Key takeaways
The sell-off was concentrated in the layers that build and power AI, not spread evenly across tech.
Spending, power supply and financing costs each pressured a different layer.
Whether AI is a bubble is debated. This article sets out both sides and gives no verdict.
What happened in October 2026
The Nasdaq Composite fell 2.6% on Thursday 22 October, its steepest one-day fall since 5 June, Al Jazeera reported. The index closed about 5.8% below its all time high. Amazon (down 4.6%) and Nvidia (down 4.4%) led the declines among large companies, while Apple rose 1.2%.
Reporters linked the fall to concern about whether the very large capital investments in AI are sustainable. Capital expenditure, often shortened to capex, is money spent on long lived assets such as chips, buildings and servers.
What is the AI stack?
The AI stack is the chain of businesses that makes AI products work. Each layer depends on the one below it. A seven layer breakdown of the AI investment stack goes into more detail and shows where listed exposure exists.
Energy and power: utilities and producers that supply the electricity AI compute needs, such as Constellation Energy, Vistra and GE Vernova.
Compute and semiconductors: designers and makers of chips and memory, such as Nvidia, AMD, Broadcom, TSMC and Micron.
Physical infrastructure: the data centres, cooling and power systems, servers and property owners that house the hardware, such as Equinix, Vertiv and Dell.
Cloud and hyperscalers: the large cloud companies that buy the chips and rent out computing, such as Amazon, Microsoft and Alphabet.
Foundation models and AI labs: the organisations training frontier AI models, such as OpenAI, Anthropic and Google DeepMind. Most are privately held.
Data and tooling: the middleware that connects models to applications, including data platforms and vector databases, such as Databricks and Snowflake.
Applications and software: the products people actually use, built on the layers below, such as ServiceNow, Salesforce and Adobe.
Why each layer moved differently
Spending and returns
Hyperscalers are committing very large sums to AI infrastructure. CNBC reported in February that Alphabet, Microsoft, Meta and Amazon planned to spend nearly $700 billion combined in 2026, with free cash flow under pressure as a result. Competition for AI computing capacity is a main reason big tech is pouring billions into data centres. When investors doubt how fast that spending earns a return, suppliers that depend on those budgets can fall faster than the spenders themselves.
Power supply
A Jefferies analyst told Yahoo Finance that power is the biggest constraint on the AI build out, and that public pushback against new data centres is part of the problem. Projects that wait on grid access or local approval can slip, which affects builders and equipment makers. The International Energy Agency reported that electricity demand from data centres rose 17% in 2025, compared with 3% growth in global electricity demand overall.
Financing costs
Data centres are capital intensive and are often funded with debt. Higher yields raise borrowing costs and reduce the present value of profits expected years from now. The Associated Press reported in June that high yields can pressure investments seen as the most expensive, and its reporting included both stocks and cryptocurrencies among the assets affected.
Is AI a bubble?
Opinions differ, and price moves alone cannot settle the question.
The concern: some critics argue that AI stock prices have risen too high, too fast.
The counterpoint: Nvidia chief executive Jensen Huang has said the technology's benefits will soon ease investor concerns.
A bubble is usually identified in hindsight. Indicators worth watching include revenue growth compared with capex, financing terms, and whether data centre projects reach completion.
What history shows
AI stocks have fallen sharply before. In 2024, Nvidia pulled back on fears that cloud companies were overspending on data centres, then returned to all time highs by October that year. The same analysis, published in March 2025, noted that Nvidia has had drawdowns of 50% or more twice in the last decade. Past performance is not a guide to future results.
How investors approach volatility
Concentration: a small number of AI companies make up a large share of major indices. RBC Wealth Management found that the 10 largest companies made up nearly 41% of the S&P 500's weight at the end of 2025. Single stocks can move far more than a broad fund.
Diversification: spreading exposure across sectors and asset types can reduce the effect of one theme falling.
Time horizon: short term price swings matter less to people investing over years than to those who need the funds soon.
Research: read company filings, check sources and understand what you hold before you invest.
Some investors build a portfolio with tokenised AI stocks, and Luno offers tokenised stocks and ETFs.
Frequently asked questions
1. What is the AI bubble?
The AI bubble is a debate over whether the prices of AI related companies have risen beyond what their future earnings can support. Critics point to heavy spending and high valuations. Supporters point to strong demand for AI chips, cloud computing and services.
2. Is AI a bubble?
Analysts disagree. Some say AI stock prices rose too far, too fast. Others point to demand for chips, cloud computing and AI services. Price moves alone cannot settle the question, so revenue growth, spending levels and financing terms are worth tracking.
3. Is the AI bubble bursting?
The Nasdaq Composite fell 2.6% on 22 October 2026, its steepest drop since 5 June, and data centre equipment, chip and power stocks fell furthest in October. One sell off does not confirm that a bubble has burst. Bubbles are usually identified in hindsight.
4. When will the AI bubble burst?
No one can say, and not everyone agrees a bubble exists. Bubbles are normally identified after the fact, often once prices have already fallen. Company earnings, financing conditions and data centre project timelines give a clearer picture than trying to predict a date.
