The massive investment frenzy sparked by billions of dollars poured into artificial intelligence is beginning to take its toll on the financial markets. Are we witnessing the collapse of the biggest technology bubble in history, or is this merely a temporary correction? Despite the technology’s groundbreaking nature, huge financial outlays are not yielding the expected returns. Economists and analysts are drawing parallels with the dot-com crisis at the turn of the century.
Anyone who has been following the performance of technology companies on the stock market recently may have felt uneasy. The share prices of major semiconductor manufacturers in Asia and the US have plummeted. Shares in South Korean giants SK Hynix and Samsung fell by 46% and 35% respectively, while the US SOXX index saw a correction of over 28% from its highs. Infrastructure suppliers have been hit particularly hard, with SanDisk’s share price falling by 58 per cent, Arm Holdings’ by 46 per cent and Bloom Energy’s by 55 per cent.
Do such drastic sell-offs signal the end of the AI boom, or are they simply a normal correction after an unprecedented period of growth? What do the 19th-century railway boom and the early internet crisis have to do with all this? And how is the software revolution colliding with the harsh reality of hardware limitations? Below, we present some facts, conclusions, and predictions regarding the future of artificial intelligence.
Huge investments, minimal returns
The value of AI-related companies has risen by 27 trillion dollars over the last three years, accounting for as much as 36 per cent of the value of the entire US stock market. The development of artificial intelligence is an enormous and undeniable technological breakthrough. However, given the level of interest, actual profits from this sector may fall short of investors’ unrealistic financial expectations. The International Monetary Fund explicitly warns that a sudden burst of this bubble would threaten global economic stability by curbing consumption and causing a collapse in lending.
The greatest threat to the sector remains the growing gap between expenditure and revenue. Analyst Ed Zitron’s calculations indicate that tech giants such as Microsoft, Meta, Amazon and Alphabet have invested around $560 billion in AI infrastructure over the past two years, generating just $35 billion in revenue from it. Furthermore, an MIT report reveals that up to 95 per cent of corporate pilot projects involving generative AI fail to deliver measurable business outcomes. OpenAI would need to generate $100 billion in free cash flow by 2030 to justify its valuation, while PitchBook analysts are forecasting a loss of between $10 billion and $30 billion in the same year.
The spectre of the dot-com bubble looms large. Why will this crisis be different?
While comparisons with the dot-com bubble of 2000 are tempting, a structural analysis reveals significant differences. Towards the end of the last century, companies with no revenue, financed mainly by retail investors and cheap credit, were achieving astronomical valuations. However, the key lesson from that crisis was overinvestment in telecommunications infrastructure.
At the time, companies built over 80 million miles of fibre-optic cable based on the erroneous assumption that internet traffic would double every 100 days. Consequently, for years after the bubble burst, 85-95 per cent of the fibre-optic cable remained unused, classified as ‘dark fibre’. Manufacturers suffered as a result: Corning’s share price, for example, fell from $100 to just $1.
Writing for Seeking Alpha, analyst Victor Dergunov points out that the current investment frenzy combines the characteristics of the dot-com bubble with systemic risks reminiscent of the 2008 financial crisis. Capital expenditure is currently financed through a complex private debt market, non-bank loans, and closed-loop financing. How exactly does this work?
Artificially inflated valuations
Tech giants are channelling capital to AI start-ups so that they can purchase cloud services from them, which artificially inflates valuations. According to Dergunov, there is enormous political and corporate pressure to increase spending on AI, even if the figures quoted by CEOs are inflated.
Nevertheless, industry leaders are emphasising the civilisational significance of the transformation taking place. In a statement released by the BBC, Sir Demis Hassabis, founder of DeepMind, which was acquired by Google, emphasises the unimaginable scale of this technological breakthrough:
‘Artificial intelligence cannot be compared to standard technological breakthroughs, even ones as significant as the internet or mobile technology. It is far more akin to the discovery of electricity or fire… Essentially, we’ve found a way to make sand think. It’s marvellous.’
However, one must ask whether such statements reflect a genuine belief among decision-makers in the technology’s revolutionary nature, or if they are intended to maintain investor and industry interest in order to sustain profits from the enormous expenditure on artificial intelligence. Furthermore, history teaches us that investors can lose fortunes even when the technology they invest in ultimately changes the world. This is exactly what happened with 19^(th)-century railway companies, for example.
The physical barrier to transformation. Will we have enough electricity to power the development of AI?
