Cognex Pays $500M to Become the Eyes of Physical AI
The factory machine-vision incumbent is buying RealSense, the Intel spinout whose depth cameras sit on mobile robots, quadrupeds and humanoids — a bet that robots, not production lines, are where vision grows next.
Cognex has agreed to buy RealSense for roughly $500 million in cash, announced September 22. The deal is funded entirely from Cognex's existing cash and investments and is expected to close in the fourth quarter.
The headline number undersells the check. Cognex is also putting up a three-year, performance-based $56.5 million cash retention program for RealSense staff and roughly $50 million in restricted stock units — which is why Israeli outlet Calcalist framed the all-in cost closer to $600 million.
For that, Cognex gets a company doing real volume. RealSense expects $80–90 million in revenue in 2026, which the companies say is more than 50% growth year over year, and reports more than 4,500 customers. Cognex sizes the robotic perception market at about $600 million today, rising to $1.6 billion by 2030 — roughly a 25% compound annual rate.
A fourteen-month round trip
RealSense's corporate history is compressed enough to be its own argument.
Intel started the depth-camera line in 2014 and spent years treating it as a side project that periodically looked like it would be wound down. In July 2025 it was spun out as an independent company with a $50 million Series A, with Intel Capital and MediaTek's innovation fund among the backers. Intel kept about a 20% stake and a board seat, according to Calcalist.
Fourteen months later it is being sold for ten times that Series A. Intel, as a remaining shareholder, participates in the upside of a business it chose not to run.
That arc says something about where robot perception value has moved. Inside a chipmaker, depth cameras were a component story — a way to sell silicon. As an independent company pitching itself as the "visual cortex of physical AI," it became a platform story, and a platform story is what a strategic buyer pays for.
The company is mostly engineers in one place: about 180 employees, roughly 135 of them in Haifa, which Cognex intends to run as a development center. A small facial-authentication unit — around 25 people, with systems deployed at Ben Gurion Airport — will be spun out separately before close, so Cognex takes only the robotics perception business.
Why a machine-vision incumbent needs depth
Cognex built its franchise on industrial identification and 2D machine vision: reading codes and inspecting parts as they pass a fixed camera on a production line. That is a mature, cyclical market tied to capital spending in electronics, logistics and automotive. It is profitable and well-defended, and it does not grow like robotics.
RealSense sells into a different geometry. Its stereo depth cameras are designed to ride on the robot rather than watch a conveyor — fixed-arm cells, autonomous mobile robots, quadrupeds and humanoids. The job there is not "is this label correct" but "where is everything around me, in three dimensions, right now." That is the perception layer every mobile or manipulating robot needs before any model can decide what to do.
Cognex CEO Matt Moschner called RealSense "one of the most attractive adjacent markets" available to the company. CFO Dennis Fehr said the deal should be strongly accretive to Cognex's growth profile — note: growth, not margin. The pitch to shareholders is that Cognex buys a faster-growing revenue line, then builds what it calls a full-stack visual intelligence offering spanning identification, 2D and 3D machine vision, depth perception and robotic navigation.
Why it matters
Three things make this more than a tuck-in.
The eyes are consolidating before the brains. Most capital in physical AI is flowing to robot foundation models and humanoid makers. But every one of those systems depends on a small number of perception suppliers, and RealSense was one of the few independent, widely adopted depth-camera vendors with a mature developer ecosystem and SDK. Putting it inside an established industrial company with Cognex's balance sheet and sales channels makes it a sturdier supplier — and a more strategic one. Robot builders that standardized on RealSense now depend on a company with its own automation ambitions.
Commodity hardware is not the asset. Stereo depth modules are not exotic; plenty of vendors make them. What $500 million buys is the installed base, the calibration and firmware know-how, and a software layer thousands of developers already code against. Switching a robot's perception stack means re-validating the robot. That stickiness is the moat Cognex is paying for, and it resembles its own moat in factory vision.
Industrial incumbents are choosing to buy in rather than wait. The reflex among automation incumbents has been to watch the humanoid wave with polite skepticism. Cognex is instead buying exposure to it at the component level — the lowest-risk way to participate. If humanoids scale, they need eyes. If they don't, autonomous mobile robots and arm cells still do. The bet pays off under most plausible futures for robotics, which is what makes it a disciplined one.
What to watch
The integration risk is cultural and commercial more than technical. RealSense's customers range from hobbyists and university labs to robot OEMs; Cognex sells to large manufacturers through an enterprise channel. Whether developer-friendly pricing, open SDK access and broad availability survive that shift will decide whether RealSense keeps its ecosystem or slowly becomes a Cognex-only component.
Watch also how rivals respond. Cognex's machine-vision competitors now face a peer with a credible depth-and-navigation line, and the robot makers that buy RealSense cameras face a supplier that could one day sell them complete perception systems.
The deal still needs to close. But the signal is already clear: in physical AI, the perception layer has stopped being a parts bin and started being real estate.