Metalenz Acquires PxE Imaging to Bring Color and Depth to Edge AI Sensors

Metalenz announced on November 11, 2026 that it had acquired PxE Imaging, an Israel-based startup whose Holocoder technology captures color and depth information with a single camera. The companies did not disclose the terms of the transaction, but the strategic logic is concrete: Metalenz extends its sensing business from near-infrared into visible imaging, while PxE gains access to a manufacturing platform that has already shipped more than 300 million metasurface optics into consumer devices.

The deal matters for the edge AI market because it targets a persistent bottleneck in on-device perception. Cameras that feed AI models typically require separate depth modules, multi-sensor rigs or active illumination systems to produce the 3D data that robotics, automotive systems and machine vision need. PxE built its stack specifically as an alternative to those approaches, aiming to reduce the cost, size and power consumption of 3D sensing. Metalenz now owns that alternative.

What Does the Holocoder Actually Do?

PxE co-founder and CEO Yoav Berlatzky invented the Holocoder, which is patterned directly on the image sensor. The optical layer exploits the wave nature of light to encode wavelength and depth information, enabling simultaneous color and depth capture in a single image.

According to PxE's technology page, the Holocoder is a transparent, static diffractive optical element (a flat layer that shapes light through engineered microstructures rather than curved glass). It replaces the Bayer filter — the standard color mosaic found in virtually every consumer camera sensor — in a compact camera module and creates a holographic image. A deterministic, real-time image signal processor and decoding algorithm stack then generates RGB, infrared and depth images from that hologram, delivering channels already fused for perception AI.

The efficiency claim is the number that stands out. Metalenz says conventional Bayer filters absorb 60% of incoming light, while the Holocoder sorts light without absorption and delivers 2.5 times more light to a standard image sensor. The same layer encodes spatial geometries aligned with the visible image, a design intended to remove the need for separate depth modules or computationally expensive sensor fusion.

For camera-module designers, that combination touches both image quality and system architecture. More light at the sensor means better performance in constrained optical packages; encoded depth means fewer components, fewer sensors and a smaller bill of materials.

Why Does This Matter for Edge AI?

The significance for edge AI lies in where the additional information gets decoded. Metalenz says decoding depth and wavelength data directly in hardware reduces downstream computation, allowing AI models to run more efficiently on devices instead of relying on heavy software reconstruction.

That distinction carries real weight. Reconstruction-based approaches — computationally deriving 3D structure from raw sensor data — tax the processors on battery-powered devices such as phones, drones and robots. Hardware-level decoding shifts that burden into the optics themselves.

Rob Devlin, Metalenz's co-founder and CEO, framed the strategy around owning the hardware layer that feeds AI systems. "Owning hardware that brings new information to sensors is the real moat for differentiated AI models," he said.

The comment signals how sensor makers position themselves in the AI value chain. As model architectures commoditize, Devlin's argument goes, proprietary input data becomes the differentiator — and the company that controls the optics controls the data.

Who Is Metalenz?

Metalenz was spun out of Harvard University in 2016. The company reports shipping more than 300 million metasurface optics in consumer devices — flat lenses fabricated with semiconductor techniques rather than traditional polishing and molding.

That manufacturing base is the acquisition's operational core. Metalenz's foundry-compatible optics are produced at semiconductor scale and integrated directly onto CMOS image sensors, the standard chip type in digital cameras. The combined platform layers PxE's visible-light color and depth capture on top of the material and surface data Metalenz already collects, targeting consumer electronics, automotive systems, drones, robotics and machine vision.

Metalenz's existing PolarEyes platform captures polarization information — the orientation of light waves — that lets vision systems distinguish materials by their physical properties. Polar ID, the first product application of that platform, brings payment-grade face authentication to consumer devices. That product line shows the company's trajectory from component supplier toward edge products, and the PxE acquisition extends the same logic into visible imaging.

What Did Each Side Gain?

The acquisition pairs an early-stage Israeli optical technology with an American manufacturing operation already built for high-volume consumer hardware. Berlatzky acknowledged that scale gap directly.

"Joining Metalenz gives us the manufacturing platform and commercial reach to bring this technology to everyday devices," Berlatzky said.

For PxE, the deal answers the question that faces most novel optics startups: turning a diffractive optical element patterned on a sensor into hundreds of millions of units requires exactly the semiconductor-scale foundry relationships Metalenz has built since 2016. For Metalenz, PxE fills a visible-imaging hole in a portfolio anchored in near-infrared and polarization sensing.

Can the Numbers Become Products?

The stated stakes are straightforward. PxE reports light-capture and depth gains — 2.5 times more light than a Bayer-filtered sensor, single-camera color plus depth, hardware-level decoding — and Metalenz reports more than 300 million optics shipped. The combined company now has to convert those figures into commercial products across the consumer electronics, automotive, drone, robotics and machine vision markets it names as targets.

The path is plausible on paper: one optical layer that replaces the Bayer filter, captures 2.5x the light, and hands AI models pre-fused RGB, infrared and depth channels, manufactured on a supply chain already proven at consumer scale. Whether the Holocoder survives contact with mass production at that scale, and at what cost per module, will determine whether Devlin's hardware-moat thesis holds.