Summary of Key Points
In June, global stocks related to AI computing power reached new historical highs, but in July they all experienced a sharp decline: overseas markets (US semiconductor stocks, South Korean companies Samsung/SK Hynix) and A-share sectors in storage and optical communications saw significant corrections, with A-share storage leaders even experiencing a 50% drop. The trigger was Meta's sale of idle computing power, which raised concerns about the sustainability of AI capital spending (the money companies use to purchase equipment for data centers). This, combined with crowded trading conditions and speculative buying by leveraged funds in South Korea, led to global volatility. However, the domestic computing power chain (Huawei Ascend, ChangXin's capacity expansion, and domestic large models) was relatively resilient due to its focus on self-reliance and control. The market is divided on whether this represents a turning point for the AI industry or just a temporary setback following excessive trading. The future will depend on the financial reports of overseas giants (to verify demand) and the progress of domestic substitution.
Detailed Analysis
1. Why Did Meta's Sale of Idle Computing Power Cause a Global Decline in AI Stocks?
As one of the largest buyers of computing power, Meta has been hoarding GPUs (the core chips for AI calculations) over the past two years. With Llama4 completed and Llama5 still in development, 35% of its computing power was idle, leading it to decide to sell this excess capacity. This news had a devastating impact:
- Market Concerns: If even Meta has idle capacity, does that mean AI demand is not as strong? Will other cloud providers reduce their server purchases? Will orders from upstream chip, storage, and optical module companies shrink?
- Crowded Trading: In June, AI stocks rose excessively, and many investors wanted to cash in. Meta's announcement provided an excuse for them to sell, and the leverage-driven buying in South Korea exacerbated the panic, which spread globally, affecting A-share markets as well.
In short, everyone assumed AI demand was unlimited, only to realize that major buyers were selling their capacity, leading to a mass sell-off due to the previous surge in prices.
2. The Difference Between Overseas and Domestic Computing Power Chains
Although both chains are related to AI, their underlying dynamics are different:
- Overseas Computing Power Chain: It relies on the capital spending of giants like Nvidia, Meta, and Google. Companies in A-share markets that supply optical modules and storage mainly receive orders from these overseas firms, so their performance is heavily influenced by Meta's decisions.
- Domestic Computing Power Chain: The focus is on self-reliance and control. For example:
- Huawei Ascend 950 Super Node: This product uses system architecture advantages to compensate for chip performance limitations (due to advanced manufacturing constraints), allowing a single cabinet to integrate 64 domestic computing cards, expanding the scale of related industries.
- ChangXin Technology's Capacity Expansion: As a leading domestic DRAM manufacturer, its expansion will boost orders for domestic equipment (etchers, cleaners) and materials (photolithography chemicals, specialty gases), reducing dependence on overseas cycles.
- Domestic Large Models: The release of the Kimi K3 model (with 2.8 trillion parameters) created a shortage of computing power, leading to increased demand for domestic AI chips.
Therefore, during July's correction, domestic semiconductor equipment companies (such as NorthStar and SMIC) were more resilient than those in storage and optical modules because their business models are less affected by overseas trends.
3. Has the AI Market Ended?
Most institutions believe this is just a temporary setback due to crowded trading, not a turning point for the industry:
- Private Equity View: June's rally was too extreme, with thousands of stocks falling daily and only AI stocks rising. The surge in financing (borrowing money for investing) made the market structure fragile, leading to widespread selling at the first sign of trouble.
- Brokerage Perspective: The trend in the AI industry remains strong—Google has increased its 2026 capital spending to $195-205 billion, and OpenAI has raised its 2030 budget to $750 billion. Historically, bull markets in internet, smartphones, and renewable energy sectors have seen 30%-50% corrections, but leaders always reached new highs eventually.
The key is whether AI demand remains strong (e.g., with ongoing development of large models and growing inference demands).
4. Opportunities for the Domestic Computing Power Chain
The domestic chain's advantage lies in its independent logic and policy support. Future developments to watch include:
- Momentum-Creating Events: The commercialization of Huawei Ascend 950, updates to domestic large models (like Kimi and WenXinYiYan), and progress with ChangXin's capacity expansion will benefit related companies.
- National Strategy: Self-reliance in computing power is essential, and there is significant room for domestic substitution across all aspects of the industry.
- Valuation Opportunities: After July's correction, many domestic computing power companies' valuations have returned to more reasonable levels, making them more attractive.
For example, once Huawei Ascend 950 begins mass production, related companies will see improved performance. ChangXin's expansion will boost orders for domestic equipment manufacturers.
5. What Comes Next?
The future of the AI market depends on several factors:
- Financial Reports from Overseas Giants: The second-quarter reports and capital spending plans of Google, Meta, and Nvidia will indicate whether demand is still strong.
- Progress of Domestic Initiatives: How well does Huawei Ascend 950 sell? Will ChangXin's expansion meet expectations? Are there new breakthroughs with domestic large models?
- A-share Mid-Year Reports: Cash flow, order volumes, and revenue growth from technology companies will provide insights into the health of the domestic computing power chain.
If these indicators are positive, the AI market will continue to grow, with opportunities focusing on the domestic chain and related sectors driven by inference demands.
In Conclusion
The short-term correction is a test of the market's resilience. The overall trend in the AI industry remains strong, and the domestic computing power chain, with its focus on self-reliance, is more resilient. The key is to monitor the financial reports of overseas giants and the progress of domestic substitution.