A couple of points are worth noting. First, assessing a bubble has more to do with the valuations than the assets. The assets might ultimately prove profitable and widely used in the global economy, but if the price paid for them is excessive, this can result in poor returns on invested capital. Second, the value of a financial asset is nothing more than the sum of its future expected cash flows discounted to today at an appropriate rate. Those estimated cash flows are, in theory, an average of different scenarios weighted by their probabilities. If the best-case scenario gets assigned a probability that is too high, then the valuation might be overstated despite such a scenario being a plausible one.
Another important factor is how buyers are funding their assets. Going back to the point regarding poor returns on invested capital, if those returns are meant to be used exclusively to fund shareholder dividends, then the timeframe for achieving those returns might be extended (provided shareholders have the patience). If there are lenders involved, however, the situation is quite different. We, as bondholders and other lenders such as banks, are not flexible at all with our investments. Therefore, if the returns on invested capital are insufficient to service debt, then shareholders might lose their invested capital as lenders seize assets to try to recoup the money lent. How much leverage is too much is an open question, but what we have witnessed is an increase in debt being raised to fund these investments. We have also noted new avenues to raise debt, such as special purpose vehicles (SPVs), which are sometimes partly owned by tech companies. These SPVs own certain assets relevant to the tech companies and then lease these assets for their use. To fund part of their investment in the assets, these SPVs can issue debt; however, this would not be reflected on the balance sheets of tech companies.
As 2026 approaches, we have begun compiling lists of things that can go well and those that may go wrong next year. One item on the latter is a correction in equity markets, led by fears that some AI investments may not deliver the cash flows that equity and increasingly debt investors are expecting. The repercussions of a correction in equity markets could be severe in theory, but at this stage, while some contagion is inevitable, we believe the impact on credit spreads should be contained. Companies in our fixed income investment universe with direct exposure to AI are generally very large investment grade issuers, with exposure among high yield issuers being quite limited. The impact on the real economy of AI investments slowing down a bit is difficult to estimate, but we also tend to think would not be too severe. Sentiment would certainly be squeezed, but for all the massive numbers in the press about AI investments, it is worth noting that the deals being announced are typically multi-year contracts and the Investment component of US GDP is about US$6.3 trillion per year ($8.2 trillion); Goldman Sachs estimated last September that in the official GDP data, AI has boosted US real GDP by just 0.2%.
A correction in valuations would be far from unexpected after the price action we’ve witnessed in the last few quarters in the tech sector. Given the direct exposure in traditional credit markets appears to be low and the impact on GDP is also not very large, we tend to think that if there is a sell-off which bleeds into credit, that might be a nice buying opportunity. Given valuations in government bonds incorporate several rate cuts already, it seems that a rally in rates in this scenario might be limited, but of course, safe havens usually benefit when risk assets are sold. Time will tell whether this scenario materialises, but predictions of a correction are increasing, and the “greed vs fear” index is edging further into fear territory.
See also: Singapore embraces AI, but workforce sentiment remains a challenge
Felipe Villarroel is partner, portfolio management at TwentyFour Asset Management
