Listen to the audio version of this article (generated by AI).
Editor’s note: “The Robot Boom’s Next Milestone Is Surprisingly Boring” was previously published in September 2026 with the title, “AI’s Next Big Opportunity Is Right in Front of You.” It has since been updated to include the most relevant information available.
Picture yourself being assigned to a new floor of a warehouse during a busy shift.
You might see a robot coming toward you with a heavy container while workers move pallets off trucks for the bots to pick up.
You suddenly step in the path of the robot, which doesn’t recognize you, and just like that, you’ve got a workers’ comp claim.
The next day, the “days since last accident” sign has to be reset to zero. Yet, if factory robots are to work as intended, they must be able to recognize people and respond appropriately.
That may sound straightforward, but making it happen reliably, around people who aren’t following a carefully rehearsed demonstration, is a major engineering challenge.
Now, think about the broader AI doomsday fears going around right now. Behind those warnings is a concern about increasingly powerful systems behaving in ways their builders cannot predict or control.
A warehouse accident and a runaway AI system are very different risks. But both bring us back to the importance of safeguards that work when something unexpected happens.
Listen, you don’t have to believe every doomsday prediction to take that problem seriously.
It’s also a business problem. Because a warehouse manager can love your technology and still have very good reasons to hold off on buying it.
That’s what caught my attention about Agility Robotics’ new Digit 5.
Digit 5 Is Designed to Work Safely Around People
The company says its humanoid can lift 50 pounds, reach 7.2 feet high, and operate for more than 20 hours a day, with rapid recharging between stretches of work. But it also has an independent safety controller overseeing its response when people get too close. Depending on the situation, the robot can avoid them, stop, or sit down.
That doesn’t resolve the broader debate over AI’s risks. It does illustrate how addressing a safety problem can help move the technology forward.
Now, think about that from an investment perspective.
Teaching a robot when to stop could help a company sell more robots. Making the technology more dependable could help customers deploy it faster.
That’s the investment question I want to focus on today: What turns a promising robot into a product customers keep ordering?
For a robotics company, the distance between an impressive demonstration and a repeat customer can be enormous. Closing that gap is where I think some of the most valuable businesses will be built.
And after what I heard at the All-In Summit, I think investors need to pay much closer attention.
Why AI Safety Does Not Necessarily End the AI Boom
Last month, I spent two days at the All-In Summit listening to some of the most powerful people in technology talk about artificial intelligence at a pretty extraordinary moment.
One morning alone, I heard from Microsoft (MSFT) CEO Satya Nadella and Nvidia (NVDA) CEO Jensen Huang. I was sitting there when President Donald Trump called Jensen onstage, and the conversation went on speakerphone.
So, yes, it was quite a time to be in that room.
Especially after what had happened just days earlier. Some of the biggest names building frontier AI had begun calling for a deliberate slowdown in the development of increasingly powerful models. AI stocks got hammered as investors tried to figure out what that could mean for the hundreds of billions of dollars pouring into chips, data centers, power plants, and everything else supporting the AI buildout.
But at All-In, I didn’t hear much talk about slowing down…
I paid particular attention to Nadella. Microsoft is one of the companies writing the biggest checks in the AI Boom. Critically, I didn’t hear Satya announce that Microsoft was cutting its AI spending or abandoning its infrastructure commitments.
Running AI for customers takes compute. Testing AI takes compute. Training robots takes compute.
So I don’t look at this debate and conclude that the infrastructure spending cycle is over. I think there is still a pathway to years of growth.
In fact, as a long-term investor, I came away from All-In more bullish about how long this AI Boom could last.
And I think companies have an opportunity here to trade short-term speed for long-term durability.
A Slower Frontier Could Produce a More Durable Boom
My concern has never been this quarter’s earnings or next quarter’s earnings. I care about what happens two or three years from now. What happens if companies build too much capacity too quickly? What happens if a serious AI safety problem scares the public? What happens if regulators come down with a hammer?
