What It Costs a Robot to Feel, See, and Think: Checking 5 Sensor, Vision, and Brain Stocks

The smarter the machine gets, the more the silicon is worth

The thing that stuck with me most while putting this together was a simple ladder.

By Allegro Microsystems' own framing, a robot vacuum carries about $5 of its chips. A factory robot carries about $55. A full humanoid, roughly $150. As machines go from vacuums to walking machines, the silicon value earned per unit climbs about 30-fold.

The key isn't unit volume, it's content per unit. Even if robot adoption forecasts come in light, rising content per machine cushions the miss.

This piece walks through five companies covering the layer where a robot feels, sees, and thinks. I covered the joints and motors in The 4-Tier Robotics Investment Framework; this is the tier above that.

1. Allegro Microsystems (ALGM): two chips in every joint

Allegro makes the sensor chips that tell every joint in a robot exactly where it is and how fast it's turning.

The structure is clean. Every joint needs one chip to sense its position and another to drive its motor, and a humanoid has dozens of joints. Allegro has already won a design that puts close to 90 of its chips inside a single robot's joints.

But this is a cyclical chip business. The last downturn was brutal enough to briefly push the company into a loss. What pulled it back out, ironically, wasn't robots at all. It was AI data centers, now a tenth of all sales after more than quadrupling, with revenue up 23% in a single year and free cash flow multiplying almost six times to a record $125 million.

The problem is that the market has already fallen in love with the story. The stock sits at the richest valuation in its history. And I'd flag one more thing: its first robot wins are with Chinese manufacturers rather than American ones. Excellent company, but I'm waiting for a discount.

There's a foreign name doing the same job over the counter. Belgium's Melexis makes the same magnetic position sensors and has rolled out a version built specifically for robot joints, precise enough to tell a motor exactly where it's turning.

2. Novanta (NOVT): selling the sense of touch

Touch might be the most overlooked sensor layer of all.

A robot that can't feel force is basically useless in the real world. It crushes the egg, loses its balance, or breaks your hand when it shakes it. So every humanoid needs force sensors in its wrists, ankles, and fingers. The big supply chain studies call this the most wide-open opportunity left, because nobody has built the standard part yet.

One of the best options right now is Novanta. It builds the leading six-axis force sensor, the exact part that gives a robot arm its sense of touch, and it also owns Celera Motion, which makes the controllers that turn that sensing into precise movement. Two of the hardest-to-source pieces in the whole robot, from one company.

The fundamentals back it up. Over the past decade revenue has grown more than 160%, from $374 million to nearly a billion dollars. It has now teamed up with Nvidia to drop its sensors straight into Nvidia's robot platforms.

One thing bothers me. Novanta does not break out its robot revenue. And its biggest near-term growth driver is actually a medical deal. You'd be buying a robot story with no robot number to check it against.

3. Vishay Precision Group (VPG): the purest bet and the most fragile one

If you want the purest force-sensing bet available, it's Vishay Precision Group. It's also the one I'm most cautious on.

VPG makes the foil strain gauge, a 100-year-old piece of precision engineering that is the actual sensing element inside most of the force and load sensors going into robot wrists. It may be the only American pure play on force sensing you can buy.

The fundamentals tell a more cautious story. Revenue has sat flat at around $300 million for years, and earnings just slid to a cyclical low, from $2.60 a share at the 2022 peak down to about $0.40. Yet the stock re-rated to an all-time high on humanoid hope alone, even though humanoid revenue is under 1% of sales.

That's the widest gap between story and results on this list, which is why I file it as a speculative swing rather than an investment.

4. Cognex (CGNX): the eyes are already earning money in logistics

Cognex is the world leader in machine vision, the technology that gives a robot its eyes. It's also one of the few names here whose earnings are already recovering without any help from robots.

