Investment Review: Maintaining High Exposure and Increasing Investment in Edge AI
In July, we maintained a high exposure, with a focus on holding positions in internet, cloud computing, semiconductors, consumer, and fintech sectors. The proportion of AI investments slightly increased to 40-45%, primarily targeting B2B applications, leading chip foundries, and edge AI. In July, we further increased our investment in edge AI. The proportion of overseas assets remained around 80%.
Market Review
In early July, the U.S. stock market experienced a significant pullback, with panic starting to spread at the beginning of August. There are many explanations for this market behavior, including the unwinding of crowded trades (such as buying U.S. stocks of M7, yen carry trades, etc.), and concerns about a recession. Generally, the market pullback was influenced by many non-fundamental reasons.
From a fundamental perspective, the leverage levels across various entities in the U.S.—corporates, households, and local governments—are relatively low and healthy. Even if economic growth slows, it is a matter of growth rate rather than systemic risk. As for the federal government, although its leverage is indeed high, the U.S. retains the strongest national credit globally and has the authority to issue currency, which mitigates the risk of a "run" on the government. However, future spending capabilities will be constrained.
From a non-fundamental angle, the unwinding of crowded trades can trigger fire sales. Similar situations occurred during the 2020 pandemic and the 2022 Russia-Ukraine conflict, with the market facing much more severe risks (indices dropped over 30%). Currently, the fundamentals are much stronger, with indices having already adjusted by around 15%. While further adjustments cannot be ruled out, excessive pessimism is unwarranted unless there is significant fundamental pressure.
Given the low risk of a blowout, the economy may still face the risk of a rapid slowdown. However, the Federal Reserve has the opportunity to counteract this by cutting interest rates. Additionally, we can mitigate risks by stock selection, diversifying our investment logic under similar alpha conditions, and implementing some short hedges. Our goal is to neutralize the impact of a 1x standard deviation risk through stock selection, position management, and diversification, and to minimize losses in the face of a 2x standard deviation risk.
Revisit June Report: Key Takeaways on AI Investment Opportunities
In our June report, we focused on evaluating AI investment opportunities, reaching several core conclusions:
NVIDIA's High Market Expectations:
Currently, the market has set very high expectations for NVIDIA. Despite industry investments in AI hardware reaching hundreds of billions of dollars, NVIDIA's revenue is only in the tens of billions range. However, NVIDIA's valuation reflects expectations that the AI market will grow to several trillion dollars or even higher. While it is possible for these expectations to be realized in the future, NVIDIA would need a substantial upward adjustment in AI market forecasts to justify further significant price increases at its current valuation. This scenario seems overly optimistic. Additionally, if the AI applications market is expected to reach trillions of dollars, opportunities should naturally spread to application areas, indicating that current applications are undervalued.
Promising Future for AI Applications:
The outlook for AI applications is very positive, as the value of human labor continues to rise. AI enhances human efficiency, substitutes for some human tasks, and even accomplishes things that humans cannot. Therefore, the potential value that AI can achieve is immense, given the large scale of the human population. High-quality AI application companies are currently not fully priced for this promising future.
Edge AI:
Edge AI is likely to become a personal AI super assistant. However, the requirements to implement a super assistant at the edge are very high, and globally, very few companies possess the comprehensive capabilities needed to achieve this.
Unchanged Views with Increased Validation from Research and Financial Disclosures
Our previous views remain unchanged, and as we delve deeper into our research and analyze company financial reports, we find that our judgments are gradually being validated. Here are some key points worth sharing:
Exponential Increase in Computing Power Demand for Model Training:
With each new generation of large model training, the demand for computing power increases exponentially. Without commercialization and sufficient cash flow, it becomes increasingly difficult to sustain model upgrades and continuous investment. For instance, based on model parameters and required computing power, training GPT-4.0 would require an investment of several hundred million dollars in GPU cards; training GPT-5.0 might cost billions; and for GPT-6.0, the investment could reach tens of billions. Continuing to GPT-7.0 could require an investment of hundreds of billions. Faced with such massive investments, any MEGA company would find it a tough decision without adequate revenue and cash flow.
