Recently, I’ve been analyzing the investment logic of leading AI stocks and found that many people are blindly chasing hot topics without understanding how this industry chain actually makes money. Instead of buying recklessly, it’s better to first understand which segments the AI industry is divided into, as each segment’s stock price drivers are completely different.

AI is not an industry, but an entire supply chain. Upstream is computing hardware (like NVIDIA, TSMC), midstream is cloud platforms (Microsoft, Amazon, Google), and downstream is application software layers. I notice many investors simply don’t grasp the logical differences among these three layers, leading to the wrong direction when selecting stocks.

In the upstream hardware segment, NVIDIA is absolutely the core. Currently, in the AI accelerator market, NVIDIA accounts for 80% to 90% of revenue share, generating over $100 billion annually just from data center GPUs. Its moat isn’t just in the chips themselves, but in the software ecosystem built over more than a decade—millions of developers are accustomed to coding on NVIDIA’s platform, making switching costs extremely high. This is the real hard-to-copy advantage. TSMC is equally critical because almost all AI chips from NVIDIA, Apple, and AMD are manufactured by TSMC. Earlier this year, TSMC announced four consecutive years of price increases for process nodes below 5nm, with AI chips seeing a 10% increase. Customers are willing to pay these increases knowing they will continue for four years. This reflects how strong the pricing power of leading AI stocks upstream really is.

The midstream giants like Microsoft, Amazon, and Google operate under a different logic. They don’t sell chips but provide computing power services and model APIs. Microsoft, through its exclusive partnership with OpenAI, deeply integrates Copilot into Windows, Office, Teams—products with billions of users—continuously unlocking monetization potential. But here’s an interesting reverse effect—when NVIDIA’s gross margin hits 75%, these cloud giants start developing their own chips (Google TPU, Amazon Trainium) to cut costs. If upstream prices rise too sharply, it can directly squeeze midstream profits, which many analysts overlook.

Downstream application layers like Salesforce, ServiceNow, and Adobe depend on enterprise adoption speed. However, this layer usually takes 1 to 2 quarters longer to reflect in stock prices because after AI chips are shipped, it takes time to build infrastructure.

Regarding the selection of AI leading stocks, I believe it depends on your risk tolerance. If you prefer less volatility, companies like Microsoft, Amazon, and TSMC are stable, with AI being just one growth driver. Even if the AI hype cools down, their core businesses can support their stock prices. If you want to capture mainstream capital, companies highly tied to AI like NVIDIA and Meta Platforms have stronger growth momentum but are more volatile. The case of Meta is particularly interesting—it doesn’t make money directly from selling AI products but uses AI to optimize ad targeting, benefiting Facebook and Instagram’s ad revenue. This is a successful example of direct AI monetization.

Taiwan’s leading AI stocks also have investment logic worth noting. TSMC is the absolute core, holding long-term technological leadership and stable pricing power, serving as the infrastructure for the entire AI ecosystem. Foxconn, as the world’s largest contract manufacturer, is a major server manufacturer for NVIDIA, but I noticed that earlier this year, its stock price weakened against the trend because its gross margin growth was much lower than expected. MediaTek is actively deploying in edge AI and automotive AI, collaborating with NVIDIA on related solutions, strengthening its competitiveness in high-end AI chips.

Honestly, buying AI leading stocks now is still feasible, but you must recognize the risks. Valuations are very clear—these stocks already reflect years of growth expectations. If growth slows, the correction could be substantial. Capital rotation is another variable; the market might shift from hardware to software, or from AI to other new themes. Geopolitical factors (export controls), increased competition (threats from AMD and in-house chip development) also cannot be ignored.

In the long run, AI’s impact on human life and productivity will likely be no less than the internet revolution. But this doesn’t mean all AI leading stocks are suitable for long-term buy-and-hold. Looking back at the internet era, Cisco’s stock soared to $82 during the peak of the 2000 dot-com bubble, only to fall more than 90%. Even after more than twenty years of solid operation, Cisco’s stock price has yet to return to that high. This historical lesson reminds us that infrastructure-type companies, even with solid fundamentals, are more suitable for phased positioning rather than long-term hold without adjustment.

My advice is to adopt a phased investment approach. Continuously monitor whether the pace of AI technology development begins to slow, whether application monetization meets expectations, and whether individual companies’ profit growth shows signs of deceleration. As long as these conditions hold, the investment value of AI leading stocks can continue to be supported by the market. Gradually build positions, wait for pullbacks, and keep individual stock allocations within a reasonable proportion—that’s a more pragmatic approach.