German company Q.ANT has demonstrated a hybrid approach to running large language models (LLMs) by offloading some layers to a photonic coprocessor. This new NPU Gen 2+ chip performs a discrete Fourier transform using light in a single pass, aiming to overcome the computational and memory limitations of traditional GPUs, especially with increasing context lengths. While Q.ANT claims this is a world first, the British startup Lumai previously launched a full LLM using optics in April 2026, achieving significant energy savings. AI
IMPACT This development could lead to more energy-efficient and powerful LLM inference, potentially reducing hardware costs and environmental impact.
RANK_REASON Research milestone involving novel hardware for LLM inference. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Mastodon — fosstodon.org →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →