After a 20% consistent pullback, the leading U.S. stocks seem to have bottomed out. Trump's tone has also become more flexible. Today after the Hong Kong market closed, $XIAOMI-W(01810.HK), as the stock king, announced a high-level share placement which sent chills down our spines. Maybe this time is different from before?

Anyway. At the opening, I sold 8% of my speculative position in $Alibaba(BABA.US) and switched back to $Taiwan Semiconductor(TSM.US) and $NVIDIA(NVDA.US).

I won't elaborate on the logic behind TSMC. As for NVIDIA, I had no position before but bought back the stock this time due to some new short-term insights. I'll make a separate post about the outlook another day. In short: NVIDIA still has cost-performance advantages in inference scaling. Demand in the training sector hasn't decreased with the emergence of deepseek; instead, it has increased because in the era of agents, the explosion in application-side demand currently appears to be happening in training rather than inference. The training sector will continue to be competitive.