From Academic Paper to Listed Firm: Silicon Photonics Pioneer Lightelligence Charts Rise of Optical Chips in AI Infrastructure

According to China Securities Journal, a research paper published on the cover of Nature Photonics in 2017 first validated the feasibility of deep learning computation powered by photons. The work was authored by a doctoral researcher from the Massachusetts Institute of Technology. Nine years on, on 28 April 2026, Lightelligence, the enterprise founded by the researcher, debuted on the Hong Kong Stock Exchange as the world’s first listed firm specialising in silicon photonic chips for artificial intelligence. The company’s growth journey mirrors a wider shift, as photonics moves steadily from peripheral systems into the innermost core of AI computing infrastructure.

The firm’s founder stated in an interview that optical chips will account for more than 30 per cent, and potentially as much as 50 per cent, of all chips deployed in computing hubs within three to five years.

The evolution towards greater optical connectivity unfolds amid rising bandwidth and tighter latency requirements generated by training and running large AI models. Interconnection architecture is evolving from a supporting component into a fundamental pillar of modern computing infrastructure. Three distinct tiers define data centre connectivity. Scale-across covers communications between separate computing clusters. Scale-out handles links between different supernodes. Scale-up governs connections within individual supernodes, the field where Lightelligence concentrates its development efforts. 

An analogy illustrates the hierarchy: scale-across corresponds to communications across an entire school campus, scale-out represents exchanges between year groups, and scale-up resembles interactions within a single classroom. While the classroom covers a limited physical space, data exchanges taking place here operate at the highest frequency and carry the densest volumes of information.

Optical transmission has long been adopted for scale-out and scale-across connections. Communications within scale-up environments, however, have historically relied on copper and electrical signalling. Before large generative AI models gained traction, eight graphics processing units inside one server could operate efficiently using electrical data transfer. Current estimates suggest more than 95 per cent of data flowing through scale-up interconnections travels via electrical channels, with less than 5 per cent carried optically. Reliability concerns have slowed wider adoption of optical technology in this segment, yet this proportion is set to shift rapidly. Optical transmission could capture between 30 per cent and 50 per cent of scale-up traffic within the next three to five years.

Physical limitations create strong incentives for broader uptake of photonics. Electrical signals weaken substantially as copper cable length increases, rendering copper impractical for transmission distances exceeding one metre under heavy data loads. Fibre optics support multiple data streams simultaneously through separate wavelengths along a single strand, whereas each copper wire can only carry one signal. Systems built around copper cabling therefore demand far greater physical space to deliver equivalent bandwidth, creating layout constraints inside server hardware.

Technological progress spanning conventional optical modules, LPO, NPO and CPO consistently shortens electrical pathways leading to primary computing chips. Standard optical modules sit roughly one metre away from main processors. LPO reduces this distance to half a metre or several tens of centimetres. NPO further compresses the gap to approximately ten centimetres, while CPO brings optical hardware within a few millimetres of processing chips. Further developments including 3D CPO will enable tighter integration between photonic and electronic components.

The deeper integration of photonics into core computing infrastructure triggers sweeping restructuring across the industrial supply chain. Large-scale roll-out of CPO architecture carries transformative implications for the whole ecosystem. Relationships between suppliers and customers remain fluid and subject to rapid change. Established optical module manufacturers are advancing in-house photonic chip research, while GPU designers identify substantial commercial opportunities within the optical interconnect sector.

Against this backdrop of industrial transformation, the independent silicon photonics developer has set clear strategic objectives. The business aims to deliver a full portfolio of photonic chip solutions for future computing hubs, covering optical switching chips, photonic computing chips and optical interconnect devices.

Market demand for photonic hardware continues to expand. The entire industrial chain faces constrained production capacity, sustaining widespread supply shortages. This supply-demand imbalance creates a valuable window for emerging participants. Industry reports note Lightelligence holds an 88.3 per cent market share by revenue within China’s independent scale-up optical interconnect solutions market. Filings published ahead of its Hong Kong listing record annual revenues of 38 million yuan in 2023, 60 million yuan in 2024 and 106 million yuan in 2025, representing a compound annual growth rate of 66.9 per cent. Optical interconnect solutions form the company’s primary revenue stream, bringing in 75.58 million yuan from scale-up products in 2025. The photonic computing division registers accelerated expansion, achieving turnover of 20.20 million yuan in 2025, a year-on-year rise of 579 per cent and accounting for nearly 20 per cent of total revenue. The firm remains in a heavy investment phase and has not yet turned profitable, with cumulative research and development expenditure exceeding 1.1 billion yuan across 2023, 2024 and 2025.

High talent density, strong learning agility and fast iterative development form the core competitive strengths supporting the technology developer amid intensifying industry competition. Before 2022, silicon photonic chips saw limited deployment within commercially available optical interconnect hardware. Lower communication bandwidth requirements meant alternative solutions remained sufficient. Young chip design firms possess greater flexibility to absorb emerging technical knowledge and implement rapid iterations amid fast-evolving industry conditions. Ongoing efforts target the construction of industrial ecosystems and durable competitive barriers.

Major industrial investors including Baidu, Tencent, Alibaba, China Mobile and ZTE Corporation have taken equity stakes in Lightelligence. Earlier this year, the firm partnered with Biren Technology and ZTE Corporation to launch China’s first commercially available 128-GPU supernode equipped with integrated optical interconnection and optical switching technology. Thousands of cards from this product line have been deployed, with native compatibility established for mainstream large AI models.

During WAIC 2026, the business unveiled its next-generation scale-up optoelectronic hybrid networking framework. The company also confirms its self-developed NPO near-packaged optical chips have completed full-link verification and are moving into large-scale commercial roll-out. A strategic co-operation framework for CPO technology has been agreed with Centec Networks.

Photonic technology continues penetrating every tier of AI computing infrastructure. The enterprise maintains sustained investment into research and commercial deployment for optical interconnect and photonic computing technologies.