1. Investment Review​

Maintained a high position level in February. Focused on investments in the internet, semiconductors, cloud computing, consumer goods, and fintech. Continued to increase investments in fintech and emerging markets. Due to slower-than-expected progress in AI applications and the impact of tariffs and geopolitics on upstream chip manufacturing, the proportion of AI investments has decreased. The main investment directions remain AI applications, leading chip manufacturers, edge AI, and robotics. The proportion of overseas investments remained around 80%, with increased investments in emerging markets.

​

2. Market Review and Outlook

Last month, we discussed the profound impact of DeepSeek (referred to as DS): 1) Inference demand will surge; 2) Short-term training demand faces disruptions.

​

• Regarding the surge in inference demand: The low barrier to entry has led to a wave of model deployment in China, but the actual application of AI both domestically and internationally still requires time.

The cost of deploying models has significantly decreased, leading to a deployment boom in China. Although deploying a full-scale DS model with 671 billion parameters is still relatively expensive, costing tens of millions of RMB, deploying a medium-sized DS model is much more affordable, requiring only a few hundred thousand RMB. For individual developers, deploying a small-parameter model can cost as little as 10,000 to 20,000 RMB.

Cloud inference costs have also dropped significantly. DS recently disclosed details on inference cost profits. Under current pricing, which is better than competitors, maintaining full capacity operation could theoretically yield an 85% gross margin. Even considering the difficulty of maintaining full capacity in practice, there is still room for further price reductions.

Overall, the lowered barriers will encourage more people to explore AI applications, and the future of AI application implementation is promising. Previously, the U.S. led the way in AI, but China is quickly catching up.

In terms of actual AI application progress, successful areas are still limited, mainly search, advertising, and code writing. The advancement of 2B applications is slow. Although enterprise customers are enthusiastic about promoting AI, integrating AI capabilities with internal data and processes still takes time. According to Q1 outlooks from various Cloud and SaaS companies, enterprise AI applications have not yet seen a surge.

​

• Short-term training demand faces disruptions:

The release of GPT-4.5 further demonstrates that the marginal performance improvement of foundational model pre-training is slowing down. On the other hand, post-training through reinforcement learning has shown significant progress and is closer to final applications. The previous "brute force R&D" model of stacking computing power may shift or divert some resources to new paths. Although capital expenditures of North American tech giants will continue to grow significantly, they have all mentioned increasing the proportion of inference.

Additionally, DS has shown that through technical optimization, the efficiency of unit computing power can be greatly improved. For example, based on recent DS revelations about inference details, it can be deduced that a cut-down version of H800 might achieve 3-4x the performance of H200 FP8, only slightly worse than B200.

Finally, chip embargoes and sanctions may lead to a split between the Chinese and overseas markets, altering the current landscape of advanced semiconductor design and manufacturing.

​

"Tariff War" Escalates Rapidly, Causing Significant Market Disturbances

The U.S. has officially implemented tariffs on Mexico and Canada, and further imposed a 10% tariff on China in early March, accumulating an additional 20% tariff since February.

The final conditions of negotiations between countries, the countermeasures from the other side, and the subsequent pace and magnitude of tariff increases are hard to predict, making it difficult to judge the impact on inflation and economic demand. This is a significant disturbance to the market. Additionally, DOGE's actions to reduce unnecessary government spending and layoffs have just begun. These are factors unfavorable to economic growth.

On the other hand, major companies have announced plans to invest hundreds of billions in the U.S., and the Trump administration's promises of tax cuts and regulatory relaxation are favorable to economic growth, but these effects have not yet started to take hold.

Currently, the U.S. economic fundamentals have both positive and negative factors, but there is no sign of crossing the threshold to trigger systemic risks. With market valuations having fallen, some risks have already been reflected. Therefore, as long as macroeconomic risks remain controllable, they do not hinder our search for structural investment opportunities. However, it is important to note that although market risk premiums have fallen, they are still at historically low levels, so macroeconomic disturbances can easily lead to significant market volatility.

​

Where Are the Opportunities?

This year, we will more actively seek opportunities from different alpha sources, focusing on new areas: AI, robotics, new consumer trends, fintech, and emerging markets:

• AI is a significant source of long-term alpha. As human value increases, AI, by enhancing human efficiency or even replacing humans, will also see its value rise. This is just the beginning, with a very high ceiling.

• The pursuit of health and self-focused lifestyle changes will continuously present consumer opportunities.

• Emerging investment opportunities in fintech.

• Opportunities in emerging markets (Latin America, Asia-Pacific, etc.).