Concerns about a potential AI bubble emerged in late January 2025 following the unexpectedly successful launch of the Chinese-made chatbot DeepSeek. These anxieties intensified on 18 November 2025, when a nearly 500-point drop in the Dow heightened investor unease. In late 2025, 30% of the United States (US) S&P 500 and 20% of the MSCI World index were held solely by the five largest companies, and share valuations were reportedly the most stretched since the dot-com bubble.
The late 1990s dot-com bubble was fuelled by investor optimism about the internet, leading to large capital flows into unproven tech startups and a sixfold rise in the Nasdaq Composite. When expectations didn’t produce sustainable revenues, the market collapsed, losing about 75% of Nasdaq’s value by 2002 and roughly $5 trillion in investor wealth.
Experts warn that the current cycle on Wall Street appears to be “very similar” to that seen before the dotcom bust in 1998 and 1999. It has been argued by analysts that the current AI boom may surpass the internet bubble of the 1990s, with the 10 largest companies in the S&P 500 now more overvalued relative to fundamentals than at the peak of the dot-com era.
The current AI boom has been acknowledged as an AI bubble by several big business leaders, such as the CEOs of Amazon, JPMorgan Chase, Google and OpenAI. In particular, Sam Altman, CEO of OpenAI, drew a parallel between today’s AI frenzy and the 1990s dotcom bubble, when internet company valuations spiked dramatically before crashing. The Bank of England and the International Monetary Fund have also voiced alarm about the stock market bubble building around AI, warning of a crash comparable to the dot-com bubble.
In contrast, Wall Street analysts remain confident that the AI boom has further to expand, with the AI revolution driving the tech bull market for at least the next two to three years. Compared to the dot-com era, this is seen as a 1996-like moment with much greater scope for growth, rather than a 1999 one. From a Wall Street perspective, there is no AI bubble yet, as tech companies at the heart of the AI boom are posting massive profits, have solid business models, and now have much stronger margin profiles than during the dot-com bubble.
However, there is general awareness that investors may be overexcited and overestimate the level of certainty, leading to a significant drop in stock prices over the next two years and losses for some investors and vendors. Overall, there is optimism about the progress of and long-term value created by AI, which will eventually outweigh short-term losses.
As investments in AI continue to grow, with spending from US mega caps projected to hit $1.1 trillion between 2026 and 2029, and total AI expenditures expected to exceed $1.6 trillion, warnings arise that current levels of investment in AI are “very similar” to the dot-com bubble. There is a tangible fear that substantial sums invested in AI may not deliver the expected results, leading to underwhelming productivity gains for the wider economy, with warning signs already appearing.
In the US, AI adoption among larger companies is reported to have already peaked, is stagnating, or even declining. Most businesses do not see a substantial return or productivity gains from their investments in generative AI tools. Software company Atlassian found that 96 per cent of companies did not achieve notable productivity improvements, and an MIT report stated, “despite US$30–40bn in enterprise investment into Gen[erative]AI, […] 95% of organisations are getting zero return.”
It also appears that AI will not widely replace human workers, at least not in the near future. Since the rollout of ChatGPT, no noticeable disruption has been observed in the labour market. Additionally, it has been estimated that AI-driven productivity will not bring about a GDP miracle but will result in only a modest increase of between 1.1 and 1.6 per cent in the US over the next 10 years.
While tech companies are now pushing for larger AI models, this scaling does not correlate with proportional gains or with artificial general intelligence (AGI), as AI researchers seem to agree. The economic underpinnings of this scaling approach are beginning to show strain, as the tech industry is projected to be about $800 billion short of the revenue needed to cover the costs of data centres and chips, which are estimated to cost $2 trillion in combined annual revenue.
Companies such as Amazon, Google, Microsoft, Meta, and Oracle have already used a record 60 per cent of their operating cash flow on capital expenditures, like data centres and chips, as of June 2025. Tech companies are reportedly resorting to “creative finance” such as circular financing deals to continue raising money for data centres and chips. Such financing models, which create complex structures that blur the lines between risks and liabilities, have traditionally been associated with bubbles.
As data centres grow, their hidden costs, particularly electricity and water for cooling, become clearer. Data centres can use as much power as entire cities, with demand expected to rise by 160% by 2030. In the US, they are the fastest-growing power consumers, and utility companies plan to invest $1.4 trillion in infrastructure by 2030, which will raise rates for all ratepayers, including households. Since half of the increased power demand will likely be met by natural gas, this has significant environmental consequences.
A burst of the AI bubble would not only affect financial institutions and venture capitalists but also wipe out $20tn in wealth owned by American households and $15tn held by investors in the rest of the world. Interestingly, it has been argued that China would not be equally vulnerable to a crash, due to its pragmatic approach to AI-centred policy and investment in integrating AI into the real economy rather than developing frontier models for their own sake.
In conclusion, a burst of the AI bubble could trigger significant global economic disruption, echoing past financial crises. Yet an extended bubble may prove equally damaging, with mounting financial and environmental costs. Either scenario suggests that the current AI growth model carries hidden risks, leaving the global economy caught between a rock and a hard place.
