Today’s AI-driven market looks less like the dot-com bubble of the late 1990s and more like a market increasingly dependent on a small number of dominant technology companies.
At the turn of the millennium, investors allocated substantial capital to internet companies that often had no revenues or profits. In many cases, simply having ‘.com’ in the company name was enough to lift share prices.
When the bubble burst in 2000, the Nasdaq index declined by more than 75%, thousands of companies disappeared and investors lost significant amounts of capital.
Today, artificial intelligence raises similar questions. Shares of companies linked to AI grew strongly between 2023 and 2025, leading to comparisons with past speculative episodes.
At Amundi, we analysed this through data and found the picture appears more balanced than during the dot-com period.
Which companies are part of the AI ecosystem
Artificial intelligence is not a standalone sector. It extends across manufacturing, healthcare, financial services and technology. We assessed how closely individual stocks are linked to the broader AI ecosystem, from chip manufacturers and data centre operators to software developers.
The performance of this AI-related portfolio has been volatile. In 2022, while non-AI stocks declined by around 12%, the AI portfolio dropped by close to 50%.
This was followed by a strong rebound: AI-related stocks increased by nearly 88% in 2023 and a further 55% in 2024. Overall, AI-related equities gained approximately 115%, compared to 72% for the broader market and 83% for the S&P 500.
Bubble then and now: What is the key difference
There are clear similarities. As in 2000, a relatively small group of technology companies plays a major role in driving index performance. Expectations around technological transformation also remain strong. However, important differences emerge on closer inspection.
The most significant difference relates to profitability. In 1999, investors were paying for future expectations rather than existing earnings, with valuation multiples frequently exceeding a price-to-earnings ratio of 50x. By contrast, today’s leading AI-related companies – including Nvidia, Microsoft and Alphabet – generate substantial revenues and profits.
Earnings growth has also been strong. As a result, even though share prices increased, valuation multiples declined. The price-to-earnings ratio for the AI portfolio decreased from around 40x in 2023 to approximately 34x in 2025. This suggests the market is paying less for each unit of earnings than before.
Data analysis further supports this distinction. During the dot-com era, prices became increasingly detached from fundamentals. By contrast, current developments suggest AI-related stock prices remain more closely linked to company performance.
A more concentrated market than in the past
The overall weight of technology in the S&P 500 is lower today than at the peak of the dot-com era, when it exceeded 44%. Today, AI-related companies account for approximately 38% of the index. However, this exposure is now concentrated in a smaller number of firms – around 44 companies compared to nearly one hundred in 2000.
In practice, this means US equity market performance is increasingly influenced by a limited number of companies. Many investors gain exposure through index strategies without fully recognising the level of concentration embedded in their portfolios.
The key risk: Large investments that need to deliver
Another important factor is the scale of current investment. Companies such as Microsoft, Google and Amazon are investing hundreds of billions of dollars into AI infrastructure, including data centres and specialised hardware. These investments have accelerated significantly and are now more than double, relative to assets, compared to the rest of the market.
The risk would arise if these expenditures do not translate into corresponding earnings over time. If the ability of AI to generate profits were to slow, or if infrastructure proved excessive, market reactions could be significant.
At the same time, balance sheets provide a more reassuring signal. The ratio of debt to total capital for AI-related companies has declined from around 35% to below 25% over the past four years. This suggests companies are financing investments largely through internal cashflows rather than external borrowing.
A system built on a limited number of players
Speculative bubbles often follow a pattern of rapid price increases, detachment from fundamentals and subsequent correction. The current AI-driven market environment does not fully replicate this pattern, although some elements – such as concentration and strong expectations – remain.
Rising earnings, lower leverage and price dynamics that remain linked to fundamentals suggest markets are reflecting real technological and economic developments. At the same time, concentration in a limited number of companies represents a structural risk.
Whether significant investment in AI infrastructure will translate into sustained profitability remains an open question. Changes in the macroeconomic environment, including interest rates, could also affect valuations more broadly.
Thierry Roncalli is head of quantitative research at the Amundi Investment Institute. The views expressed above should not be taken as investment advice.