Recently, I noticed a pretty interesting phenomenon—whenever discussing investment opportunities in 2026, AI is always an unavoidable topic. However, what truly deserves attention is not blindly chasing the trend, but understanding how this industry chain actually makes money.

Speaking of which, AI is not a single industry at all, but an entire supply chain. Upstream sells computing power, midstream sells services, downstream sells applications, each driven by completely different logic. If you don’t understand which part you’re investing in, it’s easy to get caught in valuation traps.

Let’s start with the upstream—this is the foundation of the entire ecosystem. NVIDIA, with an 80% to 90% market share of AI accelerators, earns over $100 billion annually just from data center GPUs. But even more formidable is its software ecosystem; millions of developers are already accustomed to programming on NVIDIA’s platform, with switching costs being extremely high—this is the real moat. TSMC is also an absolute key player upstream; NVIDIA, Apple, AMD’s AI chips are almost all produced by TSMC. Starting from January 2026, TSMC will implement continuous price increases for all processes below 5 nanometers for four consecutive years, with AI chip prices rising by 10%, and customers still rushing to buy. JPMorgan predicts TSMC’s revenue growth in 2026 will reach 35% in USD terms.

The midstream includes cloud giants like Microsoft, Amazon, and Alphabet. They don’t sell chips but provide computing power services and model APIs. Microsoft, relying on its exclusive relationship with OpenAI, deeply integrates Copilot into Windows, Office, and Teams, covering 1 billion users worldwide, with monetization capabilities continuously releasing. Amazon, on the other hand, has a five-layer layout binding AI startups like Anthropic; AWS is both a cloud partner and supplies self-developed chips, forming a complete closed loop. But there’s a risk here—if upstream prices rise too sharply, midstream costs will increase. Google and Amazon have already started developing their own chips to reduce costs, which could impact long-term profit structures.

The downstream is the application layer, where software companies like Salesforce, ServiceNow, and Adobe embed AI into their products. Meta is somewhat special; it is itself the largest application scenario, with ad AI optimization directly monetizing, and the targeting accuracy of Facebook and Instagram ads has significantly improved thanks to AI.

Regarding specific targets, NVIDIA is definitely unavoidable—the absolute leader’s position is hard to shake in the short term. TSMC is a foundational infrastructure company with strong long-term certainty, suitable for core allocations. Microsoft and Amazon are stable in nature; AI is just one of their growth drivers, so even if the hype cools down, their stock prices can hold. Meta is a successful case of directly monetizing AI.

Here in Taiwan, besides TSMC, Hon Hai (Foxconn) is a major server manufacturer for NVIDIA, but recently its stock price has weakened, and the market’s patience for margin improvements has almost worn out. MediaTek is also actively布局 AI chips; the Dimensity series already has built-in AI computing units.

If you want to participate in AI but prefer less volatility, choose Microsoft, Amazon, or TSMC. To capture mainstream capital, NVIDIA and Meta offer more flexibility but also come with higher volatility. If you can accept high risk and want to seize explosive opportunities, you can look at second-tier chip companies.

Honestly, AI concept stocks in 2026 will trend as “long-term bullish, short-term volatile.” Valuations have already been significantly inflated; infrastructure companies, though fundamentally stable, will also see their stock prices pull back at the peak of a major bull market. Historically, Cisco’s stock reached $82 at the peak of the internet bubble, then after the bubble burst, it fell over 90%. Even with good management later, it has yet to return to that high. This reminds us that even the best infrastructure companies’ stock prices are more suitable for phased布局 rather than long-term HODLing.

Downstream leading companies have more sustainable business models, but even top-tier competitors like Microsoft and Alphabet will see their stock prices decline at the market top, requiring a long time to recover to new highs. For most investors, a more pragmatic approach is to adopt phased investment strategies, continuously monitor AI development speed, application monetization, and individual company profit growth.

Can you still buy AI concept stocks now? Yes, but it’s better to stagger your布局, wait for dips, and control individual stock positions. Whether short-term or long-term investing depends on your style. For swing trading, you might consider flexible participation via CFD platforms—supporting long and short positions, leverage trading, with no need for complex order types.

Finally, a reminder: the main risks of AI-related stocks include four points—valuations already reflect years of growth expectations, capital may rotate to other themes, geopolitical factors could impact supply chains, and increased competition from AMD and self-developed chips. The core logic behind recommending AI concept stocks remains the same—understand the industry chain structure clearly to judge whether the stock price is undervalued or overvalued.