Chip stocks fell for eight days straight, the longest losing streak since 2008
For a year the AI trade has been the market. In the eight sessions to 16 October it was the longest losing streak in semiconductors since 2008, and the number people reached for was a trillion dollars.
The Philadelphia Semiconductor Index fell for eight consecutive sessions through 16 October, its worst week since March. The size of the hole depends on what you count: about $1 trillion across twenty large AI and chip companies over the week, or roughly $2 trillion across the index over the full eight sessions. Reporting points at three things: rate worries after a hot inflation reading, growing anxiety about how the AI buildout is being financed, and, according to Jefferies, nerves ahead of Anthropic's updated numbers. On 19 October the index rose 2.4% as Treasury yields eased and the US and China trade standoff looked less stuck.
Why this one is different
Previous AI wobbles were about capability: a model disappointed, or a demo was staged. This one is about the balance sheet. The week's other stories were a $518bn infrastructure commitment that is mostly non-cancelable, a reported shakeout that leaves fewer than ten of hundreds of AI clouds standing, and a ten gigawatt data centre entering a grid study. The market stopped pricing what the models can do and started pricing who pays for the concrete.
The doubt is not about the models. It is about the financing.
How we got here
- 18 Sep 2026Anthropic's run rate passes $100bn, with a listing reported for November.
- 28 Sep 2026Its prospectus shows $518bn of infrastructure, about 80% non-cancelable.
- 29 Sep 2026OpenAI's run rate approaches $70bn, up more than 70% in a quarter.
- 16 Oct 2026Chip stocks close an eight session slide, the longest since 2008.
What it does and does not mean
A week is not a verdict, the index bounced 2.4% three days later, and the trillion dollar figures are arithmetic on a basket somebody chose rather than a measured loss. Rate expectations moved in the same week, which moves every long-duration asset whether or not it has a GPU in it. What is worth keeping is the change of subject. The question being asked of this industry is no longer whether the technology works, but whether the spending can be financed at these rates, and that is a question the labs answer with prospectuses rather than demos.