Take shelter! The wave of AI mania is about to break

In 1925 the Soviet economist Nikolai Kondratieff published a study on prices, interest rates and output in Britain, France and the United States, stretching back two centuries. He discovered something of an anomaly. Retracing the data, he saw that there were long swings of roughly 50 years, from expansion to mania, then crisis and depression, and back to boom time, which repeated themselves. From the perspective of a Marxist, this was equivalent to heresy; capitalism was meant to be about to collapse, not merely taking a breath and renewing itself. Kondratieff was arrested in 1930, and Stalin had him shot in 1938. Yet, his key observation on the waves of capitalism – with boom turning to bust and back again – outlived him.It would be up to Joseph Schumpeter, the brilliant Austrian economist who coined the phrase creative destruction, to give these waves Kondratieff’s name and, in 1939, the mechanism by which they function. Each long wave, he would argue, was the result of a cluster of innovations that reorganise the entire economy: from steam and cotton mills in the 1780s; railways from the 1830s; oil, the motor car and mass production after World War 2; and more latterly the microprocessor and the internet from the 1970s. Every one of these technological transformations altered indelibly how people live and work. Each one also produced at least one cataclysmic financial collapse along the way.By incorporating the financial sector, the economist Carlota Perez subsequently explained why. Each technological revolution, she argued, passes through two critical phases. In the first phase speculative finance pours capital into the new infrastructure much faster than the real economy can productively absorb it. The phase ends in a mania and then a bust; canal mania in the 1790’s, the railway panic of 1847, 1929, the dot-com bust of 2000. This is known as the “installation” phase. Only after this does the “deployment” phase begin, which is when new technology is put to productive use. She argued therefore that the financial crash is not an accident, but a fundamental mechanism through which speculative capital hands the baton to productive capital.This explains why the “Is AI real” question is pointless. The questions we should be asking ourselves are, rather: Where are we are in the installation phase of AI mania? And is it about to end?The frenzy by the numbersJudging the numbers, the evidence seems to suggest that we are nearing the end of this phase. To start with the spending, it has gone stratospheric. Combined AI related capital expenditure by the major AI hyperscalers has roughly doubled, from about $450-billion-$500-billion last year to close to $1-trillion in 2026, according to Bloomberg data. This is not being funded by profits and cash flow; according to Capital Economics, the combined free cash flow of the four biggest hyperscalers is forecast to turn negative in 2027. These companies have more than doubled their bond issuance in a single year. Put simply, the richest companies in history are borrowing record sums to build capacity for revenue and demand that does not exist.Access to power drives data centre investment.While valuations are their highest since dot-com, according to Bloomberg data, there is also a particular circularity to the “earnings” of these companies that ought to worry anyone who lived through that bust of 2000 to 2002. Much of the earnings growth on one side of the income statement is simply another company’s capital expenditure. In many instances, that capex is being funded by equity investment or lending from the (usually semiconductor) vendor companies themselves. Chipmakers’ sales are the hyperscalers’ capex. If the eventual revenue does not appear from somewhere, the earnings bubble will burst, in both sets of income statements.Then there is the classic late-stage signal: that moment when insiders start selling to outsiders. This new wave of IPOs and share sales is perhaps the clearest warning yet that we have reached the peak. Historically, similar issuance booms have coincided with market peaks. Anthropic’s planned IPO this autumn might be exactly the kind of landmark listing that historians later treat as a marker, such as the infamous listing of pets.com in February 2000.The late cycle meets a hawkish FedA bubble needs a pin, and it seems as though the business cycle has already supplied one in the form of the Federal Reserve. On 16 September the Fed raised its benchmark rate by 25 basis points to 3.75-4% in a unanimous decision. This was its first increase since July 2023, with the new chairperson Kevin Warsh citing a strong labour market, stubborn inflation and the continuing energy shock from the Iran war morass.The market is now pricing in one more hike this year. The ECB had already started raising rates, the US 10-year bond yield is now firmly above 5%, its highest level since 2002, while oil is well above $100 and showing no signs of retreating. The macroeconomic context then all speaks to textbook late cycle dynamics; an economy at full employment, an energy shock, central banks tightening, long bond yields rising, and equity valuations supposedly off to the moon on a SpaceX rocket.The precedent is uncomfortable. Between June 1999 and May 2000 the Fed hiked rates from 4.75% to 6.5%, after which the S&P 500 lost half its value in roughly two-and-a-half years.The transmission mechanism is straightforward; capex funded by debt at rising rates, against cash flows turning negative, is capex that gets axed. On this reading one might expect the S&P 500 to start cracking next year, and ultimately to fall between 30-50% from its highs.The global economy is about to take another breathKondratieff’s critics have always argued that his waves are too irregular and imprecise to have any predictive power. This is missing the point. His insights are not useful for any accurate analytical reasoning, but the phenomena they describe: finance overbuilds, the price of capital rises, the frenzy breaks, but the innovations go on to change the world anyway. The question for investors is not whether AI will be transformative, it is whether they want to be holding the paper when the turning point hits. That could be sooner than many think. DM