Lightelligence raised $323 million in a Hong Kong IPO that attracted 600 times oversubscription, pricing at the top of its range and becoming the world's first publicly traded optical computing AI company.
Lightelligence, trading under 1879.HK, debuted on the Hong Kong Stock Exchange on April 28, 2026, after raising HK$2.5 billion, approximately $323 million, in an IPO backed by Tencent, Baidu, and China Mobile. Bloomberg confirmed the deal priced at the top of its marketed range, with public subscription reaching 600 times the offering size. For a company reporting RMB 106 million in 2025 revenue against RMB 1.34 billion in net losses, that demand signal reflects something beyond conventional financial metrics. Investors are betting on a paradigm shift in how AI hardware processes data, and Lightelligence is the most commercially advanced company in that bet.
Frost & Sullivan credits Lightelligence as the first company globally to achieve large-scale deployment of hybrid optical-electronic computing, a distinction that matters enormously when the field is otherwise populated by research labs and pre-revenue startups. The company has 44 paying customers, more than 5,000 card clusters deployed, and support for several thousand-GPU clusters in production environments. Those numbers convert what might otherwise be pure science into defensible commercial traction, exactly what institutional investors need to write nine-figure checks.
The company runs two product lines that address different parts of the AI data center bottleneck. Optical interconnect solutions replace copper cables linking GPUs and servers with light-based connections that deliver higher bandwidth, lower latency, and meaningfully lower energy draw. LightSphere X, launched in 2025, reportedly boosts GPU supernode interconnect efficiency by over 50% while cutting total cost of ownership on intensive compute tasks. That product alone addresses one of the most pressing constraints in scaling AI clusters: moving data between memory and processors fast enough to keep GPUs fed.
The second product line goes further. Optical computing processes data using photons rather than electrons, with Lightelligence's OptiHummingbird accelerator card featuring an on-chip optical network as the first commercial product of its kind. Its Pace 2 accelerator uses 3D through-silicon via packaging for further density gains. SCMP reported that the company has accumulated 410 patents covering both product lines, and KR Asia noted that more than half apply across both categories, creating a shared technical foundation that competitors would need years to replicate.
The Energy Crisis Driving Demand
AI data centers consumed an estimated 4% of global electricity in 2025, a figure analysts expect to double by 2028 as inference workloads scale. Optical links and optical compute both attack that number from different angles. Copper interconnects consume significant power and generate heat that drives up cooling costs. Optical fiber reduces that consumption substantially, and photonic processors eliminate the energy penalty of converting signals between optical and electronic domains repeatedly during computation. For hyperscalers facing power procurement constraints and governments imposing energy efficiency standards on new data center permits, optical hardware shifts from a performance upgrade to a regulatory compliance tool.
Lightelligence's timing reads as deliberate. Caixin Global reported the filing in March as AI workloads push data centers toward optical links. The company waited until commercial scale was provable, not just theoretical, before pursuing public markets. That discipline likely explains the valuation discipline from underwriters too: a pre-IPO valuation of RMB 7.8 billion on RMB 106 million revenue is ambitious, but the 600x oversubscription suggests the market viewed it as below fair value given the infrastructure buildout runway ahead.
What This IPO Means for AI Infrastructure Capital
The Lightelligence debut provides the first public market benchmark for optical computing as an asset class. Before this listing, institutional investors building AI infrastructure theses had no tradeable pure-play. Now they do, and the oversubscription tells you how much pent-up demand existed. Watch the post-listing trading range closely. If 1879.HK holds a premium to the IPO price in the first 90 days of trading, it validates the category and accelerates similar listings from peers including Ayar Labs, which operates in the same optical interconnect space from the US side. If it trades down under selling pressure from IPO allocations, it signals that the 600x was retail froth rather than structural institutional conviction.
For AI infrastructure startups and their investors, the takeaway is straightforward. Energy efficiency is now a commercial feature, not an afterthought. Optical computing addresses both inference performance and power consumption simultaneously, and the market just confirmed it will pay a significant premium for demonstrated traction in that intersection.
Also read: Taylor Swift trademarks her voice and face to draw a legal line against AI deepfakes • Canonical turns Ubuntu into the AI OS with native model management and edge inference • Monroe Capital's Koenig says AI supercharges only narrow software niches
