Learning from the past: What the
Dot-com bubble and beyond can tell
us about the AI Boom today 

Imagine, if you will, that you were a Dutch tulip farmer in the early 17th century. It would likely have been a peaceful if unexciting venture, cultivating flowers for a small niche of wealthy buyers. But starting in 1634, things would have begun to change for you. Almost inexplicably, an avalanche of demand for these flowers was unleashed, which pushed up prices almost two-hundred times. Traders, merchants and materialistic aristocrats were all jostling for a piece of the pie. It would undeniably have been surreal to witness such an immense boom. But in a few years, this would all come crashing down. In 1637, as inexplicably as it arose, so did the Dutch ‘tulip-mania’ evaporate. It is still unclear what drove merchants to become so fixated on the practice, nor what suddenly deterred them just a few years later. But this story does provide us one of the most illustrative examples of what we now understand as an ‘Asset Bubble’. Hype, speculation, and ‘circular spending’ - where agents fuel an unsustainable rise in prices by exchanging cash between themselves in spite of a dearth of genuine demand on either side - were key to Tulip overvaluation. And when this overvaluation was realised in the market, the ‘bubble’ burst quickly, and all at once. 

 

There is little evidence that the Tulip bubble had a significant impact on the 16th century Dutch economy - Tulip farming was hardly a systemic industry. But in the 21st century, this disconnect between traded assets and the wider economy is no longer there. Nowhere is this as potentially potent as in the current AI Boom. Driven by widespread adoption of LLM usage by both individuals and firms, AI-exposed firms such as Amazon and Google have seen exponential growth over the last 3 years. According to Reuters, market capitalisations of AI-exposed firms have increased by an average of more than 200% in the last four years. Central to this has been the ripping success of chip company Nvidia, who recently became the first four-trillion dollar company. Is all this cause for celebration? Perhaps not. Because from economists, to leading figures in tech, to punters on the stock market, everyone and their dog seems to reckon that this ‘boom’ is shaping up to be the mother of all bubbles. And with almost 50% of recent US economic growth being driven by AI expansion according to the FT, the implications of an AI crash could be destructive.

 

Determining whether a market boom is proof of a bubble is easier said than done. After all, if bubbles could be reliably predicted, they wouldn’t occur in the first place! But there are a few useful indicators. Firstly, the presence of excessive hype in the industry is a sign that related assets may be overvalued. Undeniably, there is evidence that this has been a consistent presence in the AI industry in the last few years. Specifically, excitement over Artificial General Intelligence (AGI) has been brought up by investors and industry leaders as a reason to be bullish on AI, with billions of speculative investment being directly targeted towards this goal. However, with prominent figures such as ex-Tesla AI Chief Andrej Karpathy suggesting that AGI is more than a decade away, there is good reason to believe that investors have overestimated the proximity of AGI. Secondly, a considerable presence of circular spending is a likely sign that the current boom could be a bubble. On this mark, there is further cause for concern. JPMorgan’s Asset Management division has identified that much of AI growth has been fuelled by two-way cash flows between chip firms and model developers in the form of investment and capital spending, with ‘suppliers, customers and investors’ overlapping. 

 

Perhaps the most direct parallel that can be drawn to this situation is that of the Dot-com bubble of the late 1990s, which popped in the year 2000. The Internet took on a very similar role as AI is doing in sparking investor excitement, with a flurry of new ‘Dot com’ companies being registered to take advantage of the craze. But circular spending and excessive dilution leading to asymmetric information for investors meant the surge in the stock market, which itself helped spur economic growth, wasn’t to last. The Dot-com crash led to companies such as Amazon losing 80% of their market capitalisation, which had the knock-on effect of 1.7 million jobs being shed amid a recession in 2001. Considering the similarities in the movement of stocks between 2000 and 2025, it is not unreasonable to believe that we are heading down the same path now. This is particularly alarming when considering that a potential recession triggered by market corrections now would be many scales larger than that of the early 2000s.

 

Yet, while the similarities between the situation in 2000 and 2025 may be important in understanding what we face today, the differences are in some ways just as crucial. Firstly, the Dot-com crash was immediately signalled by high equity P/E ratios - a measure of financial health calculated as Current Shares Price / Earnings per Share - and rising interest rates. This is not observed at the same level in the short-term in the United States, where most AI-related companies are listed. Indeed, the Fed recently cut rates amid slowing economic growth, while the P/E ratio of firms today remains lower than they were in 2000, albeit still rising. AI Volatility notwithstanding, the bubble bursting may not be as imminent as some believe. And what of the wider economic picture? Though the Dot-com bubble did lead to a shallow recession, the long-term productivity gains, as noted by the FT, helped improve output and revenues considerably in the two decades since. By contrast, though AI is expected to achieve productivity gains also, it is believed by bodies such as the St Louis Fed that this will be accompanied by a deep rise in long-run structural employment as more jobs are automated. This could damage government receipts through derived unemployment benefits more than productivity gains improve them.

 

Nothing in the current global markets, or indeed the current economic picture, is certain. But on the balance of odds, we can accept that there is a large chance the current AI Boom will end in the burst of a bubble. The similarities in its manifestation to that of the Dot-com bubble may provide us with a framework of what to expect. But the uniqueness of the AI Bubble leaves us much to be worried about. A lack of synonymous financial indicators can make it difficult to anticipate when the penny will drop, and the destabilising effects of AI on the job market means that the economic costs of a crash could be felt by more families for far longer than that of 2000, or indeed any prior bubble. In an ideal world, governments would be ready with potential regulations, and large fiscal stockpiles, to deal with this potential fallout. This is likely not the case. So while caution and preparation seem the optimal path, achieving this may be as difficult as predicting the timing and scale of a crash in the first place. All that leaves us to do for now is to watch and wait.