ResourceGeopolitics

Covering natural resources, energy markets, geopolitics & crypto.
Not financial advice.

[ResourceGeopolitics @ResGeoPol]

Markets don’t crash because bad news appears.

Markets crash when investors realize the story they believed is no longer enough to justify the price.

That’s why valuation always matters.

Even during revolutions.

[ResourceGeopolitics @ResGeoPol]

What’s the bigger risk over the next 5 years?

A) Persistent inflation

B) Recession

C) Debt crisis

D) AI bubble

My answer: The debt problem eventually amplifies all three.

[ResourceGeopolitics @ResGeoPol]

The biggest mistake investors make is assuming every market decline confirms a crash thesis.

The biggest mistake bulls make is assuming every dip is a buying opportunity.

The reality is simpler:

• Debt levels matter
• Interest rates matter
• Valuations matter
• Earnings matter

Narratives move markets in the short term.

Cash flows move markets in the long term.

The challenge isn’t predicting the next headline.

It’s separating signal from noise.

[ResourceGeopolitics @ResGeoPol]

🚨 The biggest threat to global food security in 2026 isn’t war, AI, or climate headlines.
It’s groundwater — the invisible resource under our feet that’s disappearing faster than we can replace it.
~40% of the world’s irrigated agriculture depends on it. And many key aquifers are in serious trouble.
Silent crisis thread 👇

[ResourceGeopolitics @ResGeoPol]

Everyone is obsessed with GPUs.

They shouldn’t be.

The next AI bottleneck may be a material most investors have never heard of: Indium Phosphide (InP).

And China just gained leverage over it.

🧵

AI scaling isn’t just about more GPUs.

It’s about moving insane amounts of data between them.

800G is here.
1.6T is coming.
Co-packaged optics is next.

At those speeds, copper starts becoming the problem.

Light becomes the solution.

And light needs lasers.

The catch?

The best lasers for next-gen AI networking depend on InP.

That’s where the supply chain gets interesting.

China controls a large share of upstream indium production.

Export restrictions and licensing delays are already creating uncertainty across the photonics ecosystem.

Meanwhile hyperscalers are racing to build larger AI clusters.

More GPUs = more optical links.

More optical links = more lasers.

More lasers = more demand for InP.

Simple equation.

The market is focused on:

→ GPUs
→ HBM
→ Power

But very few people are looking at the laser stack underneath the AI buildout.

The supply chain is surprisingly concentrated:

• InP substrates
• Epitaxy
• High-performance lasers
• Optical engines
• Co-packaged optics

Each layer has only a handful of qualified suppliers.

And qualification cycles can take years.

That’s what creates bottlenecks.

Not demand.

Qualified supply.

Names I’m watching:

• Sivers Semiconductors
• IQE
• AXT
• Soitec
• LPKF
• Riber

Not because they’re guaranteed winners.

Because they sit close to a constraint the market may be underestimating.

The biggest question:

What’s the real AI chokepoint over the next 3 years?

GPUs?
HBM?
Power?
Cooling?
Or InP-based photonics?

My view:

The market still underestimates photonics.

Interested in counter-theses.

What am I missing?

[ResourceGeopolitics @ResGeoPol]

Everyone is talking about Transformer models.

Almost nobody is talking about transformers.

That may be a $100B mistake.

🧵

The biggest bottleneck in AI might not be GPUs.

It might not be memory.

It might not even be power generation.

It may be the giant electrical transformers required to connect AI data centers to the grid.

No transformer.

No power.

No AI cluster.

No revenue.

Here’s the problem:

Building more GPUs takes months.

Building large power transformers can take years.

Meanwhile, every hyperscaler is racing to deploy larger and larger AI clusters.

The market sees NVIDIA.

The grid sees a completely different problem.

Millions of AI chips are useless if they can’t be connected to reliable power.

That’s why I think investors are looking at the wrong layer of the stack.

Everyone is focused on intelligence.

Very few are focused on infrastructure.

The biggest winners of the AI boom may not be the companies training models.

They may be the companies enabling electrons to reach the data center in the first place.

The AI trade is slowly becoming a power infrastructure trade.

What am I missing?

[ResourceGeopolitics @ResGeoPol]

@Justfacts1976 Mainly Hitachi Energy, Siemens Energy and GE Vernova for the big transformers. Also tracking Eaton, Hyundai Electric and Virginia Transformer. Solid-state startups (Heron, DG Matrix) could be interesting disruptors. Lead times are still brutal though.

[ResourceGeopolitics @ResGeoPol]

Everyone is talking about Transformer models.

Almost nobody is talking about transformers.

That may be a $100B mistake.

🧵

The biggest bottleneck in AI might not be GPUs.

It might not be memory.

It might not even be power generation.

It may be the giant electrical transformers required to connect AI data centers to the grid.

No transformer.

No power.

No AI cluster.

No revenue.

Here’s the problem:

Building more GPUs takes months.

Building large power transformers can take years.

Meanwhile, every hyperscaler is racing to deploy larger and larger AI clusters.

The market sees NVIDIA.

The grid sees a completely different problem.

Millions of AI chips are useless if they can’t be connected to reliable power.

That’s why I think investors are looking at the wrong layer of the stack.

Everyone is focused on intelligence.

Very few are focused on infrastructure.

