How can we please our future robotic overlords if we cannot communicate with them? Audio capabilities for artificial intelligence allow us to speak with AI using normal conversation while AI speaks back to us in our preferred language and accent. That’s all progressing just fine, but giving robots the ability to see means they’ll be able to navigate the world around us with ease. Autonomous cars do this today while the humanoids of tomorrow will also need “eyesight.” It’s all about sensors that allow any robotic device to see the world in the same way humans do. Today’s research piece will focus on the most visible, intuitive, and accessible robotic eyesight theme for retail investors – computer vision.
Computer Vision vs Machine Vision
Computer vision allow computers to do useful things like identify and label a real-world object, or track the movement of a product in a supply chain. The use cases range from the medical industry, where computer vision is used in imaging and diagnostics, to manufacturing where it’s used for quality control. While humans can simply use their eyes and brains to accomplish these simple tasks, computers need things like cameras, sensors, and AI algorithms.
Note that you’ll frequently comes across reference to “machine vision” which is a subset of computer vision. In other words, all machine vision is computer vision, but not all computer vision is machine vision. We’ve been following both for over a decade while tracking a variety of pure play stocks for retail investors. To keep things simple, we’ll just refer to everything as “computer vision” going forward. Today’s first computer vision stock is also the smallest of the lot.
Basler Shows Signs of Life
With a market cap just under $500 million, Germany’s Basler $BSL.DE offers a full stack of computer vision products – cameras that can handle extreme conditions, specialty computer chips for crunching all the data extracted from image processing, and 3D sensors for advanced imaging.
We’ve historically avoided the stock for two reasons. They fall under our market cap cutoff of $1 billion and they cannot grow their revenues. Perhaps when growth returns, the market cap issue will sort of work itself out. And that may be happening. After two years of declining growth due to tariffs and weakening demand in China, Basler is on track to post 21% revenue growth for 2025 at midpoint of guidance along with mention of 275 million euros in revenues for 2028. Here’s what that growth would look like:
The 2028 target now needs to increase given it represents just 7% growth over the next three years. That’s because growth is accelerating. Originally, they expected 4.5% growth in 2025, then midway through they raised guidance to 14% growth, then last quarter to 21% growth. That sort of acceleration is promising but also reflects a management team with a loose grasp on revenue forecasting.
The company attributes this growth acceleration to strong demand from the semiconductor industry where their products are used in the inspection process. This matches the success that semiconductor test equipment company Teradyne $TER appears to be having lately, seeing sequential revenue growth of nearly 40% of which 60% is said to be driven by AI demand. (That’s a company that has its own industrial robotics aspirations which still haven’t come to fruition.) Double-digit growth is the minimum we expect from disruptive companies, and while Basler appears set to achieve that, we need to see them deliver on these aspirations and then show strong guidance for next year. Gross margins in the high 40s are decent for a hardware-centric company but we’d like to see those improved with software.
All hardware companies should supplement their point-in-time sales with some sort of recurring revenue stream. Basler appears to be working on that, ramping up their AI software offerings to take advantage of the current hype demand. The latest advancement is their “pylon AI” platform which can be used to detect anomalies in electronics or classify certain types of cells and molecules in a living sample. It can also tell the difference between a pumpkin seed and a sunflower seed in your trail mix, but that seems a bit harder to monetize.
We’ll keep monitoring Basler for continued traction as they pivot into AI. Moving overseas to the U.S., Cognex $CGNX is another computer vision pure-play that’s ten times the size of Basler.
Cognex Shows Mediocre Growth
We first came across Cognex over a decade ago and remarked about what a fun place it must be to work. While the company has doubled in size to nearly $7 billion, the growth has been inconsistent and fairly dismal. Not being able to clear our 10% growth hurdle may be a function of the disparate applications their technology targets.
Fortunately, Cognex has the decency to break out their revenue into segments which we can monitor to see what’s working and what’s not. The trends imply that their recovery is being hampered by the automotive segment which is still in a downturn as the EV boom subsides and high interest rates persist.
