October 2026 was a bad month for technology shares. The Nasdaq Composite fell 5.3%, its worst month since March 2025, and the Philadelphia Semiconductor Index lost 15.4%. On 29 October Alphabet had its worst trading day on record, falling nearly 13% and losing about $482bn of market value in one session. “The AI bubble is popping” became the headline of the month.
But share prices and technology are two different things. The same weeks that hammered the stocks also brought some of the strongest demand figures the AI industry has reported, and a steady stream of new models. This explainer separates what fell from what didn't, and why.
What actually fell
- 14 October: US inflation came in at 3.7% for the year to September, above expectations. The most debt-heavy AI infrastructure firms fell hardest: CoreWeave dropped about 16% and Oracle about 9%. CoreWeave's bonds traded at yields near 15%, levels usually associated with distressed debt.
- 22 October: after Meta's results, the Nasdaq fell 2.6% and chip stocks 6.7%, their worst session since June.
- 29 October: Alphabet fell nearly 13% after a quarter in which its free cash flow turned negative.
- 30 October: chip stocks fell another 7.2%, led by Intel's weak forecast, closing the month down 15.4%.
Not everything fell equally. Microsoft barely moved on its results day, and the broader S&P 500 ended October down only about 1%.
Why: the cost of building, and how it is paid for
The pressure came from the cost side, not the demand side. The biggest technology companies are spending at a pace without precedent:
| Company | Q3 2026 capital spending | What else stood out |
|---|---|---|
| Alphabet | $51.4bn (up 114% on a year earlier) | Free cash flow negative, at about −$9.1bn |
| Meta | $34.9bn | Free cash flow about −$2.1bn; 2026 capex guided at $135–145bn |
| Amazon | $75.9bn in cash capex | Trailing 12-month free cash flow −$42.4bn; plans to spend “meaningfully more” in 2027 |
| Microsoft | $39.4bn in additions to property and equipment | First major bond sale in more than nine years, about $25bn |
That money increasingly comes from borrowing and from selling shares. Alphabet, Amazon, Meta and Oracle had issued more than $228bn of bonds in 2026 by mid-October, more than double their total for 2025, and Meta announced plans to raise up to $97.5bn by selling new shares. Higher inflation and the prospect of higher interest rates make that borrowing more expensive. On 28 October the US Federal Reserve held rates at 3.75–4.00%, with two members voting for a rise.
Some of the fall had nothing to do with AI. Oil above $110 a barrel, a 10-year US bond yield around 5% and Intel's PC-related guidance all weighed on the market.
What didn't fall: demand
On the revenue side, the AI supply chain had one of its strongest quarters:
- TSMC, which makes most advanced AI chips, reported third-quarter revenue of US$46.7bn, up 41% in dollar terms, and guided higher for the fourth quarter.
- Nvidia's most recent quarter (to July) brought $96.2bn of revenue, up 106%, with $108bn guided for the next. It reports again on 17 November.
- Cloud growth accelerated: Amazon Web Services grew 39%, its fastest in nearly five years; Microsoft's Azure 48%; Google Cloud 103%.
The models kept coming too. Between September and October, OpenAI shipped GPT-6 and GPT-6.1 tiers, Anthropic released Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5, Google launched Gemini 4 Argon and Carbon, and Moonshot, Mistral and DeepSeek all released new large models. Our AI Breakthroughs timeline lists them.
Is this the dot-com bubble again?
There are real similarities: a huge infrastructure build-out, money racing ahead of proven profits, and companies financing each other's growth. In 2000 the market rightly questioned whether all that spending would pay off, and many companies did not survive.
There are important differences too. Many dot-com companies had little or no revenue. Today the companies selling the picks and shovels (chips, data centres, cloud capacity) are already earning very large sums, as the figures above show. The open question is one layer up: whether the businesses and people using AI will earn enough from it to justify the spending. That return is real in places, such as coding, customer service and research, but it is still early and uneven.
So two things can be true at once. Investors can be right that some companies are overspending or borrowing too much, and that some share prices ran too far, while the underlying technology keeps improving and spreading.
What to watch next
- Nvidia's results on 17 November: the clearest signal of whether demand for AI chips is still rising.
- Free cash flow at the biggest spenders: whether it stays negative, and for how long.
- Borrowing costs: bond yields for data-centre builders such as CoreWeave and Oracle.
- Inflation and interest rates, which set the price of all that debt.
- Evidence of returns for businesses that use AI, not just for those selling it.
For how the money flows behind AI fit together, from circular financing to data-centre loans, see our AI Money timeline.
Sources
- TheStreet Pro — chip rout sends Nasdaq to worst month since March 2025
- investingLive — CoreWeave bonds and Big Tech bond issuance
- Meta Platforms — Q3 2026 results (SEC filing)
- Amazon — Q3 2026 results
- XTB — Alphabet Q3 2026 results
- CoinCentral — Alphabet's record one-day fall
- Finimize — Microsoft's first bond sale in nine years
- Tech Insider — Meta's $97.5bn share-sale plan
- Let's Data Science — Amazon on 2027 spending
- Vested Finance — Microsoft Q1 FY27 results
- TSMC — Q3 2026 results
- Nvidia — Q2 FY2027 results
- US Bureau of Labor Statistics — September 2026 CPI
- Federal Reserve — 28 October 2026 statement
Frequently asked questions
Is the AI bubble popping?
Share prices of many AI-linked companies fell sharply in October 2026, mainly over spending, borrowing and interest-rate worries. Demand figures from chipmakers and cloud providers stayed very strong, so in our reading the evidence points to a re-pricing of risk rather than a collapse in AI use.
Why are AI stocks falling if demand is strong?
Because building data centres costs hundreds of billions of dollars, much of it now borrowed, and investors are questioning when that spending will pay back. Higher inflation and interest rates make the debt more expensive.
How is this different from the dot-com crash?
The companies supplying AI infrastructure already earn very large revenues, unlike many dot-com firms. The uncertain part is whether customers using AI will earn enough to justify the spending.
Should I sell my tech shares?
We can't tell you that, and this article is not financial advice. If you are deciding, speak to a qualified financial adviser who knows your situation.