7 Robotics Stocks I'm Looking At

On Tuesday, we saw what the AI billionaires see coming... from robot utopia to AI apocalypse.

One thing they all agree on: Rapid change is coming.

Today we're going to see how we could make money from any coming changes so we can fund our dream doomsday shelter...

As intelligence becomes ubiquitous and takes on human form.

Remembering our discussion last week...

If I look at my garage floor and see my lost 10mm socket, I simply pick it up without thinking about it.

Not so with robots.

A robot has an entire technical pipeline. Light hits a sensor and becomes millions of numbers. Software decides the blob is a socket. A model predicts what happens if the hand reaches left instead of right.

This chain is where stocks that could become the next trillion-dollar market cap company live.

Giving robots eyes is hard, but giving them eyes that can see more than we can, a brain that understands what they see, and the ability to act on what they see is the path to life-changing wealth.

Don’t Bet Only On The Robot

Nobody knows who the humanoid winners will be.

Tesla, Figure, Unitree, Boston Dynamics or one of a handful of private companies.

The robot race has barely started.

Chinese companies shipped more than 90% of the world's humanoids in the first half of this year, while Tesla, Figure and Agility each shipped roughly 150 last year.

Tesla is building production lines designed for a million Optimus robots a year, and eventually ten million, but it hasn't said how many it will actually ship in 2027 or 2028.

But where the U.S. leads is what you won't see in an Internet demo video: the brains and the eyes.

The Robot I Already Own

I've already placed my big humanoid bet.

It's Tesla (TSLA), and I bought it to hold for ten years.

Why?

A useful humanoid isn't ONE hard problem. It's five hard problems that need to be solved together.

1. Integrate roughly 10,000 new parts.

2. Build enough initial robots to train. The first robots off of Tesla's new Fremont line aren't even going to customers! They're going to what Tesla calls the Optimus Academy, where 10,000 to 30,000 robots will live full-time and practice what humans do to collect training data.

3. Turn that data into a world model. A model that can be used as its brain for real work.

4. Run that model on a low-power chip inside the robot. Tesla designs its own, called AI5, with volume expected in 2027.

5. Then, finally, produce Optimus on assembly lines by the millions.
   Tesla is working on the entire stack.

Almost every company on this watchlist below works on just one slice of that.

To be clear up front… This will take a few years.

Musk admits production "will be extremely slow at first."

Last month, Tesla's own AI chief admitted the flashy humanoid moves you see online are "essentially remote controlled."

So why am I still confident?

Because I think the real prize is the world model.

Look at chatbots like ChatGPT, Grok or Claude. The best models require capital. The most recent version of Grok, for example, cost $400 to $500 million.

Future models will cost $1 billion or more.

Being fourth-best at intelligence means trillions in lost market opportunities.

I think robots will repeat that same pattern.

The body and eyes are hard...

But whoever builds the best brain (world model), the one that actually understands a messy garage, will capture most of the value.

Now, Tesla owns its stack, so it's not likely to be a big customer for any stock below.

So the real question for each stock below is: Can it become a go-to supplier for the second- or third-place winner?

To companies that can't afford to build everything in-house?

But first, we need to screen for them, meaning...

What does this company own that a robot maker would rather buy than rebuild?

That kills a lot of ideas fast. A plain sensor anyone can manufacture will get commoditized.

Prices fall, margins shrink, and the robot maker switches suppliers to save a few dollars.

The survivors fell into three layers...

Layer 1: Hardware: Better-Than-Human Eyes

Human eyes work in one thin band of the electromagnetic spectrum. Machines don't have that limit.

Teledyne (TDY)

For the purest version of "machines that can see what humans can't," Teledyne is the perfect match.

Its digital imaging business builds sensors and cameras across the visible, infrared, ultraviolet and X-ray spectra.

It owns FLIR, the thermal camera company that lets machines see in total darkness.

You may have one of its scopes... and its infrared detectors are on satellites and drones.

The catch: Teledyne is big and diversified, with defense at 30% to 35% of revenue.

You aren't buying a pure physical-AI play. You're buying a lower-risk full-spectrum company with a lot of other businesses attached.

Aeva (AEVA)

Ordinary lidar tells you where something is.

Aeva's 4D lidar also measures how fast every point in space is moving, instantly.

You and I estimate speed by seeing movement. Aeva's sensor reads it directly.

It's early, though.

First-quarter revenue was $6.3 million, with a GAAP net loss of about $35 million.

Layer 2: Hardware and IP: The Visual Cortex

A robot reaches for a socket. The socket slips. It has milliseconds to react.

It can't wait for a 4K video to be uploaded to the cloud and processed.

The thinking has to happen on the machine.

Ambarella (AMBA)

Ambarella may be the best fit in this layer.

It designs chips that process video and run AI locally at very low power.

