Is the "excess computing power" that dragged down the stock market a false proposition? Analysts say it's not an "overall excess of computing power," but rather a structural underutilization.

Glonghui July 2｜Meta dropped a "deepwater bomb" last night, announcing plans to enter the cloud infrastructure services sector and sell surplus AI computing resources. This news triggered a collective plunge in US-listed upstream storage chip and optical communication stocks, and further dragged down the Korean stock market this morning. For a moment, the entire market’s expectations shifted from "bottomless buying frenzy" to "computing power surplus".However, some analysts point out that Meta’s move does not indicate an "overall computing power surplus," but rather structural idleness. Why is it definitely not a comprehensive computing power surplus? Two ironclad proofs:1) Meta is still frantically purchasing computing resources. While preparing to lease out older GPUs, it continues to buy large-scale computing power from third-party cloud providers and build its own new AI data centers. Their full-year 2026 capital expenditure guidance stands at $125–145 billion, with massive new high-end GPUs added for training next-generation, ultra-large-scale models. If there were really an overall computing power surplus, they wouldn’t keep investing hundreds of billions into infrastructure.2) External enterprises are proactively offering premiums to acquire computing resources. At Meta’s shareholder meeting, Mark Zuckerberg clarified: every week, companies approach them willing to pay above Meta’s own costs to rent computing resources. The overall market supply of AI computing power remains tight, with no signs of industry saturation.