CogniFiber

CogniFiber is an Israeli deep-tech startup pioneering photonic computing for artificial intelligence inference workloads. Founded in 2018 by Dr. Eyal Cohen (CEO, hardware/algorithms engineer, Ph.D. in brain research from the Weizmann Institute) and Prof. Zeev Zalevsky (CTO, optics researcher and Dean of the Faculty of Engineering at Bar-Ilan University), the company has developed a proprietary architecture called DeepLight, which uses programmable multi-core optical fibers to perform neural network computations directly using photons instead of electrons. This fundamental shift from electronic to photonic processing addresses critical performance bottlenecks in current-generation AI accelerators: latency, power consumption, and thermal management.

The core technology leverages wave-division multiplexing and photonic switching within fiber-optic waveguides, enabling massively parallel inference operations without the cross-chip communication delays inherent in silicon. CogniFiber's publicly demonstrated systems have achieved performance levels approximately 1,000 times faster than Nvidia's latest GPUs on comparable inference benchmarks (measured in inference-per-watt and latency metrics), while consuming less than 1% of the electrical power. For example, demonstrations show the Aurora prototype system performing over 100 million AI classification tasks per second using only 180 watts of power—translating to roughly 500 million operations per watt, versus Nvidia's GPU performance in the range of 10-50 million operations per watt. This energy efficiency translates directly to operating-cost advantages and enables deployment scenarios where electrical infrastructure, thermal dissipation, or portability constraints currently limit AI adoption.

CogniFiber's commercial roadmap targets completion of a full-scale photonic processor system by 2027, with performance targets of 50 exaflops using less than 5 kilowatts of power. This would represent a multi-order-of-magnitude compression of supercomputing capability, shrinking deployments from football-field-sized data centers down to a few server racks. Near-term applications include data center inference acceleration, high-frequency trading signal processing, real-time edge analytics for autonomous systems (aerospace, defense, maritime), and embedded AI for IoT and robotics requiring ultra-low latency and power constraints.

CogniFiber is headquartered in Rosh HaAyin, Israel, and operates with a team spanning deep expertise in photonics, quantum computing, electrical engineering, and optical mathematics—many with backgrounds in Israeli defense and advanced technology institutes. The company has raised approximately $14 million across multiple funding rounds, most recently securing $5 million in Series A-II funding (July 2024) from lead investors Chartered Group and Eastern Epic Capital, with support from the Israel Innovation Authority. This capital is directed toward advancing prototype fabrication, establishing partnerships with cloud and telecommunications infrastructure providers, and scaling toward commercial deployment.

Dual-use relevance is substantial and credible. Photonic computing is foundational for next-generation defense capabilities in signal processing, sensor fusion, and real-time decision-making in contested environments. Military and aerospace applications—including autonomous systems onboard aircraft, ships, and unmanned platforms; communications signal processing; electromagnetic spectrum analysis; and situational awareness in low-bandwidth, high-latency scenarios—are natural markets for ultra-efficient, ultra-fast inference. Governments globally are investing heavily in photonic computing infrastructure as a strategic priority to maintain technological parity with peer competitors. Israel's existing ecosystem in photonics, quantum computing, and defense technology creates a natural advantage, and CogniFiber is well-positioned to serve allied government and defense-industrial requirements alongside commercial markets.

Risk factors include: prototype-to-production scaling (photonic systems require precision manufacturing and environmental stability); supply-chain dependencies for specialized fiber and photonic components; competition from alternative AI acceleration paradigms (other neuromorphic, optical, or quantum approaches); regulatory and export control uncertainty around advanced semiconductor and computing technology; customer qualification and adoption cycles (defense procurement can be lengthy, and civilian cloud infrastructure providers may be slow to adopt radically new technologies); and execution risk on achieving claimed performance targets in commercial deployment versus laboratory demonstrations. However, multiple independent sources (TechRadar, CBInsights, CTech, and academic reviews) have validated the technical foundation and preliminary results, suggesting material differentiation and defensibility of the approach.