Investments by MEGA Companies:
Annual investments by major companies appear to be in the range of $50-100 billion (with Meta investing slightly less at around $37-40 billion). However, the actual amount spent on GPU cards is much lower. For instance, Microsoft's total investment, including leasing, might reach $70-80 billion in 2024, but only about half of this is allocated to non-GPU expenses like building/leasing data centers, leaving less than $30 billion for GPU purchases. Similarly, Amazon's capex is substantial, around $60-70 billion, but a significant portion is allocated to logistics, leaving only a few billion for GPU purchases. These tech giants are enticed by the promising prospects of AGI and the potential for the first movers to win it all, leading them to join the "arms race" despite the immense pressure to scale their investments strategically.
Scenario Analysis for Model Iteration:
Conducting a scenario analysis for models iterating to GPT-6.0 (requiring hundreds of billions for GPU investment) and GPT-7.0 (requiring trillions), the number of companies capable of continuing such investments will be crucial for NVIDIA's performance in the coming years. If application commercialization remains insufficiently scaled by the time GPT-7.0 is reached, with only two companies able to keep up with the investments, NVIDIA's revenue growth will significantly slow; if only one company can continue, there will be a noticeable decline; if three companies can sustain the investments, growth can continue at a high rate.
Commercial Viability of Smaller Models:
Smaller models and those with fewer parameters are closer to commercialization, whereas training ultra-large models moves further away from it. This is because smaller models can provide faster feedback, can be fine-tuned for niche areas, exhibit fewer hallucinations, and are more reliable with much lower costs, making them more commercially viable. Moreover, the advancement of smaller models follows Moore's Law, with 13 billion parameter models (edge models) expected to reach GPT-3.5 levels this year, and potentially GPT-4.0 levels by 2026-2027. Open-source models with 400 billion parameters might already be at GPT-4.0 levels. Increasingly, companies are shifting towards embracing smaller models rather than investing heavily in training large models. Lastly, because large models are extremely costly, extending training times to reduce costs may lead to a situation where smaller models advance more rapidly than large models.
Few Companies Successfully Unifying Model Upgrades and Commercialization:
Currently, very few companies excel at both upgrading models and commercializing them effectively. In the B2B enterprise services sector, Microsoft stands out; in the internet sector, Meta is notable; and in edge AI, Apple is a key player.
As we continue our research and observe financial disclosures, these initial conclusions are being reinforced, increasing our confidence in our investment strategy focusing on AI and its applications.
Market Views & Outlook
Our views on investment opportunities in the U.S. and Chinese markets remain largely unchanged, continuing to focus on structural opportunities. Investment opportunities in China are relatively more challenging, while the U.S. offers more prospects. Additionally, markets such as Latin America and Europe also present opportunities, and we aim to identify more sources of alpha returns.
Where Are the Opportunities?
AI as a Long-Term Alpha Source:
AI is a significant long-term source of alpha. As the value of human efficiency increases and AI enhances this efficiency or even replaces human roles, its value continues to rise. Currently, we are only at the beginning of this trend, with immense potential for growth.
Changing Lifestyles and Health Consciousness:
The shift towards healthier lifestyles and greater self-awareness will continually create consumption opportunities. This trend reflects a growing market for products and services that cater to health and personal well-being.
Increased Purchasing Power:
As people's purchasing power increases, there is potential for higher pricing in certain products and services, enhancing the competitiveness of high-value offerings. This trend supports the growth of premium, high-quality products and services.
Interest Rate Cycle and Investment Opportunities:
If we enter a period of interest rate cuts, sectors previously burdened by rising interest rates, which made it difficult to assess associated risks, might present opportunities. Companies with strong competitive positions and long-term potential could benefit significantly once short-term interest rate risks are mitigated.
Fintech and Emerging Investment Opportunities:
The fintech sector and other emerging technologies continue to offer promising investment opportunities. These sectors are evolving rapidly, presenting new areas for potential growth and returns.
Opportunities in Other Markets:
Markets in regions like Latin America and Japan also present opportunities. By exploring these markets, we aim to diversify our sources of alpha and capitalize on global investment prospects.
In conclusion, while the core of our strategy remains consistent, focusing on structural opportunities and emerging trends, we continue to monitor and adapt to market conditions to identify and seize new investment prospects across various regions and sectors.
精彩评论