While Wall Street analysts debate the ratio of expenditure to profits, the real threat to AI stems from engineering and energy realities. Reports presented at the CERAWeek conference in Houston, Texas, show that global expenditure on data centre construction reached an astronomical $771 billion in 2025. This is almost equal to the annual investment of the entire global oil and gas sector ($835 billion) and the renewable energy sector ($798 billion) combined. However, the bottleneck to development is not a lack of capital or processors, but rather the physical capacity of electricity grids.
According to forecasts by the International Energy Agency (IEA) and Deloitte, global electricity consumption by data centres is expected to increase from approximately 450 TWh in 2025 to up to 1,065 TWh in 2030. If data centres were a country, they would be the world’s fifth-largest energy consumer, ranking between Japan and Russia. Furthermore, the rapid development of autonomous agent-based AI is increasing the demand for computing power by a factor of ten to twenty, completely offsetting the annual gains in energy efficiency of the chips themselves.
AI developers at Anthropic estimate that training a single flagship model in 2027 will require 5 gigawatts of power. They also predict that the entire US AI sector will consume 50 gigawatts of new electrical power by 2028: double New York’s peak demand.
The clash between technological promises and construction realities is causing serious investment bottlenecks. According to the GlobalData database, of the 583 major data centre projects announced in 2025, 90 per cent (526 projects) were still in the pre-construction phase in 2026. OpenAI, SoftBank, Oracle and MG’s flagship project, Stargate, had to be modified, and plans to build facilities in the UK and Norway were abandoned. Meanwhile, in Kenya, Microsoft and G42’s billion-dollar geothermal data centre project has been put on hold due to a lack of agreement on power allocation guarantees.
The war over resources: industry, water and local resistance
The rapidly growing energy demand of the digital sector is directly impacting traditional industries and domestic consumers. In the United States alone, data centres are expected to account for 40-50% of the total increase in electricity demand this decade. The state of Texas forecasts that, by 2028, the load generated by AI server centres will exceed 40 GW, almost half the load of the entire ERCOT grid. These constraints could lead to drastic increases in electricity prices for manufacturing plants, as well as the risk of local blackouts.
In addition to electricity, water used in cooling systems is becoming a vital yet scarce resource. In response to environmental threats, an increasing number of local authorities and governments are introducing legal restrictions. For example, the Dutch government has imposed a nine-month moratorium on issuing permits for hyperscale facilities, forcing investments to relocate to South-East Asia. Public resistance is also growing, with residents protesting against the construction of new high-voltage power lines and data centres.
Another unknown factor is the potential technological revolution coming from China. Reports of breakthroughs in China’s integrated circuit manufacturing process suggest that Beijing is becoming increasingly self-sufficient in chip design. If Chinese engineers develop efficient models that do not require such massive computing infrastructure, Western investments in gigawatt-scale data centres will become difficult to sustain financially.
Is a final crash inevitable?
The tech giants have no intention of slowing down their spending any time soon. According to a report by the World Economic Forum, global spending on AI is expected to exceed $2.5 trillion by 2026, with infrastructure accounting for over half of this figure. The five largest companies (Amazon, Alphabet, Microsoft, Meta, and Oracle) are set to spend over $700 billion collectively on capital expenditure (CapEx).
In a sense, these companies find themselves in a Catch-22 situation. Such enormous expenditure may simply fail to pay off; however, halting it would signal weakness to investors and competitors.
Victor Dergunov’s market forecasts suggest that the current corrective wave does not represent the bubble’s final peak. CapEx infrastructure expansion could drive the S&P 500 index up to 8,200 points by 2027. However, the Shiller P/E ratio is approaching 42 points, close to the dot-com bubble peak of 44 points, whilst the historical average is 17.41.
A return to the mean, coinciding with a slowdown in CapEx spending, could trigger a 50 per cent or more sell-off in the stock market. Dergunov expects a full-blown crash to occur in 2027 or 2028, when it becomes clear that returns on investment do not cover the cost of debt.
Has the AI bubble burst? For now, we are witnessing a significant correction, but the coming months could be pivotal for the sector. Eileen Burbidge, a renowned technology investor, told the BBC that, while the bubble has not yet burst, it is ‘losing air’.
The artificial intelligence sector is facing the reality of infrastructure and financial constraints. However, the server centres, power lines and power stations that have been built will not be rendered obsolete. They will form the digital backbone of the new industrial economy, just as 19(th)-century railways and 20(th)-century fibre-optic cables did. However, before that can happen, investors and companies must pass a market test to reveal which business models actually create value and which were merely a financial illusion.
