More bubble signs
Leverage, massive investment and rising debt costs
The rapid rise and fall of Situational Awareness, a hedge fund run by a 24-year old that managed to accumulate more than $20bn in assets, is a tale as old as time. A boom begins in an asset class. Some investors use a lot of borrowed money to invest in that class and generate stellar returns. This persuades everyone that they are a genius, so they attract more assets and take on more debt. All goes well until the boom falters. Then they must sell the assets to meet their debts; if the investor is big enough (or the problem is widespread) the fire sale of assets drives down their price and makes the situation worse.
Back in the 18th century, the cycle was seen in Britain and France, with the South Sea and Mississippi bubbles respectively. It occurred in 1920s America when investors bought stocks on margin (only paying part of the price upfront) and of course it was behind the 2007-2008 financial crisis when NINJA borrowers (those with no income, job or assets) were allowed to buy property and elaborate financial structures were built on the assumption that US house prices would never fall.
This time round, the speculation has been built on stocks in the artificial intelligence industry. As in the late 1990s when everyone expected the internet to revolutionise the economy and invested in fibreoptic cables, this time round the corporate sector is investing in data centres and computer chips. In a feedback loop that George Soros dubbed “reflexivity”, confidence in the economic benefits of a boom leads to increased demand for the products made by the booming sector, which appears to validate the rationale for continued investment.
The scale of investment is staggering with Amazon, Google, Meta and Microsoft alone set to spend $1.5 trillion in 2026 and 2027. The tech companies are in an awkward position. If they don’t invest, they may lose their dominant market position (which generates their profits and boosts their share price). But investment on this scale starts to undermine their appeal as well; Google burned through $6bn of cash in the second quarter, the first period of negative cashflow since its flotation. As a result, the tech companies are having to fund more of their capital expenditure through debt; Morgan Stanley is forecasting $570bn of AI-related debt issuance this year.
This is having two effects in the bond market. First, it is causing investors to become more concerned about defaults; the cost of insuring against default by a “hyperscaler” (i.e big AI spender) has gone from being a third of the investment grade average to 80% above it, according to Societe Generale. Second, the scale of the debt issuance may be putting upward pressure on yields more generally, especially as the US government’s deficit, at $1.4 trillion in the first half, is still trending higher. Higher bond yields weigh on economic growth (and make the AI capex more expensive to finance).
All this may not matter, of course, if AI is such a transformative technology that it pays off for the hyperscalers and boosts economic growth. But the evidence to date is inconclusive. A Fed study of earnings transcripts found that, while companies were very optimistic about the productivity-improving potential of AI in the future, very few had actually seen any benefit up till now. Maybe they are right to be optimistic or maybe they have just bought into the hype.
Considering all the messianic talk of AI’s transformative powers, it hasn’t worked yet, except by boosting capex spending. Indeed, on an encouraging note, the trade expert Richard Baldwin finds, by looking at the export statistics for “other business services”, that jobs in this area, which looked vulnerable to AI, are booming.
So on the bubble front, we have signs of speculative trades going wrong, massive investment being funded by more expensive debt and a lack of evidence of AI’s effectiveness. If we haven’t passed the peak already, we must be close to it.