It converts stereo vision into a full 3D representation of an environment in real time.

Its new CV7 chip, launched in January, runs AI vision models across multiple cameras at once.

But the potential moat isn't the camera...

It's efficient visual intelligence at the edge. But robotics is still a small slice of Ambarella's business.

So, robotics is an option on Ambarella's future.

CEVA (CEVA)

CEVA designs building blocks that other companies put inside their chips, then collects a royalty on every one that ships.

More than 20 billion devices carry CEVA technology.

For robots, the blocks that matter are its AI processor designs, which run vision models on the device, and its sensor-fusion technology, which helps a machine combine what its cameras, motion sensors and other inputs are telling it.

That answers the obvious objection to every chip name here: "Won't big robot makers design their own chips?"

CEVA gets paid a licensing fee up front, but the real money comes from royalties once those chips ship in volume...

But those can take years.

Cognex (CGNX)

Cognex already sells machine vision to factories, and its planned $500 million acquisition of RealSense would give it a bigger role in helping robots navigate them.

RealSense's depth cameras are already used by robots to map their surroundings and move autonomously.

The potential moat here is all of the software developers who already build around those cameras.

Once a robot's vision system depends on that integration, switching costs can be high.

That creates an appealing possibility: supplying the eyes to competing robot makers as the entire industry grows.

Ouster (OUST)

I looked at Ouster and passed on it last year.

It was a lidar company, and I think standalone lidar gets commoditized.

This is why you follow up on your watchlists.

In February, it bought Stereolabs, a 3D camera company with more than 90,000 ZED cameras shipped to more than 10,000 customers.

Ouster now sells lidar, stereo cameras, onboard AI compute and perception software as one stack.

In August, Trossen Robotics started building Stereolabs cameras into its robot-learning platforms, specifically to capture visual training data for teaching robots new skills.

Now, Ouster isn't a world-model company yet. It simply detects, classifies and tracks objects. That's the eyes, not the brain that predicts what happens next.

But it could become one.

Layer 3: The World Model (The Moonshot Opportunity Hot Zone)

A world model is the robot's working sense of what's around it and what happens next if it acts.

Besides the Mag7 companies, there are a few well-funded private companies working on this...

Including Yann LeCun's AMI Labs and Fei-Fei Li's World Labs.

This is the layer I care about most. This is where the next OpenAI or Anthropic will come from. This is also where you need billions of dollars of capital to even have a chance.

Mobileye (MBLY)

Mobileye's chips and cameras sit in millions of cars with driver-assistance systems.

Through a system called REM, they process what they see onboard, send tiny summaries to the cloud (about 10 kilobytes per kilometer) and feed a continuously updated map Mobileye calls the Roadbook.

It captures not just lanes and signs but how people actually drive each road.

See. Process at the edge. Build a representation of the world. Remember it. Use it to make future decisions.

That's much closer to a world model than any other company on this list.

And it has a real data loop. More cars create more observations. More observations make the map better. A better map makes the system more useful.

The stock is down 45% over the past year because the market doesn't see a future in this.

But, in January, Mobileye made an unexpected pivot away from just being a car-chip company... It bought Mentee Robotics, a humanoid startup, for $900 million.

Mentee's robots are designed to learn new tasks from a few human demonstrations.

Production is targeted for 2028.

Can a company that taught millions of cars to model the road teach robots to model a home?

They do have about $2 billion in revenue a year, and if we see them raise capital, they can make a run for it.

So, we'll keep an eye on them.

My Watchlist

Here are the seven, ranked by how I see them today.

Our rule is simple. The closer a company gets to the world model, or the more robot makers need specific hardware no matter who wins, the higher it ranks.

1. Mobileye (MBLY): The closest thing to a world model here, plus its own humanoid. The closest Tesla competitor on the list, but that's not saying much.

2. Ambarella (AMBA): The edge brain for every robot maker that doesn't design its own chips.

3. Ouster (OUST): Lidar won't win Tesla's business, but plenty of other robot makers want its stack.

4. Cognex (CGNX): The safest name here, and the least exciting.

5. CEVA (CEVA): They are a pure IP play. Licenses come first; royalties can take years.

6. Teledyne (TDY): The best eyes on the list, but robotics is currently a small part of a big company.

7. Aeva (AEVA): Breakthrough technology, but unproven business.
   I don't know for certain which humanoid becomes one of the top three other than Tesla (TSLA).

In fact, I'm more excited about what some privately held companies are working on.

And that's fine.

Sometimes the best investment isn't buying early. It's seeing early.

The real advantage is understanding the hidden structure of technology before the investment opportunity is obvious...

And waiting until the technology hits a tipping point.

So, for now, these seven are on my radar.

None are picks. Not yet.

And I'll keep my eye on them for you.

Until next week...

Always be prospering,

Harry Seldon

🔓 Premium Content Begins Here 🔒