Lightelligence’s Hong Kong debut turned an obscure data-centre component into a $10 billion market story. The Shanghai photonics company rose almost 400% from its offer price, despite reporting only $15.5 million in 2025 revenue and a much larger annual loss.
Investors are betting that optical interconnect will relieve a fundamental AI bottleneck. As thousands of accelerators work together, moving data between them through conventional copper consumes power, creates heat and limits bandwidth.
Optical systems replace electrical signals with light, promising faster communication with lower latency and energy use. The technology could determine how efficiently future AI clusters convert expensive chips into useful computation.
Light removes friction between AI processors
Large models divide work across many GPUs. If processors wait for data, headline compute capacity goes unused. Copper connections become harder to scale over distance and density because electrical resistance generates heat and signal integrity deteriorates.
Lightelligence builds optical links within servers and across clusters, as well as optical-computing systems that process information using photons. Its flagship LightSphere X is described as a distributed optical circuit switch for GPU supernodes.
The company says LightSphere X can improve model FLOPS utilisation by more than 50%, reducing total computing cost. The claim needs validation across customer workloads, but it captures the economic value: better interconnect can make existing accelerators behave like a more coherent machine.
Commercial leadership comes with concentration risk
Frost & Sullivan describes Lightelligence as the first company to deploy optoelectronic hybrid computing commercially at scale. It held 410 patents by March 2026, with more than half spanning both interconnect and optical computing.
Among independent providers in China’s scale-up optical-interconnect market, it held 88.3% revenue share in 2025. Huawei nevertheless dominates the overall segment with 98.4%, leaving Lightelligence as the largest third-party supplier in a highly concentrated field.
The company had 44 commercial customers supporting clusters with several thousand GPUs by the end of 2025. Its IPO investors included Alibaba, GIC, Temasek, BlackRock, Fidelity, Schroders, Hillhouse, Lenovo and ZTE.
Customer concentration is a concern: one buyer represents 40.6% of revenue. Losing or delaying business from that account could materially affect a company still building scale.
The valuation runs far ahead of the financials
Revenue rose from RMB 38 million in 2023 to RMB 60 million in 2024 and RMB 106 million in 2025, a compound annual growth rate of 66.9%. Net loss widened to RMB 1.34 billion in 2025, and liabilities far exceed assets, with a reported asset-liability ratio of 473%.
The IPO raised about HK$2.4 billion, or $310 million, after the retail allocation was oversubscribed nearly 5,785 times. That capital can fund product development and manufacturing, but the market value assumes revenue will eventually approach the scale of the technical ambition.
Founder Yichen Shen’s credibility contributes to the premium. His 2017 Nature Photonics paper helped demonstrate the feasibility of using light for deep-learning computation.
The broader AI computing and interconnect market is forecast to grow 27% annually through 2031. Lightelligence is positioned in the right constraint at the right time, but technological relevance does not guarantee profitable execution.
Its debut matters because it signals where infrastructure attention is moving. The AI race has focused on accelerator performance; the next gains may come from how effectively those processors communicate.
Optical interconnect is not visually prominent and does not improve a model by itself. It reduces the waiting, heat and energy that prevent clusters from using their full capacity. Investors have placed an extraordinary price on that possibility. Lightelligence must now prove that commercial adoption can grow quickly enough to connect a $15.5 million business with the $10 billion expectation attached to it.
