Will the artificial intelligence bubble burst? The question mark adds to the dilemmas that the crisis in the world order is fuelling with debt, rearmament, protectionism, wars, and illusory peace agreements. After many years in which stock markets have weathered persistent headwinds, becoming the emblem of capital’s resilience to its own crises, the accumulated risks are coming to the surface. High technology, the newest sector of listed companies, has become a magnet for investment, producing in just a few years giants with unprecedented market values.
The Magnificent Seven
in the sector (Alphabet, Meta, Microsoft, Amazon, Apple, Nvidia, Tesla) have a combined market capitalisation of around $22 trillion, about 35% of the total of the 500 largest US corporations. The artificial intelligence (AI) champion Nvidia alone has a market capitalisation of $5 trillion. In November, when third-quarter corporate financial statements were presented, uncertainties about capital expenditure, profitability, debt, chip obsolescence times, and the energy costs of these technologies caused significant shocks in the sector’s share prices. This bubble, like all those that preceded it, will see alternating euphoria and panic. Once again, we will witness the power and drama of capital.
Two bubbles
According to the Financial Times, since OpenAI launched ChatGPT in November 2022, AI-related companies have increased their stock market value by 165%, compared to an overall increase of 70% in the S&P 500 index. A comparison with the dot-com bubble shows that, in the last three years of the bubble, from December 1996 to March 2000 when it burst, the S&P 500 general index grew by 110%, while the NASDAQ index of technology stocks tripled its capitalisation in the last 18 months of the bubble [William Quinn and John D. Turner, Boom and Bust, Cambridge University Press, 2020]. The cyclically adjusted price-to-earnings ratio of the S&P 500 index, dear to scholars of stock market ups and downs, reached 45 in the last phase of the dot-com bubble (i.e., the average company on the Wall Street stock exchange was valued at 45 years of its earnings). Today it stands at 38.
The diversity of the geopolitical context, technological content, and economic power does not allow for easy comparisons between the two bubbles, but today’s bubble appears to be in a more lukewarm
phase than the dot-com bubble. However, it should be noted that markets are not compartmentalised. In a recent study, the Federal Reserve drew attention to the financial indebtedness of hedge funds, which has reached a record level of $6.2 trillion, up more than 25% from a year ago. At the same time, large banks have lent $1.7 trillion to non-bank financial businesses. Some of this flood of money will inevitably end up in the AI bubble. However, there are other sectors and activities that demand real or fictitious capital, not least States that need to finance their debts and the arms industry.
A sign of vulnerability to external collisions
came in January, when Chinese start-up DeepSeek launched its own AI model at a much lower cost than that of OpenAI and Nvidia’s processors. The latter, in a single day of panic, lost $590 billion in market capitalisation. But that did not stop investment flows to the magnificent
seven giants. Morgan Stanley estimates that between 2025 and 2028, big businesses will spend around $3 trillion on building AI data centres.
So far, the AI groups’ investments have been largely self-financed by the growth of their share capital and cash flows. In the four-year period indicated above, more than half of the expenditure will come from external financing, $1.15 trillion from private credit and $350 billion from bond issues and securitisation. The Wall Street Journal describes a circular
financing model among these groups, illustrated as follows: Oracle buys Nvidia chips, Nvidia commits $100 billion to OpenAI, OpenAI has long-term purchase commitments from Oracle for $300 billion. These circuits are reminiscent of the suspect relationships with which banks, non-bank financial groups, and special-purpose vehicles created an artificial market to fix the prices of toxic securitised products during the 2005-2007 real estate bubble.
Banks, builders, data centres, and turbines
Large spending and debt plans disrupted the market in November, but — writes the WSJ — any concerns on Wall Street about a possible investment bubble have largely been trumped by the fear of being left behind
. Virtually every Wall Street player is angling to get a piece of the action, from banks such as JPMorgan Chase and Morgan Stanley to traditional asset managers such as BlackRock
. The Wall Street Journal rediscovers a side of stock market psychology: the desire to squeeze every last drop out of the bull market. In 2007, the then-CEO of Citigroup Chuck Prince theorised: As long as the music is playing, you’ve got to get up and dance.
Less well-known but cash-rich companies are entering the fray, fund managers such as Blue Owl have amassed trillions of dollars of investing firepower and have been hunting for big deals where they can put that money to work
. Blue Owl is building eight data centre buildings in Abilene, on the border of West Texas, the epicentre of oil fracking, which will consume 1.2 GW of energy, enough to power about one million homes. Oracle will lease the centres for fifteen years. The construction of these infrastructures and the energy plants that will power them are the main facts supporting the thesis that what is underway is an industrial bubble
, and not just a speculative one.
The enthusiasm that inflates the bubble shines through in announcements and forecasts. Morgan Stanley signed $75 billion worth of deals in one week to finance data centres. Google, Amazon, Meta, and Microsoft have spent $350 billion on data centres this year and plan to spend another $400 billion in 2026. The Financial Times has put the surge in gas turbine production under the microscope. In connection with AI, gas turbines in the US have been revitalised after a period of stagnation during the boom in renewable energy. According to consulting firm Dora Partners, the sector received 1,025 orders in 2025, including 183 for large turbines, an increase of 50% compared to the average over the last five years. The Department of Energy predicts that data centres will consume 6.7-12% of US electricity by 2028, compared to 4.4% in 2023.
This phenomenon is not unique to the United States: the IEA predicts that global electricity consumption by data centres will double by 2030, reaching 945 TWh, more than Japan’s current energy consumption. The three big groups in the turbine sector in the US (America’s GE Vernova, Japan’s MHI-Mitsubishi, and Germany’s Siemens Energy), which were in a severe demand crisis a few years ago, are now in a supply crisis, with orders for the next three years. According to the London newspaper, this bottleneck in the supply of American and Japanese turbines is worrying emerging Asian countries, which are facing growing demand for electricity. This could also be an opportunity for China to exploit.
US-EU technology gap
The battle for artificial intelligence has an uncertain timeline; it is conceivable that it will have applications in every field of human creative and destructive activity, but essentially it is a battle for productivity and a battle for capital. The US capital market is once again demonstrating its power. Nevertheless, the fact that, after the first black
day in November, one of the most active technology companies called for government intervention is indicative of how tough the battle is.
OpenAI’s chief financial officer, Sarah Friar, called for the government to play a role, complementing an ecosystem of banks [and] private equity
, as a backstop, a guarantee that allows the financing to happen
. The White House’s artificial intelligence czar, David Sacks, immediately ruled out bailouts for AI companies: The US has at least five major frontier model companies developing cutting-edge AI models. If one fails, others will take its place
. This cynicism is part of the strength of American capitalism, but it has never prevented the government from intervening.
In his report on competitiveness, Mario Draghi denounced the innovation gap
between the European Union and its competitors, the US and China above all
. Europe did not capitalise on the first digital revolution driven by the internet and is now also lagging behind in revolutionary digital technologies. About 70% of basic artificial intelligence models have been developed in the US, and just three US 'hyperscalers' account for over 65% of the global and European cloud market. [...] Quantum computing is set to be the next big innovation, but five of the world’s top ten technology companies in terms of quantum investment are based in the US and four in China (none in the EU)
.
For now, Europe will be watching the battle for AI only from afar.