What doesn’t make sense is that the semiconductor segment is only seeing “modest growth.” We’re in ‘peak AI’ right now, with semiconductor companies of all kinds reporting record revenues and earnings every quarter. Why isn’t Cognex able to capitalize on that?
To answer that, we dug into the company’s earnings calls and found that much of their semiconductor market growth came in 2024, so 2025 is simply building off a larger base, which seems intuitive. Tariffs and trade restrictions also delayed orders for their chip-related products such as wafer readers for semiconductor fabrication plants, as noted in the diagram above.
We recently moved Cognex to an “avoid” in our Tech Stock Catalog due to this lack of consistent growth which may come from a business model that’s just too diversified. Having a wide range of end markets isn’t a bad thing; it means one underperforming segment won’t ruin the whole business. However, it makes it harder for a company to grow as they’re distracted with trying to sell into unrelated industries. That brings us to the leader in computer vision that mainly targets automotive applications – Mobileye $MBLY.
Mobileye On Double Secret Probation
This Israeli company was acquired by, then re-spun off from Intel $INTC. They develop what’s known as advanced driver assistance systems (ADAS) which they sell to automotive companies to enable autonomous features such as steering and acceleration control. While they are still working on commercializing a solution for full self-driving, they’ve successful enabled Level 2 autonomy, which is labeled as “Occasional Self-Driving.” The driver still needs to be alert and engaged, but the car will drive itself in certain conditions.
Mobileye’s “SuperVision” product is realistically more like a Level 3 solution, but Mobileye management calls it “Level 2+” for legal reasons. It enables limited hands-off driving and is found in nearly 300,000 vehicles on the road today. SuperVision isn’t a single product but rather a full-stack solution made up of eleven cameras located all over the vehicle. It’s all powered by Mobileye’s proprietary EyeQ chip which acts as the “brain” of the operation, performing the real-time computing needed for the car to react to its surroundings.
Mobileye has been hard at work closing deals with major automakers. They recently expanded their partnership with Volkswagen, then closed a deal with an unnamed automaker to supply 9 million EyeQ chips. The problem is that these deals haven’t actually translated to revenue growth yet.
The company touts a rapidly expanding backlog, with their “8-year future expected automotive revenue pipeline” recently hitting $24.5 billion. Great, but why aren’t these conversions happening in 2026? With guidance of 3% revenue growth at the midpoint, investors were understandably concerned, sending MBLY stock to new all-time lows. The stock is now down 67% since spinning off from Intel in 2022, versus a +116% Nasdaq return over the same time period.
Value or value trap? Everything depends on whether Mobileye can return to growth. As they continue spinning wheels in the automotive industry waiting for their pipeline to turn into revenues, the company hopes to expand into other areas of what they’re calling “physical AI.” They cemented these hopes with a recently announced acquisition of humanoid robot startup, Mentee Robotics, but it isn’t passing the sniff test.
For starters, Mentee Robotics was co-founded by Mobileye’s CEO who still sits on the board of Mentee. Was this acquisition simply a self-serving move by Mobileye’s CEO to provide liquidity to his side project? Additionally, this startup is still pre-revenue. Mobileye’s offer of $900 million presents a steep price point for a company that still has not proven product-market fit. If Mobileye had really wanted to expand into physical AI, we think there are better ways of doing so. Were there no robotics software companies they could acquire? Perhaps a specialized chipmaker they can tack on to their own semiconductor solutions? We’re placing Mobileye on double secret probation while we wait to see what 2027 revenue guidance looks like. We need to see a return to double-digit growth if we want to continue calling this company “disruptive.”
Ambarella Opens Up
The last time we checked in with Ambarella $AMBA in Summer of 2024 they were a mess. Fiscal year 2024 revenues declined by nearly 33% to $226.5 million. IoT revenue, which represented about two-thirds of total revenue, declined by about 40% for that year. The automotive segment, while faring slightly better, still saw a 14% decline. And one customer accounted for more than 50% of revenues, while another represented 14% of total revenues. So, what’s happened since?
