There has been an explosion of information this week. Let's start with AI applications.

To be honest, I am very optimistic about Agent AI—back in 2023 when GPT first became popular, I was already thinking about how to truly implement this technology, and I bought quite a few so-called agent concept stocks with real money. Now that it's 2025, I am even more convinced that the coming years will definitely be the year of Agentic development, which is what everyone is looking forward to as the "Year of Application." Look at how a major chip manufacturer is deploying HBM+Scram architecture—it's clearly tailored for multi-step reasoning and edge deployment.

The problem is—most of the hotly promoted companies in A-shares are not really AI applications; instead, they seem more like "AI damage concept stocks." Calling these companies AI applications is as absurd as saying a certain agricultural company is an AI application. Maybe they even have fewer servers than others, after all, they have seriously engaged in internet-based farming.

What’s more painful is whether these AI application players have really thought through their business models. From what I’ve seen of most of the application companies you’re paying attention to, honestly, researching these is even easier than hardware.

Essentially, media computing companies are doing the dirty work of "intellectual labor-intensive" tasks. The biggest AI applications like coding and text-to-image generation are precisely aimed at replacing this sector. Going deeper, it’s about the Scaling Law—directly converting energy into intelligence, saving humans over a decade of wages and education costs.