Physical AI Stocks: What High Conviction Actually Looks Like in the Data

The biggest checks in artificial intelligence are no longer being written for software. By 2026 the frontier has moved from AI that thinks to AI that acts, and the capital is following it into the physical world. Venture funding into robotics startups reached roughly $15 billion in 2025, eclipsing even the sector's 2021 peak. NVIDIA's Jensen Huang keeps calling humanoid robots a $40 trillion opportunity, and SoftBank's Masayoshi Son told CNBC he sees the next trillion-dollar company coming out of physical AI.

There's a catch for anyone trying to invest in that narrative: most of the conviction is locked inside private companies you can't buy. So this post asks a narrower, more answerable question. Among the US-listed companies you can actually own, which ones have put their own balance sheet, revenue mix, and R&D budget where the story is, and where can you see that commitment in the data?

This is a map of conviction, not a recommendation. "High conviction" here means the company's commitment to the bet, not a signal for you to act on. Every claim below is backed by a chart you can re-run yourself.

What "conviction" looks like in a filing

Hype is free; commitment shows up in the numbers. Three signals separate a genuine all-in bet from an incumbent bolting a robotics line onto a mature business:

• R&D intensity: what share of every revenue dollar goes back into building the future.
• Revenue concentration / pure-play focus: how much of the business is the bet, versus a rounding error inside a giant.
• Backlog and deployment: contracted future work and robots actually in the field, the leading indicators that the bet is landing.

Start with the simplest version of the question: is the physical-AI tailwind actually showing up in sales, or only in headlines?

The shapes tell the story. A compute platform, a surgical-robotics franchise, a warehouse-automation upstart, a cobot maker, and a machine-vision specialist all sit under one "physical AI" headline and grow nothing alike, which is exactly why a single "robotics trade" is the wrong way to think about it.

Who is actually all-in

The cleanest conviction metric is how much of each revenue dollar a company plows back into R&D. Plot it and the hierarchy of commitment becomes obvious:

One name towers over the profitable incumbents. Lidar maker Ouster pours an outsized share of revenue into R&D: nearly 40% in fiscal 2025, against low-double-digit ratios for the established franchises. That is the signature of a company still building, not yet harvesting. NVIDIA, by contrast, spent $18.5 billion on R&D in fiscal 2026 (up 43% year over year). That is a far smaller fraction of a vastly larger business, but the largest absolute physical-AI R&D budget on this list.

Concentration is the other tell. Symbotic built its entire warehouse- automation business around a single customer: about 85% of its FY2025 revenue came from Walmart, against a backlog of roughly $22.5 billion. That concentration is its biggest risk and its clearest signal of conviction at the same time; the two are the same coin.

The honest split: what already works vs. what's still a bet

Here's where a credible map has to draw a hard line. Some of this "robotics" exposure is already a real, profitable business. The rest is cash burned today on a promise about tomorrow. Start with the proven side, the names that already convert physical-AI revenue into real operating profit:

These names are unglamorous by physical-AI standards. Intuitive Surgical has spent two decades compounding the da Vinci franchise; procedures grew about 18% in 2025. Cognex makes the machine vision that gives factory robots their eyes, at a healthy operating margin on roughly $994M of revenue. These are robotics businesses that already print profit, proof the category can pay even if the humanoids haven't yet.

Now the other side of the line, the all-in bets still spending far more than they earn:

The gap is stark. Serve Robotics booked just $2.7M of revenue against a $101.4M net loss in 2025 while scaling its sidewalk-delivery fleet. It spent more than seventeen times its entire revenue on R&D alone, the purest expression of conviction-before-product on this whole list. Archer Aviation ended the year pre-revenue with about $2.0B of cash to fund its eVTOL bet. This is what conviction looks like before it works: deep losses funded by a balance sheet, in service of a market that doesn't exist yet. Neither chart is a verdict. They just keep the "proven" and the "promised" honestly separate.

The forward signal, and the cautionary tape

Backlog and deployments are the leading indicators that conviction is paying off. Symbotic's $22.5B backlog, Intuitive's growing installed base, Serve's expanding fleet, and the order-driven growth at autonomous-defense maker AeroVironment (+14.5% revenue in FY2025, almost entirely from autonomous systems) all show up first as a revenue trajectory:

But being early and all-in is not the same as surviving to scale. Lidar pioneer Luminar, once a flagship of the autonomy trade, filed for Chapter 11 in December 2025 and expects to be delisted. It's the live reminder that the data can show commitment; it can never promise the outcome.

How much of the future is already in the price?

A great technology story and a great investment are not the same thing; the difference is the price you pay. The pre-revenue conviction names tend to carry the richest multiples, because the market has already capitalized years of flawless execution into today's price. Setting each company's valuation against its growth, side by side, shows how much of the thesis is already spoken for:

And because the cohort spans a $200B+ compute platform down to single-digit- million-revenue robot fleets, lumping it into one "physical-AI trade" hides enormous dispersion in what investors have actually lived through:

How to think about it

The physical-AI thesis has real substance: the foundation-model layer shipped, private capital is flooding in, and deployments are scaling. But the honest, investable version of the trade looks like this:

• The proven, cash-generating names are the unglamorous ones: surgical systems, machine vision, warehouse automation, not the humanoids in the demo videos.
• The most exciting names aren't buyable. Figure, Unitree, Apptronik, and Agility are private or foreign; the headline humanoid stories are out of reach on US markets. This post deliberately works with what you can actually own and chart.
• A lot of the conviction is already priced in, and concentration cuts both ways. Symbotic's single-customer bet and Luminar's bankruptcy are the upside and the downside of going all-in.

None of that is a buy or sell signal. It's a framework for asking sharper questions. Every chart here started as a plain-English question put to mrmarket.ai, so don't take these numbers on faith. Bring your own questions and research the companies for yourself.

---

This article is for informational purposes only and is not investment advice, a recommendation, or an offer to buy or sell any security. Charts reflect data as of the date shown. Do your own research.