Lyte raises $165M to build the sensor stack robots keep rebuilding

Why it matters

Lyte's valuation prices the Apple and PrimeSense veterans as infrastructure founders, betting that robot makers will buy an integrated perception stack instead of assembling their own.

Lyte CEO and co-founder Alexander Shpunt, Arman Hajati and Yuval Gerson, the former Apple engineers behind the company, have raised about $165 million at a $1.6 billion valuation, Bloomberg reported on September 2nd. Maverick Silicon led the financing, with Fidelity Management & Research, Atreides Management, Key1 Capital and Ora Global returning as investors.

The round brings Lyte's total funding since its 2021 founding to $272 million and more than triples its previous valuation, according to Bloomberg. Lyte had disclosed $107 million in aggregate backing when it emerged from stealth on January 5th, meaning the new round adds more capital in one shot than Lyte accumulated during its first four years.

That money is following a founding team with an unusually direct claim on the history of machine vision. Shpunt co-founded PrimeSense, served as its chief technology officer and pursued PhD studies in physics at MIT, according to Lyte's founder biographies. PrimeSense developed the 3D-sensing technology used in Microsoft's original Kinect before Apple acquired the business in 2013 for a reported price of roughly $350 million. Shpunt then became a director, distinguished engineer and head of Apple's Advanced Technologies Group.

Hajati, Lyte's chief technology officer, holds a PhD in electrical engineering and computer science from MIT and led the architecture of Apple's Taptic Engine across several generations of the iPhone and Apple Watch. Gerson, Lyte's vice president of engineering, previously worked in research, architecture and mechanical systems at PrimeSense, Apple and Rafael. Lyte says he has more than 25 years of advanced research and development experience.

A second attempt to make machine vision foundational

Shpunt's first sensing business helped machines interpret human movement. Lyte is taking that experience into a harder environment: robots operating around moving equipment, changing light, clutter and people who do not behave like controlled test subjects.

Lyte's pitch starts with a recurring engineering problem the founders encountered after leaving Apple. Robotics developers were buying sensors from several vendors, calibrating them separately and writing software to reconcile data captured at slightly different times. Shpunt wrote in Lyte's January launch post that teams were spending years rebuilding these perception stacks.

The founders responded by putting sensing hardware, custom silicon, edge compute and perception software into one system. Lyte's technology stack combines what it calls coherent 4D vision with RGB imaging and inertial measurement. Lyte says its silicon fuses those inputs into a spatially aligned timeline before passing the data to a robot's software stack. The intended result is less calibration work and faster decisions when objects are moving.

Lyte introduced the system as LyteVision in January and has positioned it for autonomous mobile robots, industrial arms, quadrupeds, robotaxis and humanoids. Its current website also describes a broader Lyte Galaxy platform built around Lyte Vision sensing modules, compute, software and algorithms. The product strategy is clear even as the branding expands: Lyte wants to own the layer between the physical world and the models controlling a machine.

That layer becomes more valuable as robotics labs pour money into foundation models and general-purpose autonomy. A model can plan a movement only from the evidence its sensors provide. Delayed, misaligned or incomplete inputs leave even capable software reasoning from a stale picture of the room.

The competition is integrating too

Lyte is entering a market where established sensor vendors have reached the same conclusion about integration. RealSense completed its spinout from Intel in July 2025 with a $50 million Series A and sells depth cameras, processors and developer tools for robotics. RealSense says it serves more than 3,000 customers, giving it a distribution base Lyte has yet to demonstrate publicly.

Ouster moved in the same direction when it acquired Stereolabs on February 4th, 2026. The combined offering joins lidar, stereo cameras, edge AI compute, sensor-fusion software and perception models. Orbbec also sells structured-light, stereo, time-of-flight and lidar products for robot vision. The field is converging around bundled perception rather than standalone cameras.

Lyte's differentiation rests on its coherent 4D sensing approach, custom silicon and the founders' experience shipping sensing systems at consumer-electronics scale. Investors are assigning substantial value to that history before Lyte has supplied comparable operating metrics. At its January unveiling, Lyte said it had secured design wins with leading physical AI developers, without naming the customers or specifying deployments, revenue or units shipped.

The $1.6 billion valuation therefore prices in execution that still lies ahead. Custom chips and integrated sensor modules demand repeated hardware iterations, manufacturing capacity and long qualification cycles with robotics customers. The new capital gives Shpunt, Hajati and Gerson room to complete those cycles without narrowing Lyte into a conventional camera supplier.

Lyte's stated thesis is to make robots understand the physical world safely, and the company describes LyteVision as a "visual brain" for robots in its founder materials. Lyte's founders have already helped teach game consoles and phones to see depth. The next test is whether robotics manufacturers will hand them control of the entire perception stack.