The biggest winners of the AI boom may not be the companies training models.

They may be the companies enabling electrons to reach the data center in the first place.

The AI trade is slowly becoming a power infrastructure trade.

What am I missing?

[ResourceGeopolitics @ResGeoPol]

@aleabitoreddit Data centers are just one piece of the puzzle - groundwater depletion is already a quiet crisis threatening global food security. Around 40% of the world’s irrigation relies on it.

[ResourceGeopolitics @ResGeoPol]

🚨 The biggest threat to global food security in 2026 isn’t war, AI, or climate headlines.
It’s groundwater — the invisible resource under our feet that’s disappearing faster than we can replace it.
~40% of the world’s irrigated agriculture depends on it. And many key aquifers are in serious trouble.
Silent crisis thread 👇

[ResourceGeopolitics @ResGeoPol]

Everyone is obsessed with GPUs.

They shouldn’t be.

The next AI bottleneck may be a material most investors have never heard of: Indium Phosphide (InP).

And China just gained leverage over it.

🧵

AI scaling isn’t just about more GPUs.

It’s about moving insane amounts of data between them.

800G is here.
1.6T is coming.
Co-packaged optics is next.

At those speeds, copper starts becoming the problem.

Light becomes the solution.

And light needs lasers.

The catch?

The best lasers for next-gen AI networking depend on InP.

That’s where the supply chain gets interesting.

China controls a large share of upstream indium production.

Export restrictions and licensing delays are already creating uncertainty across the photonics ecosystem.

Meanwhile hyperscalers are racing to build larger AI clusters.

More GPUs = more optical links.

More optical links = more lasers.

More lasers = more demand for InP.

Simple equation.

The market is focused on:

→ GPUs
→ HBM
→ Power

But very few people are looking at the laser stack underneath the AI buildout.

The supply chain is surprisingly concentrated:

• InP substrates
• Epitaxy
• High-performance lasers
• Optical engines
• Co-packaged optics

Each layer has only a handful of qualified suppliers.

And qualification cycles can take years.

That’s what creates bottlenecks.

Not demand.

Qualified supply.

Names I’m watching:

• Sivers Semiconductors
• IQE
• AXT
• Soitec
• LPKF
• Riber

Not because they’re guaranteed winners.

Because they sit close to a constraint the market may be underestimating.

The biggest question:

What’s the real AI chokepoint over the next 3 years?

GPUs?
HBM?
Power?
Cooling?
Or InP-based photonics?

My view:

The market still underestimates photonics.

Interested in counter-theses.

What am I missing?

[ResourceGeopolitics @ResGeoPol]

Follow @ResGeoPol for more on critical resources, energy markets, and real geopolitics.
RT if this made you rethink food security.
Save for later — this topic isn’t going away.

[ResourceGeopolitics @ResGeoPol]

🚨 The biggest threat to global food security in 2026 isn’t war, AI, or climate headlines.
It’s groundwater — the invisible resource under our feet that’s disappearing faster than we can replace it.
~40% of the world’s irrigated agriculture depends on it. And many key aquifers are in serious trouble.
Silent crisis thread 👇

[ResourceGeopolitics @ResGeoPol]

This isn’t a distant 2050 problem. Depletion is accelerating in real time.
What do you think?
Is groundwater the most under-discussed risk to global stability right now?
Drop your thoughts below 👇

[ResourceGeopolitics @ResGeoPol]

@LynAldenContact Exactly.
Nominal GDP in dollars grows mainly due to money printing. In gold (a hard, non-manipulable measure), the real picture after 2000 looks much weaker.

[ResourceGeopolitics @ResGeoPol]

Scandium just leveled up the game.
@SunriseMetals invests in @agnisemi to develop AlScN ferroelectric diode memory chips that stay stable and non-volatile up to 600°C (vs ~200°C for conventional chips). Ultra low power, perfect for AI compute-in-memory, geothermal, hypersonics, satellites and more.
@robert_ivanhoe connects the dots straight to his G-Pulse deep geothermal drilling tech — deeper access to abundant clean baseload power.
This is the real flywheel: critical minerals → extreme semiconductors → efficient AI → more clean energy.
The materials revolution is quietly rewriting what’s possible.
What other hidden synergies do you see in the critical minerals stack?

[Robert Friedland @robert_ivanhoe]

We are excited and privileged to have invested in Agni Semiconductors LLC.

@agnisemi is a US­-based, private company developing the next-generation of non-volatile memory chips using ferro-diodes manufactured with aluminium scandium nitride (AlScN).

Normal memory chips disintegrate at 200°C. Agni are developing chips that operate at up to 600°C, making them ideal for heavy duty, high temperature applications, such as geothermal drilling, hypersonics, ballistics and satellites.  As a form on non-volatile memory, these chips are extremely energy-efficient, addressing one of the key physical constraints to AI, which is energy consumption.

The investment by @SunriseMetals will increase its exposure to emerging scandium-enabled semiconductor technologies and advances our strategy of participating in the downstream value chain for scandium produced from our 100% owned Syerston Scandium Project in Australia.

The successful development of Agni’s technology will also have significant benefits in developing our G-Pulse geothermal drilling technology, which we are advancing at @IPulseGroup. This technology will enable us to drill deeper into our earths crust, unlocking significant, abundant geothermal electrical power in places where we currently can’t reach.