Last week’s big tech earnings (googl tsla intc) indicated AI capex is accelerating. Why did semiconductor stocks react so negatively?

Google, Tesla, and Intel earnings last Wed/Thurs all indicated AI related capex exceeded expectations in the previous quarter and that future capex will also be higher than expected.

Yet chip-related stocks fell sharply on Friday (SOXX -4.4%, memory -8+%, neoclouds -10+%). It's important to note that before these ERs, many of these stocks already fell 30+% since mid-May.

I'm having a hard time finding reasons why chip stocks continued to drop after the big tech ERs and other positive sector news last week (Samsung/SK Hynix deals with Nvidia/Broadcom/Anthropic). Could something be happening behind the scenes that suggests AI spending is about to drastically slow down?

Please weigh in if you believe you understand last week's continued chip stock sell off after big tech released bullish ERs. Would appreciate any insights.

Preemptively addressing explanations I expect to see:

1. Escalating Iran War and high oil/interest rates:
I think the war is partially responsible for the pressure on chips. However, the extent of the chip sell off is confusing because other stocks that are historically sensitive to war/interest rates didn't drop materially on Friday. Small caps normally fall as interest rates rise, but were barely down on Friday. Gold also generally falls, but actually finished positive. Oil fell 2%.

The immaterial reaction from other interest rate sensitive stocks suggests that the chip stock massacre was more likely due to chip sector specific developments.

2. Bullish ERs being 'sell the news' events:
I understand chip stocks have gone on a historic run and good news can serve as profit taking events. However, chips already took a huge haircut prior to big tech ERs. Many fell 30-50% between mid-May and July 21 (pre-big tech ERs). It's hard to understand why semis would fall another 5-10% on Friday right after the market discovered chip spend will rise even more than expected.

3. Introduction of low cost Chinese AI models (Kimi K3):
To me, this might be the most likely explanation of the reasons I've considered. Compared to US models, Chinese models are able to generate tokens at a fraction of both compute and memory costs. US companies may try to replicate the results, leading to a reduction in future chip spend.

However, I've read cheaper priced Chinese models are met with so much user activity that compute and memory demand ultimately rise despite less being needed per token (Jevons' paradox). Plus, predictions that chip spend will decline are speculation so far, as both Google and Tesla raised capex guidance.