To The Moon

I started this rabbit hole looking at GPUs.

Somehow I ended up researching zinc refineries.

Welcome back to ADZ5150. 😂🐅

Everyone knows the AI arms race.

NVIDIA GPUs.

Hyperscaler capex.

Gigawatt data centres.

Power shortages.

Memory.

Cooling.

But something caught my attention.

In March, NVIDIA announced a US$2 billion investment in Lumentum.

On the exact same day, it announced another US$2 billion investment in Coherent.

US$4 BILLION.

Not into another AI model.

Not another GPU company.

Into companies that make lasers and optical technology.

And both agreements went beyond investments.

They included multibillion-dollar purchase commitments and future access to manufacturing capacity.

That made me ask the obvious GAZADZ question:

🐇 WHY DOES THE WORLD’S MOST IMPORTANT AI CHIP COMPANY CARE SO MUCH ABOUT LASERS?

The answer sent me much further down the supply chain than I expected.

⸻

🧠 THE GPU IS ONLY HALF THE PROBLEM

Building a giant AI cluster isn’t simply about stacking more GPUs into a building.

Those GPUs need to communicate.

Constantly.

And as clusters scale from thousands toward potentially enormous numbers of accelerators, moving data between processors, racks and data centres becomes increasingly difficult.

Copper electrical connections are brilliant over shorter distances.

But eventually distance, bandwidth and power consumption become problems.

That’s where light enters the story.

💡 Optical interconnects.

Instead of moving all that information electrically, increasingly sophisticated AI networks move enormous quantities of data using photons through fibre.

Suddenly the AI arms race isn’t only about:

How fast can we compute?

It’s also:

How fast can we move the result?

⸻

🔦 NVIDIA ISN’T WAITING AROUND

This is where things get interesting.

NVIDIA didn’t simply place bigger optics orders.

It put US$4 billion directly into Lumentum and Coherent, while securing purchase commitments and future capacity rights.

Then this week NVIDIA went further.

It announced a huge supply-chain expansion across substrates, packaging, testing and optical engines.

For co-packaged optics specifically, NVIDIA and SPIL plan to increase optical-engine manufacturing capacity by approximately:

15× BY MID-2027.

And they’re targeting another doubling by the end of 2028.

Jensen Huang’s explanation was beautifully simple:

“Capacity is the constraint.”

That sentence got my attention.

Because suddenly this doesn’t look like optics merely benefiting from AI.

It looks like NVIDIA is actively trying to prevent optics from becoming a constraint on AI.

⸻

🚨 AND LUMENTUM IS ALREADY TALKING ABOUT SHORTAGES

Earlier this month, Lumentum CEO Michael Hurlston said some of the company’s optical-component capacity was effectively sold out through early 2029.

For certain products, the company reportedly cannot satisfy around 70% of expected demand next year.

Read that again.

We’re talking about infrastructure required to connect the machines powering the AI boom.

And customers are already fighting over capacity years into the future.

So naturally I kept digging.

And that’s where this rabbit hole got weird.

⸻

🔬 MEET INDIUM PHOSPHIDE

Many of these advanced optical systems require specialised lasers.

Those lasers can use a compound semiconductor called:

INDIUM PHOSPHIDE, or InP.

Silicon is fantastic at computing.

It isn’t particularly good at emitting light.

InP is.

Coherent operates what it describes as the world’s first and largest volume-production 6-inch indium-phosphide manufacturing platform in Texas.

Most global InP manufacturing historically operated on smaller wafers.

Moving toward larger wafers means more devices per manufacturing run and potentially much greater scale.

Coherent is expanding.

Lumentum is expanding.

NVIDIA is helping fund capacity.

The US government is even supporting domestic InP manufacturing expansion through the CHIPS program.

At this point I thought I’d reached the bottom of the rabbit hole.

😂

Nope.

⸻

⛏️ WHERE DOES THE INDIUM COME FROM?

This is where an AI story suddenly turns into a critical-minerals story.

According to the US Geological Survey, China accounts for roughly 70% of global refined indium production.

The United States recorded 100% net import reliance for indium in 2025.

And China introduced new export restrictions covering indium in February 2025.

There’s another complication.

Indium generally isn’t mined from giant dedicated indium mines.

It’s primarily recovered as a by-product of processing other metal ores, particularly zinc.

Meaning if AI suddenly needs dramatically more InP, supply can’t necessarily respond by simply opening another pure indium mine.

