The "time lag" in AI infrastructure development is starting to materialize. The concept of a "time lag" was mentioned in July, its core being: large tech companies pay upfront (capital expenditures are spent first), then computing power comes online, orders convert to revenue, and profits and free cash flow are realized last. This "spend money first, then see how efficient and fast you make money" model is the essence of the time lag. Mid-2025 saw a significant acceleration in capital expenditures by large tech companies, meaning that after a year of construction, the funds invested in data centers in the second half of 2025 will be gradually completed and put into operation from the second half of this year to next year. Previous quarterly financial reports consistently stated that demand far exceeded computing power supply, implying that if computing power was sufficient, actual performance would be significantly higher. Even better. In the coming quarters, we should see cloud vendors accelerate their computing power supply (which corresponds to the accelerated capital expenditure a year ago). With demand still growing rapidly, naturally, as much computing power as available can be sold, making accelerated revenue growth for mega-cloud vendors inevitable. This was also discussed here last week: https://(x.com)/qinbafrank/status/2112476585408557223?s=46&t=k6rimWsEbo2D2tXolYcM-A. Large-scale commercialization of AI has only been underway for two quarters and is far from saturation. This post is sponsored by @bitget_zh, "Bitget Buys US Stocks: Instant Entry, Smooth Trading"