Photo via Pexels
Photonic tensor cores use light-based operations, specifically matrix multiplications, to accelerate AI computations directly on-chip. These devices leverage the inherent parallelism of optics to perform computations at the speed of light, typically using Mach-Zehnder interferometers or other integrated optical components. Companies like Lightmatter and Luminous Computing, alongside university research groups at MIT and Stanford, are at the forefront of this development. The technology is currently in advanced research and prototype stages, with early demonstrations showcasing impressive computational speeds. In 2023, Lightmatter announced its 'Enlighten' photonic AI accelerator chip, which demonstrated over 100 TFLOPS (teraflops) of optical computation while consuming significantly less power than comparable electronic GPUs. This offers a potential leap in performance and energy efficiency over traditional electronic GPUs, which are bottlenecked by electron movement and heat dissipation for large matrix operations.
Editorial check
How this page is checked
Source trail
lightmatter.ai
External links are separated from Surfaced commentary.
Reader safety
Context before clicks
Product links and external services are not presented as guarantees.
Monetization
No affiliate flag
Ads and commerce links are kept distinct from editorial text.
Surfaced take
Why It Matters
The escalating computational demands of large AI models like LLMs are driving up energy consumption and costs in data centers, making AI training increasingly unsustainable and expensive, potentially costing billions in electricity annually. When photonic tensor cores become mainstream, AI models will train faster and with drastically less energy, enabling more complex AI to be deployed ubiquitously and affordably. Hyperscale cloud providers, AI startups, and research institutions will win big, while traditional GPU manufacturers might need to pivot or acquire photonic capabilities. Technical challenges include manufacturing precision, converting electrical signals to optical and back efficiently, and developing robust software toolchains for optical hardware. We could see early commercial products within 5-8 years, with widespread adoption taking longer. The US (Lightmatter, Luminous) and China are heavily investing in this next generation of AI hardware. A second-order consequence is the potential for truly on-device, real-time AI processing in mobile and edge devices, enabling highly intelligent personal assistants and real-time inference without reliance on cloud connectivity.
Development Stage
Related
Anyword
Anyword is an AI writing assistant developed by Anyword Ltd. that generates and optimizes marketing copy across various formats, with a unique focus on…
Scrintal
Scrintal is a visual knowledge canvas and note-taking tool developed by a startup, designed to help users think spatially and connect ideas on an infinite…
Enjoyed this? Get five picks like this every morning.
Free daily newsletter — zero spam, unsubscribe anytime.