To The Moon

AI’s Next Bottleneck Isn’t GPUs — It’s Power

The AI trade may be entering its next phase: from chips to infrastructure.

Everyone knows the AI story:

More models → more GPUs → more data centers.

But the bottleneck is changing.

As AI clusters become larger and power-hungry, the question increasingly becomes:

Can the physical infrastructure deliver enough electricity, cooling and networking to run them economically?

That’s where I think investors should start looking beyond the obvious AI names.

⸻

💡 Why this idea?

AI data centers are becoming increasingly power intensive.

The investment chain is evolving:

AI models → GPUs → networking → power → cooling → data centers

This creates an interesting second-order opportunity.

Instead of trying to predict which AI model wins, investors can look at the companies selling the infrastructure required by all of them.

⸻

🔎 Where I would look

1️⃣ Power infrastructure

AI data centers require enormous amounts of electricity and increasingly sophisticated power-management equipment.

2️⃣ Cooling

Higher rack densities create a physical constraint: traditional cooling becomes less effective.

3️⃣ Networking / optical

Thousands of accelerators need to communicate with one another extremely quickly.

4️⃣ Custom silicon

Hyperscalers are increasingly designing specialized chips to improve AI performance per dollar and per watt.

⸻

📈 One company I find particularly interesting: $AVGO

Broadcom isn’t simply an AI-chip story.

Its exposure spans:

Custom AI accelerators + networking + optical infrastructure

That’s important because the AI industry may eventually become less about buying the maximum number of GPUs and more about building the most efficient AI system.

My current reference data:

* Price: ~$371

* Morningstar: ⭐⭐⭐⭐⭐

* Morningstar Fair Value: ~$663

* Analyst average target: ~$523

* Analyst implied upside: ~41%

But I wouldn’t buy simply because the target says +41%.

The real question is whether custom silicon + networking + AI infrastructure growth can justify the valuation.

⸻

🐂 Bull case

If hyperscalers continue increasing AI capex while optimizing:

performance / watt + performance / dollar

Broadcom could benefit from multiple layers of the infrastructure stack.

⸻

⚠️ What could go wrong?

The biggest risk isn’t that AI disappears.

It’s that AI capex grows slower than investors expect.

Other risks:

• Hyperscaler capex cuts
• Customer concentration
• Custom-chip projects delayed
• Nvidia retaining more of the economics
• AI infrastructure returns disappointing relative to capital invested
• Valuation already pricing in substantial growth

⸻

🚨 What would change my view?

I’d become much more cautious if we saw:

AI capex ↓
Custom-chip demand ↓
Networking growth ↓

at the same time.

One weak quarter wouldn’t necessarily break the thesis.

A synchronized slowdown would.

⸻

🎯 How I’d approach it

I wouldn’t chase a stock simply because AI is trending.

I’d watch:

AI ecosystem → infrastructure bottleneck → company economics → valuation → portfolio fit

That’s the sequence I use before putting capital to work.

For someone already heavily exposed to AI/technology, adding another AI stock may actually increase concentration rather than diversification.