Lyte

**Product and the problem it solves.** Lyte builds LyteVision, an end-to-end perception platform for "Physical AI" — the class of machines that must move and act in the real world rather than merely process data. Today a robot maker, autonomous-vehicle team, or humanoid developer must assemble perception from a fragmented supply chain: LiDAR from one vendor, cameras from another, radar or time-of-flight from a third, plus the calibration, time-synchronization, sensor-fusion middleware, and compute needed to reconcile them into a single coherent model of the environment. That integration is slow, brittle, power-hungry, and expensive, and it is widely regarded as the primary bottleneck holding back reliable autonomy. Lyte's thesis is that perception should be delivered as unified infrastructure — "4D vision, RGB imaging, and motion sensing into a single platform to deliver unified spatial and visual data through one connection" — so that developers receive fused spatial-plus-visual data out of the box instead of building the stack themselves. The company frames LyteVision as the "perception foundation for Physical AI," analogous to how a smartphone camera stack abstracts away the underlying optics and signal processing.

**Core technology and how it works.** LyteVision combines three modalities that are usually separate. "4D sensing" refers to depth/range measurement over time (the fourth dimension being motion/temporal change), giving the system a live volumetric picture of scene geometry and how it is changing; RGB imaging supplies high-resolution color and texture for semantic understanding; and motion awareness provides egomotion and dynamic-object cues. Fusing these at the hardware and firmware level — rather than bolting them together downstream — is the technical core, and it is precisely the discipline Lyte's founders mastered building structured-light and depth systems at PrimeSense and Apple. On top of the sensing layer sits what the company calls an "AI-driven operating layer that continuously advances alongside breakthroughs in vision, language, and action models," positioning LyteVision to feed and be upgraded by the fast-moving world of vision-language-action foundation models rather than being frozen at a fixed capability. The output is a single, calibrated, time-aligned spatial-and-visual stream delivered "through one connection," which reduces the integration burden, power draw, and latency that plague multi-vendor perception. The platform is designed to be platform-agnostic across form factors.

**Market, customers, and go-to-market.** Lyte targets the broad and fast-growing physical-automation market: autonomous mobile robots (AMRs) in warehouses and factories, robotic arms, quadrupeds, robotaxis, and humanoids. Its go-to-market is a components/infrastructure model — selling perception as an integrated module to the many companies building robots and autonomous vehicles, rather than building an end robot itself. That "arms-dealer to the robotics industry" positioning is strategically attractive because it can attach to every autonomy program regardless of who wins the end-market race, and it directly addresses the labor-shortage and automation-pressure tailwinds pushing manufacturers, logistics operators, and mobility firms toward robots. The addressable pull is real, but the segment is also crowded with LiDAR makers, camera and image-sensor incumbents, radar vendors, and full-stack sensor-fusion players, so Lyte must convert its integrated-stack advantage into design wins with named OEMs. As of its stealth emergence the company had not publicly disclosed shipping customers, production timelines, or revenue, which is the central commercial unknown.

**Traction, funding, and third-party validation.** Lyte emerged from stealth on January 5–8, 2026 with $107 million in aggregate funding and immediate third-party validation: at CES 2026 LyteVision won the Best of Innovation Award in Robotics and was named an Honoree in Vehicle Tech and Advanced Mobility, selected from a record field of roughly 3,600 submissions. The cap table is a strong signal of institutional conviction: founding investor Avigdor Willenz's group, Fidelity Management & Research Company, Atreides Management, Exor Ventures (the Agnelli family's investment arm), Israeli venture firm Key1 Capital, and Venture Tech Alliance. Willenz is one of the most successful semiconductor entrepreneurs in Israeli history — founder of Galileo Technology, Annapurna Labs (acquired by Amazon), and Habana Labs (acquired by Intel) — and his backing plus Key1 Capital's participation anchor Lyte firmly in the Israeli deep-tech financing ecosystem. The presence of Fidelity and Atreides (a crossover fund known for hardware/AI conviction bets) at the stealth stage is unusual and reflects the caliber of the team more than de-risked commercial traction.

