Are AI Stocks Really a Bottleneck Trade?

Using recent earnings releases from NVIDIA, Microsoft, Samsung Electronics, and SK hynix, this post breaks down how I think about AI investing across the US and Korean markets.

Contents

AI-related stocks can get confusing fast.

At first, it looks like all I need to watch is NVIDIA. Then Microsoft and cloud capex enter the picture. After that come HBM, memory, power, data centers, cooling, networking, and equipment suppliers.

The question changes pretty quickly.

"AI is growing. But where does the money actually show up?"

When I look at AI stocks, I try not to start with ticker names. I try to start with the bottleneck. If AI demand keeps growing, not every company benefits equally. Money usually flows first to the part of the system that is most constrained.

This is not a buy or sell list. It is closer to the mental map I use when trying to understand the AI theme. If this map were perfect, I would probably be quietly retired already.

Why Bottlenecks Matter

From the outside, AI growth can look like a software story.

But underneath it, the story is very physical.

• More GPUs are needed.
• Those GPUs need networking.
• Memory bandwidth has to keep up.
• Data centers need to be built.
• Power and cooling become harder constraints.
• Cloud and software companies have to turn all of that cost into revenue.

So AI investing is not just about finding "AI companies." It is about asking which resource becomes scarce as AI demand grows.

NVIDIA's fiscal 2027 first-quarter results make that fairly clear. Revenue was $81.6 billion, and Data Center revenue was $75.2 billion, up 92% from a year earlier. NVIDIA now looks much more like an AI infrastructure company than a gaming company.

The first thing I take from that is not "buy NVIDIA." It is:

"The market is paying heavily for the compute bottleneck."

The US Market Is Closer to Demand and Platforms

In the US market, AI exposure often starts with demand and platforms.

NVIDIA sits at the center of GPUs and networking. Microsoft absorbs AI demand through Azure and AI services. In Microsoft's fiscal 2026 third-quarter release, Microsoft Cloud revenue was $54.5 billion, and Azure and other cloud services revenue grew 40% year over year.

That makes it hard to think about AI only as a chip story.

For US stocks, I would separate the questions like this.

The strength of the US market is that many companies sit close to end demand.

The risk is that expectations are also high. A great company can still be a difficult investment if the price already assumes too much.

A good industry and a good entry price are not the same thing. I keep forgetting that too.

The Korean Market Is More Supply-Chain Sensitive

The Korean market has a different shape.

There are fewer direct AI platform plays. Korea's AI exposure is more tied to the components and supply chain behind AI infrastructure, especially HBM and server memory.

Samsung Electronics' first-quarter 2026 results emphasized AI-driven memory demand. Samsung discussed strong demand from AI infrastructure expansion and plans to provide HBM4E samples.

SK hynix also emphasized AI demand in its first-quarter 2026 results. It reported revenue of KRW 52.6 trillion and operating profit of KRW 37.6 trillion, citing expanded sales of high-value products such as HBM, high-capacity server DRAM modules, and eSSD.

So I would not read Korean AI stocks the same way I read US AI stocks.

The US market is more about who captures AI services and platforms. The Korean market is more about who supplies the parts that AI infrastructure needs.

For Korean AI exposure, these questions matter more:

• Is HBM supply still constrained?
• Is revenue too concentrated around a few customers?
• Can the company keep up with the next product generation?
• Are equipment and material companies turning the theme into real orders?
• Where are we in the memory pricing cycle?

Memory is both structural and cyclical.

Even if AI demand stays strong, memory companies still face supply expansion, pricing cycles, and inventory risk. So for Korea, I think the practical question is whether AI demand and the memory cycle are aligned at the same time.

What I Miss If I Only Watch NVIDIA

NVIDIA is obviously important.

But if I only watch NVIDIA, the AI investing map becomes too narrow.

AI infrastructure is not built by one company alone.

The better question is not "which ticker is best?" It is where each bottleneck sits in the cycle.

When GPUs are scarce, accelerators can lead. When GPU supply expands, HBM, networking, power, cooling, and data-center efficiency can matter more. Later, the market may care more about who turns AI features into real software revenue.

That sequence will not always be clean. But it is still better than treating every AI stock as the same trade.

My Checklist

AI investing conversations can quickly turn into "what should I buy?"

I would rather start with five checks.

The last one is the hardest.

AI can be a real long-term theme and still be a bad entry at the wrong price. A strong industry does not automatically solve valuation. On the other hand, if the structure is intact, a pullback can become a chance to look again.

So I try to write the investment case as a sentence:

"This company solves this bottleneck in AI infrastructure, and that bottleneck is still scarce."

If that sentence does not come naturally, I am probably chasing the theme more than understanding the business.

Companies I Would Watch Next

The obvious names are already well known.

In the US, NVIDIA, Microsoft, Amazon, and Google are hard to ignore. In Korea, Samsung Electronics and SK hynix naturally come first. They are core names in the AI cycle, but they are also names the market already watches closely.

So the next layer I would spend more time on is the less obvious bottleneck companies. This is not a buy list. It is a watchlist for checking whether revenue, orders, and margin actually follow the AI story.

The two areas I find most interesting are connectivity and power.

First, AI servers are not only about having more GPUs. GPUs, memory, CPUs, storage, and networks all need to move data quickly. That makes companies like Astera Labs and Credo interesting as a second-layer question after NVIDIA.

Second, AI data centers consume a lot of power. LS ELECTRIC and HD Hyundai Electric are not semiconductor companies, but they sit near the physical infrastructure bottleneck that AI creates.

These names can also be volatile.

If expectations move faster than earnings, the stock can move first and the business has to catch up later. If contracts are delayed, financing gets harder, or customer concentration becomes an issue, the downside can also be sharp. So for these names, I would not stop at "the story looks good." I would keep checking whether orders, revenue, and margins are actually showing up.

My Current Takeaway

AI is likely to remain an important market theme.

But a large theme does not make every related stock attractive. The US market is more directly tied to GPUs, cloud, and software monetization. The Korean market is more tied to HBM, memory, equipment, materials, and power infrastructure.

My takeaway is simple.

AI stocks should not be grouped under one label. I need to ask where the bottleneck is.

Then I need to ask how much of that bottleneck is already priced in.

In the end, the hard part is not finding a good theme. The hard part is holding the right part of a good theme at a price I can live with.

That is why I keep writing this down.

Earnings Releases I Checked

• NVIDIA FY2027 Q1 results
• Microsoft FY2026 Q3 results
• Samsung Electronics Q1 2026 results
• SK hynix Q1 2026 results
• Astera Labs Investor Relations
• Credo Technology Investor Relations
• Nebius AI infrastructure agreement
• LS ELECTRIC AI data center power infrastructure contract
• HD Hyundai Electric official news