Artificial Intelligence : When One Risk Hides Another

Hi everyone and welcome back to think macro for our 10th episode. It's just me today and I want to start by looking back to a year ago in our autumn podcasts in which we looked at the issues at stake in the artificial intelligence revolution. The challenge facing AI engine manufacturers, the hyperscalers, you know, the Microsoft, Google, Meta, Amazon, Oracle, they're challenged to monetize their enormous investments. That was a considerable challenge due to the colossals cost involved. The competition as well, both among themselves and with China. And above all, the difficulty of maintaining leadership in a world where it's easier to switch AI systems than to change car or telecom provider. As with all capital intensive tech revolutions, the first immediate winners were expected to be the component supplies, the big and trouble makers, the Samsung Enix, TSMC, ASMR, micron, whilst in the long term, businesses, consumers like UNI were expected to benefit from the tremendous productivity gains offered by AI. And in between manufacturers and hyperscalers found themselves in a tricky position when it came to eventually recouping their massive investments. At the time, the market was moving in the opposite direction, piling up all the ingredients of a speculative bubble. That raised fears that it might pose a systemic risk to the financial markets, firstly because of its sheer size, but also because US households, US savings were heavily exposed to it. So against the backdrop of sluing consumption and persistent inflation, that was a serious concern, one that could drag the country and possibly the world into recession. All the ingredients for a bubble were in place. On the one hand, you had the Federal Reserve and the government that were still accommodative, you had the former openly favoring growth over inflation and the latter that just had launched its third fiscal package in less than five years. So capital, which is the fuel of financial markets, was both cheap and readily available. Now, on the other hand, you had the extreme valuation of hyperscalers that assumed that high margins, we had seen in the previous decade in social medias or online advertising, could just be sustained indefinitely. The market anticipated an AI revolution without a struggle or casualties. The psychology as well was a course for concern. AI had all the ingredients of a powerful narrative, a new, fascinating technology that was difficult to grasp in which seemed to offer a historic opportunity to those who were able to understand it before others. So you simply could not afford to miss it. And how could one not believe these producers, these hyperscalers who had created so much value over the past 15 years when they claimed to be writing the next chapter of history and winning a global race. These are the finest companies in the world after all. And finally, possibly more subtly, why would the leaders of these companies remain objective when their new aim is just no longer to reward their investors through dividends or share buybacks, substantial profits, but to raise as much capital as possible as quickly as possible in order to cross the finish line before the others. The priority has shifted. The story then is presented in a more favorable light, a time-lapse or shortened, uncheckable profits can be put forward. Risks, competition, are glossed over. It's all about making people dream. And then a more technical factor further exacerbated this bubble risk in our view. The continued rise of passive investment, combined with regulations that make it difficult and costly to take positions opposite to the consensus, it's making the market increasingly inert and momentum driven. Most systematic funds operate on this principle. They are followed by a number of hedge funds and also by circle from the mental management firms which in reality simply go with the flow. And of course, you have index tracking funds that simply amplify the phenomenon by their very nature, as they're forced to buy the largest constituents of indeed index, the ones that are rising and to sell companies whose weight is decreasing, the ones that are falling. passive investment now accounts for nearly half of assets and the management in the United States. Combined with systematic funds and similar strategies, these mechanical investments account for the majority of market movements, more than three quarters in volatile years. This mechanization of investment amplifies both downward but also upward trends. This is enough to fuel a budding bubble. All the ingredients were in place and yet the market refused on each of these points to get carried away beyond what was reasonable or to do so sustainably. The risk of alpha bubble and behind it, the risk of contagion spreading to the wider economy and other countries has diminished significantly. We have avoided a repeat of the dot-com bubble or the sub-prime crisis. Challenges remain of course regarding the distribution of profits and the economic gains on AI as well as the reallocation of capital but the system is no longer at risk. Firstly and we were the first to be surprised, the market has stopped