And THAT is how I started researching GPUs and somehow ended up staring at zinc refineries. 😂

⸻

🐅 THE ADZ5150 X → Y → Z

❌ X: WHAT THE MARKET IS WATCHING

Who makes the fastest GPU?

How many Blackwells can NVIDIA ship?

How much are Microsoft, Meta, Amazon and Google spending?

How many gigawatts will AI require?

⸻

👀 Y: WHAT MAY BE OVERLOOKED

More GPUs create another problem:

moving enormous amounts of information between them.

As AI clusters become larger, optical connectivity becomes increasingly important.

That requires lasers.

Those lasers require specialised compound-semiconductor manufacturing.

And NVIDIA is now spending billions securing that ecosystem while suppliers race to expand capacity.

⸻

💡 Z: THE UNEXPECTED CONSEQUENCE

Part of the physical ceiling on the AI boom may eventually sit surprisingly far away from the GPU itself.

Potentially all the way down here:

GPU

⬇️

NETWORK

⬇️

OPTICAL INTERCONNECT

⬇️

LASER

⬇️

INDIUM PHOSPHIDE

⬇️

INDIUM

⬇️

GLOBAL CRITICAL-MATERIAL SUPPLY CHAIN

The world’s most advanced AI infrastructure may ultimately depend partly on a metal most investors have barely thought about.

That’s pretty fascinating.

⸻

🇺🇸 AND NOW THERE’S A GEOPOLITICAL TWIST

The US wants more domestic AI infrastructure.

It wants resilient semiconductor supply chains.

It wants advanced manufacturing brought back onshore.

Coherent is expanding InP manufacturing in Texas.

Lumentum is increasing US capacity.

NVIDIA is helping fund the buildout.

Yet the upstream indium market remains heavily concentrated in China.

Reuters has already reported that Chinese InP export licensing has created delays and attracted attention at senior levels of US-China trade discussions.

So this isn’t merely:

“Optics demand is growing.”

The more interesting question is:

Can America build a sovereign AI supply chain when obscure upstream materials still pass through concentrated global supply chains?

That is a very different AI thesis.

⸻

⚠️ NOW FOR THE BEAR CASE

Before everybody runs out and searches:

“BEST INDIUM STOCK 🚀🚀🚀”

😂

Slow down.

This thesis can absolutely be wrong.

Capacity is responding aggressively.

Coherent is expanding 6-inch InP manufacturing.

Lumentum is expanding manufacturing.

NVIDIA and SPIL are targeting a gigantic increase in optical-engine capacity.

Alternative indium suppliers exist outside China.

New networking architectures can reduce component requirements.

Engineering improvements could dramatically improve efficiency.

Today’s shortage could eventually become tomorrow’s oversupply.

And perhaps most importantly:

identifying a bottleneck does NOT automatically identify the winning investment.

The economics could accrue to laser manufacturers.

Or substrate producers.

Or optical-module companies.

Or packaging.

Or testing.

Or perhaps nobody earns extraordinary returns because NVIDIA and its partners expand supply fast enough.

That’s the next rabbit hole.

⸻

🐯 MY TAKE

The first AI trade was obvious:

Buy the chips.

Then investors discovered AI needed:

⚡ electricity

🏗️ data centres

💳 enormous amounts of capital

🌐 networking

Now another layer is emerging.

LIGHT.

NVIDIA investing US$4 billion into optics companies isn’t proof that we’re about to run out of lasers forever.

But it is a pretty enormous clue.

And NVIDIA saying “capacity is the constraint” while helping expand optical-engine manufacturing 15× deserves attention.

The most interesting part of AI investing right now isn’t asking which company says “AI” the loudest.

It’s following the supply chain until you find the thing everybody suddenly realises they need.

This time…

the rabbit hole went:

GPU → LIGHT → LASER → InP → INDIUM → ZINC REFINERY.

😂🐅💡

I definitely didn’t expect that when I started.

And that’s exactly why I kept digging.

⸻

🗳️ YOUR TAKE

Where does the real AI-optics opportunity sit?

🔵 A: Lumentum / Coherent and laser manufacturing

🟢 B: Optical networking and modules

🟡 C: InP / critical-material supply chains

🔴 D: Nowhere, capacity catches up and the bottleneck disappears

I’m still digging.

🐅 ADZ5150

Just my research and opinion. Not financial advice. Do your own research.