The company saw impressive revenue growth of 26% in 2025 and last quarter’s revenues of $108 million are a new record for the company.
It turns out all you need to do to achieve revenue growth is rebrand yourself as an “edge AI semiconductor company.” Computer vision clearly wasn’t hot enough, so they needed both buzzwords du jour: ‘AI’ and ‘semiconductor.’
In their most recent quarterly earnings report, management noted that edge AI revenue had set its sixth consecutive revenue record. This segment also comprises the majority of their revenue these days, over 70%.
“Edge AI” simply means artificial intelligence workloads are being performed on the device itself, at the “edge” of the network, rather than in a remote data center somewhere. Some practical use cases for Ambarella’s chips include facial detection for security and driver safety applications. The chips can also help autonomous drones navigate the skies by processing visual data locally. This eliminates the need for cloud connectivity, making their technology ideal for robotic applications in the future.
Regarding customer concentration risk, it might more nuanced than it initially seems. In their recent 10-K filing, Ambarella noted that WT Microelectronics accounted for 63% of total revenue. WT is their “non-exclusive sales representative” in Asia excluding Japan. WT isn’t necessarily a customer, but rather they’re a distributor, helping Ambarella reach their end customers: product manufacturers. If we look at Ambarella’s true end customers, the top ten make up 59% of total revenue. That’s a bit more palatable, though an exact breakdown would be nice to see.
Ambarella additionally cites that customers may cancel, change or delay product purchase commitments with little or no notice and often without penalty, potentially making their revenue more volatile. That likely explains the inconsistency they’ve seen in the past. Is Ambarella turning the ship around, or is this a cyclical company experiencing an up cycle? It feels a bit more like the latter, and we’ll be keeping Ambarella as an “avoid” despite the glimmer of hope in revenue growth and check back in a year.
Other Computer Vision Stocks
In compiling this list of computer vision stocks we’ve excluded names without meaningful revenues ($10 million per annum). That eliminates companies like MicroVision $MVIS which is said to dabble in computer vision, but we found numerous problems with their business last time we looked. We concluded, “if you’re holding shares of MicroVision right now, sell them and move on with your life,” and since then shares of MVIS have plummeted 96% as the company continues to not sell anything in meaningful amounts.
We look for leaders in any given domain, and computer vision is mature which means the leaders should have already emerged. Some don’t provide pure play exposure and trade on international stock exchanges, like Japan’s Omron $6645.T $OMRNY which is considered a leader in automation-integrated vision systems for quality inspection and robotics. Since they don’t break out what revenues come from computer vision, investors don’t know the extent to which they’re getting exposure. And when total revenues aren’t growing at all, does it even matter?
Another computer vision leader out of Japan – Keyence $6861.T $KYCCF – popped up on a list of 14 Promising Robotics Companies we compiled several years back. They’re a global leader in factory automation and industrial inspection equipment which raises a question about form factors. If we consider self-driving cars to be robots, should highly automated equipment on the factory floor be considered the same? Doesn’t really matter, because Keyence also doesn’t break out revenues directly attributable to computer vision while overall revenue growth is lackluster as well.
Searches for pure play computer vision stocks often turn up another company with no pure play appeal – Teledyne $TDY – whose Digital Imaging segment makes up about half of total revenues and encompasses a broad range of imaging-related products. Again, a case of not being able to track the exposure we’re looking to get. Another company building computer vision at the edge for mobile products would be Qualcomm $QCOM, though again the exposure is not likely to be meaningful. We’d rather hold QCOM because they’re a dividend contender (they’ve increased their dividend for 23 consecutive years.)
Conclusion
Computer vision sounds appealing to investors because it’s how we as humans think a robot ought to see. In practice, there are numerous other ways for machines to “see” such as radar, LiDAR, and depth cameras. When companies like Basler and Cognex are too diversified across industries this can mute their growth. While Mobileye is a leader in automotive computer vision, their growth has stalled and we’d like to see at least some industry diversification to supplement their vehicle autonomy ambitions. If they can’t turn the growth ship around by next guidance, it may be time to classify them as an “avoid.”