The "Picks and Shovels" Play for the Age of Optimus

Most people missed this about the Optimus Revolution.

Elon Musk was on a Hollywood stage in October, walking through Tesla's brand-new AI5 chip — the brain going into the next generation of cars and the Optimus humanoid robot.

Buried in the middle of his pitch, he named the components going into the chip. Memory. Accelerators. PCI-X blocks.And one more thing. A name he said in passing, like it was nothing.

But that one name is the entire reason I'm writing this.

Because it isn't Tesla. It isn't Nvidia. It isn't any company most investors associate with the robot story. It's a business that gets paid every single time Tesla builds an Optimus — and every time Figure, Boston Dynamics, 1X, or Apptronik builds theirs too. It’s hiding in plain sight. And almost nobody is talking about it.

Listen, Tesla just confirmed Optimus production starts at Fremont this July or August. [3] The company is killing the Model S and Model X to free up the line.

That's a wild moment. Tesla is shutting down its flagship sedan to build robots.

But here's the thing. Everyone already knows the Optimus story. Tesla stock has been pricing in the robot dream for two years. Every Musk tweet about it gets dissected by ten thousand analysts.

You're not getting a deal at $400 a share.

Today, I will share two plays that are the "picks and shovels” plays in the Optimus revolution.

Let's start with ARM Holdings (ARM).

Most people think ARM is a chip company.

It's not. Not really.

ARM doesn't make a single chip. The company designs the blueprint — the underlying architecture that other companies use to build their own chips. Apple uses ARM in every iPhone. Nvidia uses ARM in its Grace data center chips. Samsung. Qualcomm. Amazon. Google. They all use ARM.

Every time one of them ships a chip, ARM gets a small slice off the top as royalty.

Now apply that to humanoid robots.

Tesla's AI4 chip — the brain currently running inside every Optimus prototype — has 20 ARM Cortex-A72 cores.[4] The AI5 chip taping out right now will use ARM cores too. So will AI6. Musk has confirmed that AI6 is no longer planned for cars at all. [5]. It's going straight into Optimus and Tesla's data centers.

So Tesla pays ARM every single time it builds a robot.

But it gets better. Because ARM isn't just inside Tesla.

Look at any other humanoid robot project on the planet. Figure AI. Boston Dynamics. 1X. Apptronik. Agility. Unitree. Sanctuary. They almost all run on ARM-based chips.

It doesn't matter which one wins.

ARM gets paid either way.

The stock isn't cheap.

It trades at a premium because Wall Street already knows it's a great business. But the robot angle is barely priced in. Most analysts model ARM as a phone-and-laptop story. The robot side may not be fully priced in yet.

That's where the upside lives.

Now here's the under-the-radar one.

A robot that can't see is a useless $20,000 technology.

So Optimus has cameras. Six of them, based on Tesla's patents. [6] And those cameras feed into chips that turn raw images into decisions in milliseconds.

If a robot drops a coffee cup, it can't send the signal to the cloud and wait for a reply. The cup will already be on the floor. The robot has to think on its own right there.

That's called "edge AI."

And it's brutally HARD. You need a chip that uses very little power, fits in a tight space, and processes video fast enough to run a humanoid in real time.

There's a small company in Santa Clara that's been quietly building exactly this kind of chip for almost twenty years.

It is Ambarella.

A lot of folks know Ambarella from the early GoPro days. Older investors might remember the stock running from $10 to over $130 a few years back. Then it crashed. Then it sat in the bargain bin while everyone chased Nvidia.

Quietly, Ambarella pivoted hard.

The company now makes low-power vision chips designed specifically for robots, drones, and self-driving systems [7]. Their newest chip can run a vision model that used to need a desktop GPU. It does it on a piece of silicon the size of a coin.

Here's the kicker. Ambarella's chips are also ARM-based.

So when an Ambarella vision chip ships inside a humanoid robot, two things happen at once. Ambarella books the sale. ARM books a royalty.

It's the same play stacked twice.

And Ambarella isn't the next Nvidia. Nobody is. Nvidia is a $4 trillion company.

But Ambarella's market cap is a tiny fraction of that. The whole company is worth less than what Tesla spent building one factory line. If Ambarella's chips end up inside even three or four of the major humanoid platforms, that changes everything for the stock.

When a new technology takes off, the biggest gains usually don't go to the brand on the box. They go to the parts inside.

Cisco didn't make websites. It built the routers that ran the internet. The stock went up 100x in the dot-com era while everyone else was buying Pets.com.

Qualcomm didn't make phones. It made the modems inside them. The stock printed shareholder money for a decade.

Nvidia didn't make LLMs. It made the chips the LLMs ran on. We all know how that ended.

ARM and Ambarella are running the same play, twenty years later, on a different technology.

ARM is the architecture every chip designer needs. Ambarella is the vision processor that lets a humanoid robot avoid breaking your dishes.

[1] Notebookcheck, "Tesla AI5 FSD computer to run inference 10x cheaper than Nvidia AI chips," October 23, 2025. Direct quote from Elon Musk's AI5 walkthrough.

[2] InvestorPlace, "Humanoid Robot Stocks: CES 2026 Reveals Commercial Breakout," January 15, 2026.

[3] Electrek, "Tesla pushes Optimus V3 reveal later this year — again," April 22, 2026. Coverage of Tesla's Q1 2026 earnings call, including Musk's "literally impossible to predict" quote.

[4] Wikipedia, "Tesla Autopilot hardware," accessed April 2026. Cross-referenced with RobotDyn, "Tesla's AI4 Chip: The Redundant Brain Behind FSD and Optimus," February 21, 2026.

[5] Electrek, "Tesla announces HW4 Plus with doubled memory — will HW4 follow HW3 to the grave?," April 23, 2026. Confirms AI6 is no longer planned for vehicles, only Optimus and data centers.

[6] Electric Vehicles, "Tesla Files Patents Revealing Optimus Gen 3 Mechanical Blueprint," April 2026. Cross-referenced with SETI Park's analysis on X documenting Optimus's six-camera vision array.

[7] Ambarella corporate disclosures and product documentation. Company has publicly positioned its CV-series chips toward robotics, autonomous vehicles, and edge AI applications since 2022.

[8] Tesla, Inc., Form 8-K, Q1 2026 Update Letter, filed April 22, 2026. Confirms first-generation Optimus production line replacing Model S/X line at Fremont, with second-generation 10-million-unit facility under construction at Gigafactory Texas.