What sets it apart is the AI inside. Its edge learning software trains a system to tell a good part from a flawed one off just a handful of images, with no data scientist required. Its newest cameras put that AI directly onto the camera itself, one line running on Qualcomm chips and another on Nvidia. That turns a camera from a simple sensor into an edge computer that thinks at the exact spot it sees.

The moat is real too. Once one of those systems is wired into a production line, ripping it out means re-validating the whole line, so it stays for years. That's the kind of moat four decades of vision software and a brand synonymous with the category buys you.

On the numbers, revenue has roughly doubled over the past decade to nearly a billion dollars. It moves in cycles, but it's climbing out of its 2023 low with revenue up more than 24% and earnings more than doubling from the year before. What matters is where the recovery comes from: a single end market, logistics and warehouse automation, now its largest sector and growing double digits for nine straight quarters. None of that has touched humanoids yet.

5. Nvidia (NVDA): only 6% of the cost, collected on both ends

The brain is about 6% of a robot's cost. But that 6% is what makes these machines mainstream, and Nvidia wins on both ends of it.

On the robot itself, it sells Jetson Thor, a small onboard computer that lets the machine see, think, and react in real time. Back in the data center, it sells the second brain, the one that trains these robots and coordinates whole fleets of them. And those data centers already run on Nvidia chips.

Everybody is chasing a cheaper brain, of course. Qualcomm is building a humanoid-specific chip called Dragonwing, and the big automotive names, NXP, Texas Instruments, and Infineon, are all now shipping silicon aimed at robots. But if you want the fastest learning, most efficient brain on the market, Nvidia still has the goods.

The five in one table

Company	Ticker	Role	Recent results	My read
Allegro	ALGM	Position sensing, motor drive	Revenue +23%, record $125M FCF	Watch list, all-time-high valuation
Novanta	NOVT	Six-axis force sensors, motion control	Decade revenue +160%, ~$1B	Watch list, no robot revenue disclosed
Vishay Precision	VPG	Strain gauges (the sensing element)	Revenue flat ~$300M, EPS $2.60 to $0.40	Speculative swing
Cognex	CGNX	Machine vision (the eyes)	Revenue +24%, earnings more than doubled	Watch list, logistics-driven recovery
Nvidia	NVDA	Onboard + data center brain	Still dominant in AI infrastructure	Core holding, long-standing

How I actually use this watch list

Looking at these five together, one pattern jumps out. The business quality is mostly excellent, but the prices have moved a little faster than the actual momentum justifies.

So in this tier I emphasize discipline over buying. Allegro, Novanta, and Cognex are terrific businesses on the watch list waiting for a pullback. Vishay is a speculative swing because the gap between story and results is too wide. Nvidia is the one I've been holding and talking about for a long time.

And here's the checkpoint I'd offer: in this tier, the real signal isn't humanoid enthusiasm, it's the moment a company starts breaking out actual robot revenue as its own line. Right now almost none of them do. Not disclosing means the number is still small. Starting to disclose means the story turned into results. Pairing this with Optimus V3 and Unsupervised Robotaxi also gives you the demand-side timeline.

FAQ

Q: What does the 30-fold silicon jump per robot actually mean? A: Using Allegro's ladder, a robot vacuum carries about $5 of chips, a factory robot about $55, and a humanoid about $150. Even if unit forecasts disappoint, rising content per machine defends revenue, so in this tier you watch machine sophistication more than unit counts.

Q: Why call touch sensing the most open opportunity? A: Because no part has become the industry standard yet. Position sensing and vision already have clear winners, but the force sensors going into wrists, ankles, and fingers have neither a settled spec nor a settled supplier. That's why Novanta and Vishay keep coming up as candidates.

Q: Will Nvidia keep winning the brain? A: Qualcomm's Dragonwing, NXP, TI, and Infineon are all coming in with cheaper brains. But on learning speed and efficiency Nvidia is still ahead, and right now it's the only one collecting revenue on both the robot itself and the data center that trains it.

This is educational analysis, not financial advice.