2025 · 06· Nature Communications · 16:6097· Co-first author· IF 14.7

Seamless optical cloud computing across edge-metro network for generative AI

S. Xing*, A. Sun*, C. Wang*, et al. (* equal contribution)
Fudan University · University of Cambridge · Zhangjiang Lab

DOI → PDF Scholar

TL;DR An optical cloud-computing architecture seamlessly deployed across an edge–metro network. Inputs and AI models are modulated onto light, letting edge nodes directly access an optical computing center over the edge–metro fibre. Experiments measure 118.6 mW/TOPs energy efficiency — two orders of magnitude lower than electronic cloud computing — and the same architecture runs multiple complex generative-AI models in parallel for image generation.

Edge-metro optical cloud computing for generative AI

Background

The rapid rise of generative AI demands ever-larger computational capacity, and cloud computing has become the backbone of this transformation. Yet today's electronic cloud relies on enormous data centers and incurs significant power consumption and security exposure. Reducing energy use while continuing to scale compute is a persistent challenge for cloud computing.

Architecture

We propose and experimentally demonstrate an optical cloud-computing architecture that can be seamlessly deployed across an edge–metro network. By modulating both inputs and AI models into light, a wide range of edge nodes can directly access the optical computing center via the edge–metro network.

Highlights

118.6 mW/TOPs energy efficiency, two orders of magnitude lower than electronic cloud-computing baselines
Seamless deployment across an edge–metro network, eliminating mid-link electronic regeneration
Parallel optical computing runs multiple complex generative-AI models for image-generation tasks

Citation

S. Xing*, A. Sun*, C. Wang* et al., "Seamless optical cloud computing across edge-metro network for generative AI," Nature Communications, 16:6097, 2025.