社区

I. Investment Review: Increased Investment in AI Applications and Emerging Market Super AppsIn November, we maintained a high portfolio position and increased investments in AI applications and emerging market Super Apps. From an industry distribution perspective, the focus remains on holding internet, cloud computing, semiconductor, consumer, and fintech sectors. The proportion of AI investments remained above 30%, with a focus on B2B applications, chip design and manufacturing, and robotics. The proportion of overseas asset investments was 70-80%.

II. Recent Insights

1. The AI Industry Value Chain is Being ReshapedPreviously, the value distribution in the AI industry was heavily tilted towards Nvidia. For example, when cloud providers spent $10 on chips, they might only earn less than $1, with $8-9 of the $10 going to Nvidia, and a few dimes to a dollar going to TSMC.However, with the emergence of Scaling Law in post-training and inference, and as post-training and inference become closer to final applications (and hence closer to "money"), the situation is shifting. Conversely, the investment in pre-training for foundational models is starting to face bottlenecks, as each generation of intelligence increases the marginal investment, while the returns are diminishing due to data and other constraints. The time required for improvement also increases. Large companies have started shifting their investments towards post-training and inference, with Agent 2B applications already seeing success and commercialization.As investment shifts towards post-training and inference, the demand for in-house or co-developed chips by large companies will increase. The value distribution for in-house or co-developed chips differs significantly. For example, with a $10 investment in chip production, large companies can hope to generate $10 in revenue, with design companies taking 20-50% of the $10, and the remainder (around $2-4) going to TSMC. In terms of value distribution, the share for large companies and TSMC will significantly rise.Even if the capex growth of large companies slows or stagnates, it will impact Nvidia’s growth but won’t significantly affect the commercialization of large companies or their AI revenue growth, nor will it affect the growth of chip manufacturers like TSMC.

2. Agent Applications at the Tipping PointAgent applications are at a tipping point due to:

• A sharp decrease in inference costs, making economic sense possible.

• A significant improvement in agent application experience, allowing economic effects to be assessed.

1. Human tasks often need to be broken down into several steps to be completed. In the past, AI was not strong enough to complete the entire task or lacked accuracy. Now, tasks can be completed with higher accuracy.

2. The natural language interaction mode lowers the barriers for highly specialized tasks, such as programming, designing workflows, and data analysis. When the barrier is high, only a few people can use the relevant tools, limiting demand. When the barrier lowers, demand is released.

3. Unlike RPA (Robotic Process Automation) applications with fixed rules, agents can adapt to more tasks.Compared to human processing and RPA applications, agents are at a critical point of experience.Because agents can complete entire tasks and their effectiveness can be assessed, customers can calculate the ROI. Agents are often compared to human value, especially in the U.S., where the average GDP per capita is $60,000. A 20% improvement in efficiency could add $12,000, and taking 20% of that results in over $2,000 annually—around $200/month. Many SaaS companies have ARPU levels of $100/month or lower. Even if SaaS demand is entirely replaced by agents, the TAM (Total Addressable Market) significantly increases.

1. The Year of Humanoid Robots is ApproachingNext year, we may see thousands or even tens of thousands of humanoid robots enter production lines and work alongside humans.Currently, limitations in speed, yield, and energy replenishment mean it may take 3-4 robots to match the work of one human worker. For example, on automotive production lines, humanoid robots may only be able to handle 10-20% of the work.However, subsequent iterations may be quick, and in the near future, we could see tens of thousands of humanoid robots entering production lines, even being sold to the market:

• One reason is that visual-based perception, recognition, and execution capabilities, which have already been established for autonomous driving in the automotive industry, can be reused for robots.

• Another reason is that industry leaders are gradually solving data-related shortcomings through optimal application scenarios.If the capability issue is solved, the economic case for humanoid robots in factory settings is easy to make. In the U.S., unionized automotive workers earn $60,000-$70,000 annually (with starting salaries of $40,000-$50,000). The cost of a humanoid robot is around $35,000-$40,000, and it can be used for several years. As mass production and design optimization progress, costs will decrease.

