Recently, I've noticed the AI market rally. Many people are still blindly chasing the trend, but understanding the industry chain structure is actually the key to investing in AI stocks. AI is not a single industry, but the entire supply chain; the profits for upstream, midstream, and downstream are completely different, and the stock price driving logic also varies greatly.

First, let's talk about upstream hardware. NVIDIA, as the absolute leader in AI chips, holds about 80-90% market share, and just from data center GPUs alone, it can generate over $100 billion in annual revenue. But the real moat isn't in the hardware itself; it's the software ecosystem built over more than a decade—developers are accustomed to programming on NVIDIA platforms, making switching costs extremely high. TSMC is also crucial here; almost all chips from NVIDIA, Apple, and AMD are produced by TSMC. By 2026, TSMC's HPC business will account for as much as 58%, with a 48% annual growth rate, making it the strongest growth engine. Moreover, TSMC has already initiated four consecutive years of price increases; customers know this will continue for four years, and they are not holding back, which shows how strong the demand certainty is.

The midstream is the cloud platform players. Microsoft, Amazon, Google, and Meta are expected to have capital expenditures approaching $600-700 billion, directly determining the demand for upstream chips. Microsoft, due to its exclusive partnership with OpenAI, has deeply integrated Copilot into Windows, Office, and other products used by over a billion users, continuously unlocking monetization potential. Amazon's AWS has tied up with Anthropic, forming a complete closed loop—serving as a cloud partner and supplying self-developed AI chips, with cash flow returning as infrastructure costs. This model is very clever. Meta directly monetizes through AI-optimized advertising and Llama open-source models, greatly improving the accuracy of Facebook and Instagram, which is directly reflected in revenue.

The downstream application layer includes software companies like Salesforce, ServiceNow, and Adobe. They embed AI into their products, focusing on enterprise adoption speed and IT budget growth. However, downstream usually lags upstream by 1-2 quarters because after chip shipments, it takes time to build infrastructure.

If I were to recommend AI stocks, I would categorize them by risk level. For more stable options, Microsoft, Amazon, and TSMC are solid; AI is just one of their growth drivers, and even if the hype cools down, their core businesses can still support them. For capturing mainstream capital, NVIDIA and Meta are highly tied to AI, with rapid growth but also higher volatility. For those willing to accept high risk for explosive opportunities, second-tier AI chip companies or innovative startups in applications could be considered.

In Taiwan stocks, TSMC is the foundation. The 2nm process and CoWoS advanced packaging are irreplaceable standards, with stable long-term pricing power. Hon Hai (Foxconn) focuses on system integration, but recently its stock price has weakened, and the gross margin improvement has not met expectations. MediaTek has laid out in edge AI and mobile AI; the Dimensity series now includes built-in AI computing units, and they are collaborating with NVIDIA on automotive and edge solutions. For cooling solutions, Chicony and Shuanghong are seeing structural growth due to demand for liquid cooling systems, which also presents opportunities.

But honestly, recommending AI stocks isn't without risks. Valuations have already significantly increased by 2026, and many company stock prices already reflect years of growth expectations. Competition risks are also present, with AMD and Google TPU competitors catching up. Capital rotation risk is even greater—once market sentiment shifts to other themes, capital outflows can happen quickly. There’s also geopolitical risk; export controls can directly impact the supply chain.

Looking back at the Internet era, Cisco’s stock price surged to $82 during the 2000 dot-com bubble, but after the bubble burst, it fell below $8, and it hasn't returned to that high in over twenty years. This reminds us that infrastructure companies, even with solid fundamentals, may be better suited for phased positioning rather than stubbornly holding on.

Therefore, my advice is to adopt a phased investment approach. Gradually build positions, wait for pullbacks, control individual stock allocations, and continuously monitor the speed of AI technology development, application monetization, and company profit growth. As long as these conditions remain, the investment value of AI stocks can continue to be supported. Short-term volatility is inevitable, but the long-term trend still leans toward growth.