The real logic behind the rumor of Meta renting out computing power may be

that it has completely fallen behind in the large model race
Meta is gradually falling behind Anthropic, OpenAI, and Google in the AI large model competition
Due to its current stock of older-generation GPUs like H100 and H200, which offer limited help for training next-generation models and face depreciation losses, Meta hopes to generate cash flow by renting out this computing power
The company's overall spending has not decreased; the cash flow is mainly used to continue purchasing new-generation high-end chips like GB200 for training
Renting out computing power was actually a proposal by Zuckerberg at an internal meeting in late May. In a moment of crowded trading in the entire semiconductor industry, the media exaggerated it as an overcapacity of computing power, which was then captured by quantitative algorithms and led to a sell-off
In essence, this is a ghost story designed to wash out weak hands
The current Q index has been hijacked by semiconductors
Semiconductors account for an extremely large weight of over 20% in the Nasdaq 100 index
This also means that even if most of the Magnificent Seven are rising, as long as semiconductors do not rise, the Q index will still be dragged down
Other tech sub-sectors are affected: the sharp drop in semiconductors directly dragged down related concept sectors such as optical modules, data centers, nuclear power, and some small-cap stocks
Personally, I feel that the current negative news is more about triggering panic selling among retail investors when positions are crowded. The overall trend in Q3 is still oscillating upward
In terms of operations:
Buy the dip in semiconductor equipment, then wait for the market to return to rationality