buying into the AI narrative without batting an eyelid. Last September when the hyperscalers exited the equity market to seek capital in the debt market, the reaction was immediate. In the city, it's often said that equity investors look up towards the stars and the associated profits. When bond investors look down, searching for hidden floors that could jeopardize the repayment of their loans, the perspective is different. When Meta or Oracle seek to borrow the question sourced on the longer the same, it's no longer about staggering profits but about risks. What about Chinese competition and deadlines? You're all racing against each other and you all think you'll win but there are bound to be losers. The credit market immediately sends that uncertainty and the risk premium, you have to attach to it. It refused to buy into the narrative in questioningly and demanded a higher price. The Big Five's borrowing cost has risen sharply since then and their stock market performance has stagnated. This credit pressure has sung doubt in market sentiment. The words of the Big Five CEOs are no longer taken at face value. They're being challenged and subject to scrutiny. The significant proportion of circularity in profits which we estimated at 85% at the time has been brought to light. Profits boosted by capital gains or contingent liabilities are no longer regarded as equivalent to cash sales, the market no longer looks surly at the explosion in revenue, that is real, but at the ability to convert it into profits. It is asking the right questions and returning to the reality of figures, it wants concrete results. Yet we had one final scare. The reallocation away from hyperscalers had gained so much momentum, such inertia that as summer approach the bubble had shifted, it moved to suppliers. So the price of these suppliers and the pickage of all makers had literally skyrocketed, fueled by systematic funds, by retail speculators, by passive investment managers. You're talking 2, 3, 500% in less than a year. So again, excesses, fears of a bubble, but again, that was quickly corrected. And due to somewhat unexpected mechanism, the one that comes with the way indices are built. The explosive performance of suppliers had increased their weighting in the indices to such an extent that it forced many investors to sell in order to rebalance their portfolios, often under the constraint of concentration limits imposed on them. So this resulting mechanical sell-off, which at some point even twigged some panic, in July brought the price of these suppliers to far more reasonable levels. And actually even today you can even argue that the companies in question are trading below their profit growth rates, they're cheaper than they have ever been. And on top, unlike previous cycles, they have secured their order books for several years, which is new in these very cyclical companies. There's been a great deal of volatility, but ultimately, once again, a bubble has been effectively averted. And finally, there was one last source of excess authorities, the US government, and the Fed, the policy of the zero-rate policy, the QE infinity launched in 2012, or the massive whatever it takes, fiscal stimulus package of 2020. And again, in 2023, played a major part in driving assets price up by making capital cheap and available in unlimited quantities. It is estimated that nearly half of the liquidity injected during the COVID crisis was invested in the financial markets. There's no better fuel for a bubble than free capital. There were fears with the Fed as well through its resolutely growth-oriented policy, all for the US government to decide to win back the hearts of its disillusioned voters might also be fueling the emerging bubble with their power and their excesses. And once again, not the case, the government has merely implemented the fiscal measures past last year, and the Fed has recently adopted a tougher stance on inflation. From three rate cuts in February, the market is now six months later, expecting three rises. Internal interest rates have risen by nearly one percent on average in the United States. Besides, the new governor, Kevin Walsh, goes even further by scrapping forward guidance. This communication tool designed to guide the market on the future path of interest rates is reintroducing uncertainty into the market. And the result is reducing systemic risk by seizing to do the work for investors, for us. The Fed is forcing us to carry our own analysis instead of a single line of thinking that everyone follows, we will most likely have as many viewpoints as there are analysis. Admittedly, there will be more volatility short-term, there will be more noise, but consensus excesses will be less frequent, and passive fund managers will find fewer trends on which to rely to trigger such excesses. This uncertainty paradox is a powerful mitigator of systemic risk. By introducing short-term noise, we reduce the risk of a sustained and pronounced divergence from fundamentals, which could end violently, as demonstrated by the subprime excesses of 2008, or more recently, the Fed's error in 2022, which neither the market nor other central banks dare to challenge at the