Has the AI bubble begun to deflate?
On the rise of the largest technology companies, the repricing of 2026, and what a European business should take from it.
Last reviewed: 4 September 2026
Since 2023, one question has kept returning: will the hundreds of billions being put into AI infrastructure ever be earned back, or is the market building the largest bubble since the turn of the century?
In 2026 a partial answer arrived for the first time. Not in the form of a collapse, but in the form of a repricing. This article sets out what happened, what it does and does not prove, and why it matters to a business deciding today where its AI runs.
How the concentration came about
At the end of the 1990s, the technology landscape was dominated by a different generation of companies. When the dot-com bubble burst, a large part of the value on the Nasdaq disappeared. The companies that survived the storm filled the vacuum their bankrupt competitors left behind.
From around 2013, economists began to name the concentration of market power among a small number of players. The term Big Tech emerged in that period, with a deliberate parallel to earlier power blocs such as Big Oil and Big Tobacco: sectors where the US Congress eventually moved to regulate.
The label Magnificent Seven is more recent. It was coined in 2023 by an analyst at Bank of America and refers to Nvidia, Apple, Microsoft, Alphabet, Amazon, Meta and Tesla.
The scale, measured correctly
On 3 September 2026, the combined market capitalisation of those seven stood at approximately 24.12 trillion dollars. In April 2026, their share of the total value of the S&P 500 was estimated at around 34 per cent.
The second figure says more than the first. A third of the value of the most important US stock index sits with seven companies, which are moreover investing in the same technology.
What you often read in articles of this kind is a comparison with the gross domestic product of the European Union. That comparison does not hold. Market capitalisation is the valuation of all expected future profits; GDP is the output of a single year. The figures are of the same order of magnitude, but they measure different things. We do not use them.
What happened in 2026
Until the end of 2025, high investment was rewarded by investors. In 2026 that reversed.
In April, all seven were down for the year to date, together approximately 2.1 trillion dollars lower than at the turn of the year, and were underperforming the average share in the S&P 500. In June, approximately 2 trillion dollars evaporated in a single month; Microsoft stood at that point more than 20 per cent below its early-June peak.
The sharpest moment came on 23 July. In one day the seven lost approximately 797 billion dollars, the largest single-day fall since April 2025. The trigger was the quarterly results of Alphabet and Tesla. Alphabet raised its investment budget for 2026 to a maximum of 205 billion dollars and saw its free cash flow turn negative in the second quarter for the first time since its flotation. Tesla warned of exceptionally high capital expenditure.
Nvidia had in 2025 become the first company in the world to pass the 5 trillion dollar mark, briefly, and subsequently fell back to below 4.25 trillion.
What this does and does not prove
- What it proves. Investors have stopped reading the scale of investment as evidence of future profit. Where high capital expenditure was previously valued as a lead, it is now set against visible returns. A negative free cash flow at one of the most profitable companies in the world is an event that puts the model under pressure.
- What it does not prove. This is not a collapse. Valuations are higher than a year ago. The large providers finance their roll-out largely from their own cash flows rather than from debt, which makes their resilience materially different from that of the dot-com companies. Anyone arguing that this is a bubble has to explain why parties with this cash surplus would be mistaken.
There is, moreover, an argument the companies themselves make and which deserves to be taken seriously: in infrastructure markets, whoever owns the capacity ultimately sets the rules, and the risk of being late may weigh more heavily than the risk of overspending.
We do not settle that debate here. What is established is that the outcome does not lie in the hands of the European businesses running on that infrastructure.
Why this matters to your organisation
The temptation is to read this as stock market news. For a business deploying AI on its own corporate data it is something else: it is information about the stability of your own cost base.
Anyone procuring AI from a large cloud provider is buying capacity whose price follows from the investment decisions of a small number of parties. As long as investors reward those investments, there is room to offer services below cost in order to win market share. Once investors demand returns, that room disappears. The first visible consequence of pressure on this model is not insolvency, but a revision of rates, terms and product portfolios.
Three questions follow from this for your own environment:
- What happens to your business process if the cost of your AI service changes substantially within eighteen months, or if the terms are amended unilaterally?
- What happens if the model you are building on is phased out at a moment you do not choose?
- And do you know today which of your processes have come to depend on such a service?
Anyone who cannot answer the third question does not have a pricing problem but an inventory problem. That is precisely what an AI Act audit maps as a first step: which AI your organisation actually uses, through which suppliers, and where the data is processed.