AI Infrastructure Investment Traps and a Bottleneck Evaluation Framework

TL;DR Two traps dominate AI infrastructure investing: the "every AI company wins" fantasy and the "find the next Nvidia" scavenger hunt. Proven bottlenecks (Tier 1) and emerging ones (Tier 3) don't deserve the same confidence. One question — "Which bottleneck does this company control?" — can fundamentally improve your investment decisions.

I've noticed a pattern while analyzing AI-related stocks recently. Even investors who genuinely understand AI infrastructure's importance tend to fall into the same traps when converting that understanding into actual investment decisions.

There are two common traps. And there's a practical framework for avoiding them.

Trap 1: The "Everything AI Infrastructure Wins" Fantasy

As interest in AI infrastructure grows, a predictable phenomenon emerges: every company with any connection to AI infrastructure suddenly gets treated like a winner.

This is a dangerous oversimplification.

Just because a chokepoint genuinely exists doesn't mean every company near that chokepoint captures the same economics. Lumping TSMC with every foundry-adjacent name is wrong. Equating Micron with every memory-related company is equally wrong. Treating Broadcom the same as every networking name is also wrong.

Don't flatten leaders, challengers, and adjacent suppliers into one giant bucket. There's a fundamental difference between a company that "controls" a bottleneck and one that merely "sits near" it.

Trap 2: The "Next Nvidia" Scavenger Hunt

Be wary of this framing. The "let's find the next Nvidia" approach almost always drifts toward vague hype and weakly-evidenced small-cap names.

My approach is different. Start with proven bottlenecks first. Then expand carefully into surrounding names. This keeps analysis grounded in what the AI system actually needs, rather than whatever story is being pushed hardest on social media this week.

Let me be clear about the distinction: "an interesting area worth studying" and "an area with enough proof to deploy capital right now" are two different things. Take optics as an example — it's absolutely worth deeper study. But assigning it the same conviction as foundry or HBM would be premature.

Proven Bottlenecks vs. Emerging Bottlenecks

Dimension	Proven (Tier 1)	Important (Tier 2)	Emerging (Tier 3)
Layer	Foundry · HBM · Power/Cooling	Packaging · Networking	Optics/Photonics
Key Companies	TSMC · Micron · Vertiv	TSMC · Broadcom	Lumentum · Coherent
Physical Constraints	Very clear	Clear	Becoming more apparent
Replaceability	Extremely low	Low	Still fluid
Investment Conviction	High	Moderate	Observation stage
Appropriate Approach	Core position	Selective position	Research / Watchlist

This table matters because it prevents the common mistake of treating all bottlenecks with equal weight.

A Practical Evaluation Framework: Four Key Questions

Stop asking "which AI stock is next." Instead, ask these four questions:

1. Which bottleneck does this company control? Being near a bottleneck and controlling it are different things. Can the AI system function without this company in this layer?

2. How hard is that bottleneck to replace? How many years would it take a competitor to reach the same level? What's the capital requirement? What are the switching costs for existing customers?

3. How essential is it to the full AI buildout? If this layer disappeared, would the entire AI system halt, or would performance merely degrade slightly?

4. Can that control translate into pricing power or staying power? Controlling a bottleneck doesn't automatically guarantee good economics. Can prices be raised? Does the advantage hold through technology transitions?

These four questions alone can filter out a significant amount of poor AI investment thinking.

Bull Case vs. Bear Case

Bull case: As AI data center spending expands, the market may keep rewarding companies that control the parts of the system nobody can skip. When demand ramps, the strongest bottlenecks become real toll roads. That's the second-order opportunity hiding behind the obvious names.

Bear case: Some of this thesis may already be understood by the market. Some names may already be crowded. Some adjacent suppliers may not capture nearly as much economics as investors expect. The theme can be right while individual stock picks still go wrong.

The right takeaway: the AI story is bigger than any single chip company, and many investors are still thinking about it too narrowly. If the market continues broadening from obvious winners into hidden infrastructure layers, this bottleneck framework becomes more valuable, not less.

FAQ

Q: Can this framework be applied to small-cap stocks? A: The framework itself applies, but small caps tend to have weaker evidence for "controlling" a bottleneck. Understand the large-cap anchors first, then approach small caps as research subjects rather than conviction positions.

Q: Does this bottleneck framework apply beyond AI? A: Absolutely. EV battery supply chains, defense industry, energy transition — any complex system buildout involves the same "bottleneck dominance" logic. The key question is always: "What part of this system can't be skipped?"

Q: What if the bottleneck is certain but the stock is already expensive? A: Bottleneck certainty and current valuation are separate issues. A certain bottleneck means high business quality — it doesn't guarantee a fair price. Valuation assessment must be done independently.