What Is Zuckerberg's Bunker For?

Mark Zuckerberg says he's building a storm shelter.

What do you think?

His "storm shelter" is underground in a 1,400-acre Kauai compound.

It has a tunnel. An escape hatch. Its own clean water system. Food production. A steel door filled with concrete.

My storm shelter has a golf sim in it and is also called the "garage."

So naturally, I have a question: What is Zuck worried about?

More importantly, what are the people closest to AI, the ones seeing what's happening inside the labs first, worried about?

I think this story takes us somewhere much more useful as investors than a doomsday fantasy...

Because one of the biggest things happening inside those labs right now is this: AI is learning to see.

The Optimists and The Pessimists

As an optimist...

I'll start with everyone's favorite polarizing one.

Elon Musk sees a future of robots and abundance. More robots than people. Robots doing all the work. Humans pursuing passions.

That's one end of the spectrum.

We saw how Zuckerberg is prepping.

Sam Altman, the CEO of OpenAI, is more interesting.

Back in 2016, Altman revealed to The New Yorker how he was preparing for social upheaval.

Guns. Gold. Potassium iodide. Antibiotics. Batteries. Water. Gas masks.

And land in Big Sur he could fly to.

Sensible. I've read The Black Swan and prepped too... But my getaway isn't on one of the most beautiful coastlines on Earth.

In 2023, Sam was asked about bunkers again.

He said, "None of this is gonna help if AGI goes wrong."

And just two weeks ago he urged all of the labs to slow down AI progress to focus on safety (more on this in a future issue).

Finally, let's check in with Silicon Valley titan Peter Thiel. He has more escape hatches than a Range Rover has warning lights...

In New Zealand, Uruguay, Argentina and next door to Zuck in Hawaii.

Why? Last fall Thiel delivered four sold-out lectures in San Francisco about...

The Antichrist.

His argument is not "AI is the Antichrist." It's the opposite.

Thiel worries that nonstop fear of existential risks, from AI to nuclear war, could become the excuse for a global authority promising peace and safety in exchange for control...

And that outcome could freeze technological progress.

Finally, there's Dario Amodei, who runs Anthropic. No bunker that I can find.

But he runs a company and product that has control of the future of billions of people's lives.

Anthropic and Claude.

In 2025, he warned of a "White Collar Wipeout" that could eliminate half of all entry-level white-collar jobs... And push unemployment to 10% to 20% within one to five years.

He has toned things down a bit recently.

(I'm guessing his investors gave him the talk.)

Five men. Five very different pictures of the future.

What they all agree on is ONE thing: Intelligence is going exponential and the world WILL fundamentally change.

The First Robots Don’t Walk

Amodei's warning points to a topic most people fail to understand. The first robots coming for white-collar jobs don't walk.

If your job happens inside email, Excel, Salesforce, Slack, GitHub, Adobe or a browser tab...

AI doesn't need a body to replace you. Think about the contrast with our discussion last Thursday.

Optimus needs motors, batteries, hands and a factory. Claude or ChatGPT just needs an update.

Humanoid robots will take years to find all of their homes in the world's economy.

But a new model from Anthropic or OpenAI can show up on billions of devices tomorrow that's better at a task than every human in the world today.

And we're starting to see the cracks.

Researchers at Stanford's Digital Economy Lab studied payroll data covering millions of U.S. workers.

They found no evidence of widespread, economy-wide job loss from AI yet.

But among workers ages 22 to 25 in the jobs most exposed to AI... Employment was running almost 20% below where it would have been.

Experienced workers in those same jobs didn't show the same gap. That makes sense when you think about the repetitive tasks junior knowledge workers do.

Research. Drafting. Basic coding. Data cleanup. Customer support. Routine analysis.

That's exactly the work AI replaces first.

Which leads to the part that should keep a lot of people up at night...

AI may not be replacing the senior employees yet... But it's now replacing junior employees.

The Physical World Has a Password

White-collar work is already digital.

The spreadsheet is data. The email is data. The contract is data.

AI can read it, change it and send it without ever touching anything.

Physical work isn't the same.

Remember when we were thinking about what an Optimus robot would have to do to get me a 10mm socket?

Before it could hand it to me, it had to answer a surprisingly long list of questions.

Where is it? Is it 1/4 drive or 3/8? Regular or impact? How far away is it? Can I reach it? Is something blocking it? How should I grab it?

You, fellow human, answer all of that on autopilot.

A robot has to work it out from raw data, every single time, in a world that constantly changes.

That's the password to the physical world.

Before AI can replace much of physical labor, it must see.

So... Language gave AI access to information. Agents give AI access to digital work. Vision gives AI access to the physical world.

Each step unlocks the next.

We're finishing the second step now... The third is where the next wave of money gets made.

Physical Work is a Vision Problem

In robotics, hands, battery and safety are hard...

But vision is the key missing piece.

A useful physical machine needs a stack of things working together.

Cameras to capture light. Depth sensors to know how far away things are. Lidar, which bounces laser pulses off the world to build a precise 3D map. Radar, which keeps working in dust, dark and rain.

Then you have to merge all those signals into one coherent picture.

Engineers call this sensor fusion.

Finally, the picture has to be processed right there on the robot, not in a data center 800 miles away, because a robot can't wait for a cloud server to tell it what to do next.

That's edge AI: specialized chips doing the processing locally.

Then perception software turns pixels into meaning: That's a socket. That's my hand. That's a jack stand.

But recognizing the world isn't enough.

The robot also has to predict how the world will change when it acts.

If I move my hand left, do I grab the socket... knock it off the bench... or knock over the can of oil?

That's where a world model comes in: an internal model of the scene that helps the robot anticipate what happens next.

Human workers come with two cameras, plus a pretty good processor right behind them. Robots have to "buy" theirs. And robots don't have to settle for our model of the world.

Human eyes pick up a thin sliver of the electromagnetic spectrum, the band we call visible light. That's it. No heat. No ultraviolet. No way to see through smoke or measure distance with a laser.

A machine can "shop" for all of those. And more.

Which means we aren't only teaching AI to see like us. We may be giving it better eyes than we have.

What Should Investors See?

So, as investors worried about the future of intelligence and robotics... what do we do?

A good start: Buy Tesla (TSLA), as I wrote last week. Plus NVIDIA (NVDA), too.

Both of these companies will be at the core of this transition, and barring a black swan event, they will be more valuable in 10 years than they are today.

But to build life-changing wealth in a short period of time, you need to uncover tomorrow's AI moonshots.

Unfortunately, most of the real moonshot opportunities are still private.

Still, I went looking for publicly traded companies that could benefit in the coming years...

And found a few worth looking at.

So if the next phase of AI is machines that see, understand and act, somebody should be getting paid to provide the eyes, the vision processors and the perception software.

Smaller and less obvious businesses selling three things robots are going to need:

1. Eyes that see more than ours

2. Chips that make sense of what those eyes capture

3. And software that turns sight into an understanding of the world.
   I found seven U.S.-traded names worth keeping an eye on.

None are official recommendations. But several are much more interesting than I expected.

So, today we asked what these billionaires are worried about: Rapid change.

And on Thursday we'll look for opportunities to profit from it...

By looking into seven publicly traded companies that help robots see.

Always be prospering,

Harry Seldon

🔓 Premium Content Begins Here 🔒