The review of March market predictions happened to be a prelude to the global tech stock frenzy in April. The script of the optical communication sector shifting from NVIDIA to LITE during the GTC conference, placed against the backdrop of Zhongji Xunchuang's 262% profit surge and Yuanjie Technology becoming the "stock king" of A-shares today, is even more worth revisiting in detail.

Core view: The prediction results reveal a recurring phenomenon—the tension between the "certainty trap" and the "minority logic." When expectations are highly concentrated, excess returns often hide in overlooked corners.

In the March GTC prediction, NVIDIA received the highest support at 33% but fell 4.19% that week; the eventual champion was LITE with only 9% of the votes. This is no coincidence—when consensus is too concentrated, most bullish funds have already entered the market, leading to insufficient follow-up buying pressure, making it vulnerable to short-term pressure. Additionally, conference positives are often priced in early, and counterforces appear after the event materializes. The overall surge in the optical module sector now shows that the underdog's comeback back then was the market "correcting" consensus in the right direction.

The Hong Kong stock earnings week was similar. Tencent and Alibaba received 42% and 21% support respectively, both declining that week, while the only gainer, Horizon Robotics, rose 1.24%. Large-cap blue-chip earnings were already fully priced in before the reports, making it hard to achieve excess returns even with in-line performance; smaller, more focused stocks are more likely to exceed expectations.

This reveals a framework: consensus ≠ wrong, but excess returns often lie where consensus doesn't reach. With AI computing power demand surging now, the Nasdaq hitting new highs for days, and funds continuing to bet on the AI theme—the sector is indeed good, but the leading stocks at different stages are completely different. The small wins of LITE and Horizon Robotics in March were precisely the prelude to the broader rally in April.

Looking ahead, AI is transitioning from "training-driven" to "inference-driven," with computing power demand continuing to expand. Optical modules and domestic computing chips are far from peaking. However, a clear macro trend doesn't mean any stock will make money at any point—the "anti-consensus" of the March predictions serves as a reminder to still identify stage crowding and pricing gaps even in the right direction.

Predictions use task coins, but real trading relies on fundamentals. If one goes all-in on underdogs just because of prediction results, they fall into another consensus trap. The March predictions provide an observation window, but decisions ultimately require one's own framework.

Finally, a small suggestion: if future events include consumption recovery, new energy, pharmaceuticals, and other sectors, the battlefield for predictors will be richer.

If selected as a high-quality comment, 1000 task coins will be reserved for the next round of "anti-consensus."