time. So, as the summer drew to a close, equity and bond indices showed little movement and even exchange rates remained fairly stable, but the market's risk profile looks much healthier. It has perched itself from its excesses regarding hyperscalers, its complacency over interest rate, and more recently this summer, its excessive euphoria regarding supplies. At the same time, the global economy is showing a degree of resilience. Consumers, businesses have absorbed the shock of energy prices surprisingly well. An employment is falling again, and business sentiment is broadly optimistic across the world. So, that's good news for the world, which is both seeing the risk of US contagion fade and its fundamentals improve. In contrast, in the United States, the economic problem has just been postponed. Whilst households are seeing their real incomes rooted by inflation, the economy is only staying afloat thanks to a wave of investment in AI. At the cost of mounting risks, a productive capital expenditure continues to rise, but that is largely due to rising costs. These includes, of course, the cost of components and infrastructure, which supplies of regularly raising prices, but also financing costs, driven by rising bond yields, but also credit spreads. This expenditure is popping up the economy in the short term, yes, but is weighing on future margins. Above all, by persisting in the AI race, whatever the costs, the hyperscalers are now having to resort to debt. Their huge profits are no longer enough. They must now raise capital in the credit market, and given the sums required, they're doing so swiftly and on a massive scale. Just this year, in 2026, it's $400 billion of new Asians from AI-related debt. Probably almost twice that next year. That's going to come bigger than the government's net debt issuance next year. It's just enormous. Of course, this has many implications, particularly for the future of the US economy. First, we knew that the link between the economy and AI was strong is going to get even bigger with all the risks. This leverage makes growth increasingly dependent on the future of AI profits and potential productivity gains. Any disappointment or delay will have significant macro consequences. Second, more of a side effect. This rapid reliance on debt is putting further pressure on the interest rates. You competing with governments, and they are already under pressure. If you're talking about the most indebted countries, France, the United States, the United Kingdom, it's already very difficult for them to convince investors to lend to them. Now, if they come to compete with Microsoft of Google or Google, of course, investors will have an easy choice. Finally, this false accumulation of debt re-exposes hyperscalers to interest rates. In the past, because they had a master in enormous profits over the years, they were well positioned to benefit from rising rates. That, for instance, allowed them to invest their cash reserves profitably. Now, in less than two years, they will become debtors, and they will therefore be exposed to falling rates. From being receivers, they become payers. And this is crucial, because the US economy drew significant strength from these hyperscalers when they were net receivers. They were able to help stabilize the economy when interest rates and inflation weighed on hassles thanks to their strong profits, also thanks to their stock market performance which boosted hassles wealth. They provided a strong kind of cyclical force in this K-shaped economy. As AI producers as hyperscalers take on debt, this counterbalance disappears. So after the government and hassles, it's now the entire economy that is becoming rapidly fully dependent on interest rates. For decades, cheap capital had fueled the US economy whilst the balanced provided by hyperscalers gave it its resilience. Now with these hyperscalers accumulating debt, the US economy is in the process of losing both. To conclude, equity markets have quickly crossed the scale of the AI revolution. It's winners and its future risks. They've not fallen into the trap set by hyperscalers. The bubble has been averted. The risk of contagion akin to the dot-com bubble or the supreme crisis has disappeared. That's good news for the global economy which should demonstrate its momentum in the coming quarters. In the United States, however, this remains a wave of investment that is veering towards success and rapidly exposing the entire economy to interest rates that are already in the pressure. The pressure is intense and becoming widespread and given investors' reaction, quick reaction, the risk is rising probably more rapidly than the market expects and it will most likely slow down the economy before the trillions of dollars and ounce have been spent. Now with the highest starting point, more leverage and the less robust economy, this will no longer be a mere slowdown. The US has managed to prolong the cycle once again, but it's just postponing the inevitable adjustment it needs and that will be all the more painful for it. So that's it for today. Thank you for tuning into Think Macro. If you enjoyed this episode, don't forget to subscribe and see you soon.