Slowing down at the frontier could reduce some of those risks. We may give up some speed in the short term, but I think the industry has an opportunity to make this boom more durable over the long run.
Think about a workout.
If you push too hard, you’ll pull a muscle and you’re done. Manage the pace, and you give yourself a better chance of finishing.
Robotics gives us a practical example of what making AI more dependable can accomplish.
And for investors, that could shift some of the biggest opportunities toward companies finding valuable new ways to put AI to work.
One robotics company gives us a good look at what it takes to turn an impressive machine into something customers will actually fork over their hard-earned cash for.
What Separates a Great Robot Demo From a Great Business
An AI-powered robot working on a warehouse floor doesn’t have to contemplate the fate of humanity, but it does have to recognize a person who accidentally steps into its path, so it can stop before it runs him over.
Imagine you’re the manager deciding whether to expand a robot trial across your operation.
A machine might handle a container perfectly when the aisle is empty. But your warehouse has people moving around, awkwardly placed pallets, and shifts that need to stay on schedule.
Learning a task is only the beginning. You need to know how much useful work it completes in a shift, how often an employee has to intervene, and what it costs to keep running.
Agility says the previous generation of Digit logged more than 65,000 hours with customers. That’s experience with actual operating conditions, actual customer requirements, and actual problems to fix.
Digit 5 still has to deliver, though. Early access is expected in the first half of 2027. And the reported $300 million in orders comes from one unnamed customer and depends on hitting milestones. Those orders aren’t guaranteed revenue, so we still need to see execution. I want to see conditional demand turn into deliveries, productive use, and customers coming back for more.
That progression would tell us much more about the business than a video of a robot completing one difficult task.
Vision-Language-Action Models Help Robots Understand the Job
Across the industry, vision-language-action models, or VLAs, are helping developers address these challenges. Put simply, these systems connect what a robot sees with an instruction and the actions needed to carry it out. Nvidia is developing tools that help robots make those connections.
But understanding an instruction is only part of the job. The machine also has to carry it out reliably under changing conditions. Then you have simulation-to-real training, or Sim2Real.
Instead of doing every practice run with a physical robot, developers can train in virtual environments. Nvidia’s Isaac Lab supports running simulated environments in parallel, giving developers a way to generate training experience at scale.
There are still gaps between simulation and reality, and a successful virtual run doesn’t prove the physical machine will perform reliably. This is why real-world testing remains essential.
For investors, I think the important question is whether that training produces a machine customers can deploy with less setup and less supervision.
If every new installation requires an engineering team to spend weeks adapting the product, expansion could become expensive. A company that can reduce that burden may have a better chance of growing profitably.
What Investors Should Look for Before a Robotics Company Scales
This is how I’m thinking about young robotics companies: Can they turn one successful installation into many without letting service costs swallow the gains?
A customer expanding from one location to several would be an encouraging sign. So would a machine completing more work with fewer interruptions.
I also want to know whether the company can support those additional customers without hiring people faster than it grows revenue. Selling more robots and building a profitable robotics business are separate accomplishments.
And no one on Earth is spending more to answer that question than Elon Musk.
Optimus gets the demo videos. But watch what’s being assembled around it: the AI models, the training compute, the chips, the factories to mass-produce all of it. Musk is building the entire Physical AI stack at once – and the sheer scale of that project is why it sits at the center of what I call XPANSE.
Musk himself has suggested this effort could produce 1,000-fold gains for early investors. Perhaps that’s true; but no honest analyst can promise that. What I can offer is the map.
In my new briefing, I walk through the three steps I’d take today to get positioned for this shift, and I give away the name and ticker of the one stock I believe sits closest to the center of the buildout. The briefing is free. So is the pick.
You’ve just seen the questions I ask before trusting any robotics company with capital: real deployments, repeat customers, scaling without runaway costs. Apply that same lens to the XPANSE supply chain, and one name stands out…
