when there is a major technological breakthrough, and that the value added by the few who survive and thrive will more than outweigh the temporary losses from those which fail. The productivity gains across the economy from the rapid developments in AI are set to be transformative, and it is necessary to invest in many startups as it is not possible in advance to anticipate which will become the unicorns. 

This is a credible argument, but there are trends both within the AI investment patterns, and a parallel trend in private credit, that hold the potential for wider, and bigger, negative economic shocks. 

First of all, the scale of the over-investment is unprecedented. Ten AI start-ups have seen their valuation rise by nearly $1 trillion in 2025. They are all loss-making. Venture capital investment in dotcoms at the start of the century was $10.5 billion, or $20 billion allowing for inflation. The equivalent figure for AI is $200 billion. An AI start-up with a $5 million turnover is valued at $500 million.

Secondly, there is the circular nature of much of the investment. Recent deals announced by OpenAI constitute an example. In September the company – one of the ten loss-makers and the developer of ChatGPT – announced a partnership with Oracle, in which it would pay the older tech company over $300 billion to expand its data center capacity as part of the ambition to create super-intelligence. OpenAI has also confirmed a deal with the chip maker AMD to buy $90 billion worth of its AI graphics processing units, and was granted warrants to own 160 million AMD shares, giving it a potential 10% ownership. AMD shares rose nearly 24% on the news.

OpenAI does not have the capital or recurring revenue for these commitments, but it has attracted investment of up to $100 billion by Nvidia, another specialist chipmaker for the AI industry. OpenAI will deploy 10 GW of AI data centers using Nvidia systems as part of the partnership.

Oracle, AMD and Nvidia are all publicly quoted, so if the privately owned OpenAI fails to achieve its ambitions for revenue and profits, there is a risk of stock market contagion.

A third risk factor underlying the bubble is the exposure of retail investors, which means that a crash could have direct and negative impacts on consumer confidence and demand – this point was made at a conference session I attended in Washington. Unsophisticated investors tend to be naïve and pro-cyclical, driven by a ‘fear of missing out’ (FOMO). Compared with historic bubbles and crashes, it is easy for the ordinary citizen, anywhere in the world, to invest in the US stock market. You can do so on an app, exactly like making a bet on a sports betting app. 

Linked to the amount of retail investing is the extent of leverage in exchange traded funds (ETFs), which tends to exacerbate volatility. The leverage can be up to five times, and ETFs now have extensive coverage. This risk was also discussed in Washington.

Fourthly, there are indications of a speculative bubble in crypto assets. This is fuelled partly by the Presidency of Donald Trump. He is unapologetically pro-cyclical as regards both crypto and the stock market, and will most likely loosen policy to shore up asset prices. This is a boon in the short term, but also a risk factor.

Finally, there is the unquantifiable, parallel risk emerging in private credit. As with the circularity of AI investments, there is complexity and opacity. The collapses of the companies First Brands and Tricolor, linked to debt-based funding and subprime loans, resulted in losses for mainstream banks. The two cases are described as ‘idiosyncratic’, and they are, but the opacity of private credit makes it impossible to know how many other high-risk idiosyncratic situations there are. 

As for the banks, the largest US banks are well capitalized, but there is concern about regional and the smaller national banks. In mid-October Zions’ stock fell 13% on news of a $50 million loss on two loans in its California division, and Western Alliance fell 11% after alleging fraud by Cantor Group over a bad loan of $100 million. Stock markets fell as a consequence, led by banking stocks.

Looking at behavior as well as the data, there have been lines of people wanting to buy physical gold and its price has risen above $4,000 per ounce. This does not happen when there is confidence in paper money, the stock market and the wider economy.

/ends