**Founders and team background.** Lyte was founded in 2021 by Alexander Shpunt (CEO), Arman Hajati, and Yuval Gerson — the engineers behind two of the most consequential consumer depth-sensing products ever shipped. Shpunt co-founded PrimeSense, the Israeli company whose structured-light 3D sensing powered the original Microsoft Kinect and was acquired by Apple in 2013; he served as PrimeSense CTO and then as a Distinguished Engineer at Apple working on the depth/perception platform underpinning Face ID. Gerson is likewise a PrimeSense and Apple veteran with a MEMS and systems-architecture background. Hajati, educated at the University of Tehran, built his career in the US as a lead system architect and senior engineering manager at Apple focused on integrated sensing. This is a rare "second-time depth-sensing" founding team applying hard-won lessons from mass-market consumer optics to the robotics era — the "Israeli-Iranian founding team" framing in the Israeli press underscores both the diaspora deep-tech pedigree and the unusual cross-border composition. Team headcount was not disclosed.

**Competitive dynamics.** LyteVision competes along several fronts at once, which is both its opportunity and its risk. Against LiDAR-first suppliers (e.g., Ouster, Hesai, Innoviz) it argues that range alone is insufficient and that fused RGB+depth+motion is what autonomy actually needs; against camera/image-sensor and vision-AI approaches (Mobileye's camera-centric autonomy, Tesla's vision-only stack, and numerous computer-vision startups) it argues that pure vision lacks robust metric geometry; and against sensor-fusion middleware and full-stack autonomy platforms (NVIDIA's robotics/AV perception stack, Qualcomm, and in-house OEM teams) it argues for a purpose-built, integrated, lower-power alternative. Its differentiators are: (1) hardware-level multimodal fusion rather than downstream stitching; (2) a founding team with a proven record of productizing and mass-manufacturing depth sensors; (3) an AI operating layer meant to ride the vision-language-action model wave; and (4) a components business model that can attach across many end platforms. The counter-risk is that well-capitalized incumbents — especially NVIDIA and large image-sensor makers — can bundle "good enough" perception and out-scale a startup on cost and ecosystem.

**Defense, security, and resilience dual-use relevance.** Perception is the foundational bottleneck for battlefield and security autonomy, which makes LyteVision structurally dual-use even though Lyte markets to commercial and industrial robotics. The same fused 4D-plus-RGB-plus-motion stack that lets a warehouse robot navigate clutter is directly applicable to unmanned ground vehicles, tactical quadrupeds, counter-UAS interceptors, autonomous logistics convoys, perimeter-security robots, and ISR platforms that must perceive and track in degraded, GPS-denied, low-light, or contested environments — exactly the conditions where a robust multimodal, vision-plus-geometry stack outperforms single-modality sensing. Robotaxi-grade perception maps cleanly onto autonomous military mobility, and humanoid perception onto CBRN and EOD applications that keep humans out of danger. The founders' depth-sensing lineage is itself dual-use heritage (structured light and time-of-flight are staples of targeting and terrain sensing). The honest calibration: as of its stealth emergence Lyte had announced no defense customers, programs, or certifications, is US-headquartered, and positions itself as a civilian/industrial perception vendor; the dual-use case here is strong enabling adjacency (perception as the key input to autonomy) rather than a fielded defense capability, and any allied-defense relevance would depend on export posture, ITAR/EAR considerations, and deliberate program engagement that has not yet been demonstrated.

**Growth stage, trajectory, and key diligence risks.** Lyte is early-stage but unusually well-capitalized: founded 2021, five years of deep pre-revenue R&D, and a $107M war chest at stealth exit with a marquee cap table and CES recognition, but no publicly disclosed shipping customers, revenue, or production volumes. The trajectory it must now prove is the classic hardware-plus-software chasm — converting an award-winning platform and elite team into design wins, manufacturable modules at competitive cost, and repeatable OEM revenue. Key diligence risks: (1) commercial traction is unproven — awards and funding are not orders; (2) intense, well-funded competition from LiDAR, image-sensor, and full-stack (NVIDIA) incumbents that can bundle and undercut; (3) hardware execution and supply-chain/manufacturing risk inherent to integrated perception modules; (4) capital intensity — perception hardware demands sustained investment before scale economics arrive; (5) the "Israeli" thesis fit here is founder lineage plus Israeli-VC backing rather than domicile (HQ is Mountain View), so strategic-alignment framing should be calibrated accordingly; and (6) dual-use relevance is latent, requiring deliberate defense engagement and export-control navigation to become real. For an Israeli-deep-tech-and-allied-autonomy thesis, Lyte is a high-quality, high-optionality watch-list entry whose value hinges on execution against a crowded field.