Is the American tech market currently in a speculative bubble driven by artificial intelligence (AI)? In this article, I analyze the mechanisms through which bubbles form and how one can anticipate the stage they are in, and when they might burst.
A financial bubble is generated by a strong surge in demand for a class of assets to the point where their value loses touch with reality, driven by speculation and collective hysteria, yet appearing purely rational from the perspective of each economic agent. Bubbles occur cyclically in the economy, and although each seems different (from tulip bulbs in the 17th century to dotcom stocks in 2000), the process of bubble formation follows a common and recognizable pattern.
A bubble begins when the financial valuation of an asset class rises sharply. This valuation usually has a real foundation and stems from an event or disruption so profound that it shakes the economic world as we know it. The disruption came with the launch and rapid adoption of ChatGPT in November 2022. This singular moment marked the revolution of AI as a powerful and general-purpose technology with transversal applications across all sectors, particularly in services and knowledge industries, which now represent over 70% of modern economies. Less than three years later, ChatGPT claims 800 million weekly users, with its competitors collectively matching that number. That means more than 1.5 billion people, nearly 20% of the world’s population, regularly use AI tools.
In the face of such disruption, no one can remain indifferent. Suddenly, every company must declare it is investing in AI tools; every major tech firm becomes an AI company, making massive investments; and every startup finds an AI angle, even when it has little to do with AI. This is the AI-crazed world we have lived in over the past two years.
Such a disruption creates major market winners. In a gold rush, the real winners are not the miners (only one in twenty will find gold) but the sellers of pickaxes and digging machines. In this case, the pickaxes are microprocessors, and the digging machines are data centers. Nvidia, which develops the best processors, has achieved billion-dollar revenues and margins, becoming the most valuable company in the world (worth over $4 trillion), while OpenAI, which launched ChatGPT, has become the world’s most valuable startup (already worth over $500 billion). Tech giants that own massive cloud data centers, such as Google, Microsoft, and Amazon, have seen their revenues and margins rise, while those without them (like Oracle) have decided to enter the market. Producers of processor-making equipment, such as ASML, are also thriving. Even energy suppliers that power data centers benefit, companies like EDP Renewables, capable of rapidly building solar power plants. And the biggest winners of all are AI specialists, with AI architects and programmers commanding compensation packages worth tens of millions of euros.
But is all this a bubble? Not if the disruption is truly real. Yet this is where collective hysteria comes into play. We know that the impact of technological revolutions tends to be overestimated in the short term and underestimated in the long term. That means AI will change the world in 20 years in ways impossible to foresee today, just as the internet changed the world in 20 years, though not in the ways predicted during the Dotcom era. Indeed, most of us still go to the supermarket for groceries, but a whole dimension of our lives has moved completely online. In the case of AI, there will be excessive investment in AI technologies and applications, which will disappoint in the short term until a new wave of startups, business processes, and models is created from scratch for an AI-driven world.
In the meantime, it becomes a race no one wants to lose. At first, valuations move due to results exceeding expectations for companies in the AI ecosystem. But as the bubble progresses, stock prices begin to move not based on results, but on inflated expectations. Being cautious with projections is not a winning strategy for companies that want to appear at the forefront of the AI race.
In the next phase, asset valuations are driven not by results or expectations but by the announcement of mega-investments and partnerships of completely unreasonable scale yet interpreted as signs of companies’ commitment to winning the race. Announcements that would normally alarm investors, such as hundreds of billions invested in physical assets with no clear return, companies investing millions in their own clients to finance purchases of their own products, or multibillion-dollar races among tech giants to achieve artificial general intelligence that no one fully understands or can value, are now celebrated. This is the phase we are currently experiencing, as recent announcements reveal. The notion of value begins to drift away from reality, and euphoria spreads to increasingly unrelated assets. Holders of privately valued assets (the world of private equity) start offloading them into the inflated public market to realize liquidity and gains. This is what we are witnessing today, with IPO requests reaching historic highs.
At this stage, all information inconsistent with the prevailing euphoria is ignored. The fact that Chinese companies are developing powerful AI with limited computing resources is ignored. The fact that many companies are struggling to monetize their AI investments is ignored. The fact that the proposed investment levels by AI firms lack economic rationality is ignored. Anyone who questions the dominant optimism is accused of failing to grasp the importance of the transformation underway. During this stage, we see high market volatility and sharp drops followed by recoveries to even higher levels.
The ingredients for a bubble are therefore in place, except for one that determines the bubble’s intensity and likelihood of bursting. That last factor is easy credit. When companies’ billion-dollar investments are financed by debt, and when investors buy inflated assets using borrowed money to profit from expected price increases, the perfect conditions for a speculative bubble to expand abnormally and burst dramatically are met.
Unfortunately, regulation in the American market is now so weak that we do not know how much of the current bubble is fueled by credit, though recent reports on the poor quality of regional bank loans and the historically low premium demanded for junk bonds are causes for concern. The reckless use of credit will determine how close we are to the end of the party, and how violent the final explosion will be. The party will typically end not due to an external shock but when the disappointing results of major AI companies outweigh the euphoria. The ensuing fall will be all the steeper the greater the use of credit to inflate the bubble, potentially even triggering a financial system crisis.
In that scenario, we will all bear the cost of the financial hangover and economic destruction. Yet we will be left with massive investments in processors and data centers that will last more than twenty years, and with AI engines that will transform the world in ways we cannot yet imagine.
Filipe Santos, Dean of CATÓLICA-LISBON