1. Emerging Market Super AppsWe’ve observed that in many emerging markets, where giants in mature markets fail to create Super Apps (combining digital finance, e-commerce, local life, travel, utility payments, etc.), local companies have successfully achieved this. There are several reasons:

1. The population size may not seem large enough to attract giants, but local companies have achieved high user penetration and strong user stickiness, solving scalability and profitability issues through the breadth and depth of their services.

2. The required infrastructure for applications is not third-party driven, but local platforms have pioneered this.

3. Geographical and cultural factors may make it difficult for global companies to replicate their success, but local companies have optimized solutions.Once a local Super App emerges in an emerging market and competition concerns are alleviated, the scalability and profitability of the market tend to be very promising.

III. OutlookThe U.S. election in November has concluded, but its effects may just be beginning. The new president’s policy direction includes both inflationary pressures (such as higher tariffs) and measures to stimulate the economy (e.g., tax cuts, increased government efficiency, deregulation) alongside spending cuts that may reduce economic growth. The combined effect of these factors, as well as the pace and magnitude of their impact, remains difficult to predict, with additional uncertainties about how other countries will respond. Therefore, market volatility in the future may increase.Currently, U.S. risk premiums are at historically low levels (2002-2022), and although there is some downward room for risk-free rates, it’s hard to expect significant downward movement in market discount rates. The risk premium for Chinese assets has also significantly decreased, and before long-term issues are resolved, we do not expect large valuation increases.Looking ahead, we need to increase sources of alpha, and opportunities lie in new sectors, including AI, humanoid robots, new consumption, and emerging markets.

Where are the opportunities?

• AI is a significant source of medium- to long-term alpha. As human value increases, AI, which enhances efficiency or even replaces humans, will continue to rise in value. This is just the beginning, and the ceiling is extremely high.

• Changing lifestyles focused on health and self-awareness will continue to create consumer opportunities.

• Increased purchasing power will enhance the pricing potential of certain products/services, boosting the relative competitiveness of high-value products/services.

• Emerging investment opportunities in fintech, etc.

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

对投资BEDROCK产品感兴趣的同学，欢迎后台留言，或添加微信BedrockPartners，请注明来意，我们会尽快回复 

欢迎关注网站：https://www.bedrock-invest.com/?lang=zh

BEDROCK播客已经开通，视频号、小宇宙、喜马拉雅、Youtube已上线，不定期分享对投资、产业、公司的理解和感悟，欢迎大家关注。

BEDROCK播客 E12-聊聊自己：上半年投资回顾&展望

BEDROCK播客 E11 - 聊聊日本

BEDROCK播客 E10 -聊聊拉美

BEDROCK播客 E9 -聊聊半导体

BEDROCK播客 E8 -再聊AI

BEDROCK播客E7-聊聊AI 2

BEDROCK播客E6- 聊聊AI 1

BEDROCK播客E5- 聊聊自己，回顾与展望

BEDROCK播客E4- 聊聊天2

BEDROCK播客E3-聊聊天1

BEDROCK播客E2-聊聊出海电商

BEDROCK播客E1-聊聊美国通胀

欢迎阅读往期内容：投资思考随笔

欢迎直接订阅该话题！

往期文章：

美国餐饮为何牛股辈出？

大模型训练烧钱还能烧多久

A short story of BEDROCK

Edge AI 初步探讨

拉美投资（3）- 电商巨头Mercado Libre

拉美投资（2）- 最大的数字银行Nubank

拉美投资（1）-宏观初探

BTC未来可能值多少钱？

BEDROCK 2023 Review and 2024 Outlook

人口与投资

Once Macro Turns- 23年回顾及24年展望（1）

通胀的“后遗症” -美国调研总结（1）

Some Thoughts on TSM

Some thoughts about AI

对AI LLM的一些猜想

ChatGPT研究及投资猜想