I've been pondering a question recently: Are people still blindly chasing AI themes?

Honestly, last year when AI stocks soared to the sky, many people couldn't tell what they were actually buying. As a result, after a correction, a lot got trapped. But the true AI concept stocks aren't that complicated; the key is to understand the logic of the industry chain.

AI is not an industry but an entire supply chain. Upstream is computing hardware, midstream is cloud platforms, and downstream is application software. Different segments make money differently, and the driving factors for stock prices are also completely different. That's why when NVIDIA rises, Microsoft doesn't necessarily follow; conversely, when cloud providers' costs increase, midstream stocks might be temporarily suppressed.

The most direct logic for upstream is the supply and demand of GPUs and AI chips. When the four major cloud giants (Microsoft, Google, Amazon, Meta) plan to invest a total of $600-700 billion in capital expenditure by 2026, this is the most direct order signal for NVIDIA and TSMC. TSMC is now raising prices for all processes below 5nm for four consecutive years, with AI chip prices rising by 10%. Customers know prices will increase for four years but still rush to buy—what does this show? It indicates how strong the pricing power of upstream suppliers is.

The midstream story is a bit more complex. Microsoft, Amazon, and Google sell computing power services and model APIs. Their profits depend on cloud revenue growth and the return cycle of capital expenditure. The market is increasingly questioning, "Invest so much, when will we see a return?" Interestingly, when NVIDIA's gross margin reaches 75%, some cloud customers start developing their own chips to cut costs—Google’s TPU and Amazon’s Trainium are examples. This will have a long-term impact on the profit structure of midstream stocks.

The downstream application layer usually lags 1 to 2 quarters behind upstream. After AI chips are shipped, infrastructure must be built first, and only then will revenue from applications reflect the benefits. The penetration rate of AI features in software companies like Salesforce, ServiceNow, and Adobe directly depends on enterprise IT budgets and how much extra clients are willing to pay for AI functionalities.

If I had to pick the true leading AI concept stocks, I would see it like this:

NVIDIA is currently the absolute leader, capturing 80-90% of the AI accelerator market, earning over $100 billion annually just from data center GPUs. Its moat isn’t just in hardware; a software ecosystem built over more than a decade makes developers accustomed to programming on NVIDIA platforms, creating high switching costs.

TSMC is the foundational infrastructure of the entire AI ecosystem. Nearly all high-performance AI chips—NVIDIA’s chips, Apple’s processors, AMD’s server chips—are produced by TSMC. In Q1 2026, HPC (High-Performance Computing) business will account for 58% of TSMC’s revenue, with a 48% annual growth rate, making it the strongest growth engine. JPMorgan projects TSMC’s revenue growth to reach 35% in 2026.

Microsoft is the leading platform for enterprise AI transformation. Through Azure AI, Copilot, and exclusive collaborations with OpenAI, it seamlessly integrates AI into the workflows of global enterprises. As Copilot is integrated into Windows, Office, and Teams—products with over a billion users—the monetization potential continues to unfold.

Amazon’s advantage is often underestimated. AWS is a major cloud partner for Anthropic and also supplies its own AI chip, Trainium, forming a complete closed loop. When the market re-focuses on the monetization of AI infrastructure, Amazon’s advantages will become more apparent.

Meta is a representative of the AI application layer, directly monetizing through ad AI optimization and Llama open-source models. The accuracy of ad targeting on Facebook and Instagram has greatly improved thanks to AI, directly reflecting in revenue. This is a successful case of “AI direct monetization.”

Here in Taiwan, TSMC is the absolute core. Regardless of which model wins the AI race, all high-performance AI chips must be built on the most advanced processes and CoWoS packaging. TSMC not only holds a long-term technological lead but also has stable pricing power, playing a role closer to the foundational infrastructure of the entire AI ecosystem.

Hon Hai (Foxconn), as NVIDIA’s main server manufacturer, saw its stock price weaken in early 2026. Market patience with Hon Hai is wearing thin; the key issue is that its gross margin improvement is far below expectations. Currently, Hon Hai’s cooperation with NVIDIA remains mainly at the foundry assembly level, with no significant technological leap.

MediaTek has laid out in edge computing and AI chips. Its Dimensity series mobile platforms now include enhanced AI computing units, and it is also collaborating with NVIDIA to develop automotive AI solutions.

Regarding investment strategies, if you want to avoid too much volatility, companies like Microsoft, Amazon, and TSMC are solid choices—AI is just one of their growth drivers. Even if the AI hype cools, their core businesses can still support their stock prices.

To capture the mainstream capital flowing into AI, NVIDIA and Meta Platforms are highly tied to AI growth, with strong momentum but also higher volatility. Suitable for investors who can tolerate fluctuations and are willing to hold long-term.

A special reminder: AI concept stocks have accumulated huge gains over the past two years, and many stock prices already reflect years of growth expectations. Once growth slows or market sentiment shifts, the correction could be substantial. Overvaluation, capital rotation, geopolitical risks, and intensified competition are all factors to watch.

Long-term, AI’s impact on human life will likely be no less revolutionary than the internet, creating enormous economic value. But history reminds us that infrastructure companies, even with solid fundamentals, might be better suited for phased positioning rather than long-term buy-and-hold without adjustments.

My advice is to adopt a phased investment approach. Continuously monitor whether the pace of AI technological development is slowing, whether related applications are delivering expected monetization, and whether individual companies’ profit growth is decelerating. Only if these conditions remain favorable can the true AI concept stocks sustain market support. Staggered buying, waiting for dips, and controlling individual stock positions—these are pragmatic strategies.