Lightelligence soars 400% after Hong Kong IPO: betting on optical interconnect for AI

Lightelligence, a developer of photonic chips from Shanghai, rose nearly 400% on the first day of trading after its IPO in Hong Kong. Investors essentially bet that the next bottleneck in AI infrastructure would not be the GPUs themselves, but the data transmission channels between them.

Why the stock soared

On paper, the story looks almost paradoxical: a company with annual revenue of just $15.5 million briefly received a market valuation of around $10 billion. Lightelligence placed shares at the upper end of the range — HK$183.

2, raising HK$2.4 billion, and on the opening of trading the price jumped to HK$880. The retail portion of the placement was oversubscribed nearly 5,785 times.

This is not a reaction to current financial results, but a bet that the AI infrastructure market is ready to pay a premium for technology capable of removing the next systemic constraint. Additional effect came from the company's status itself. Lightelligence became the first mainland Chinese manufacturer of photonic chips to list on the Hong Kong Exchange.

For investors, this is a rare public story at the intersection of AI, photonics, and data centers. In essence, the market is buying not only revenue, but also the company's potential place in the supply chain for increasingly large GPU clusters, where the speed of data exchange between accelerators begins to affect overall performance no less than the power of the chips themselves.

Why light is needed

Modern models, including large language models and image generators, are trained and operate on arrays of accelerators connected into a single cluster. The problem is that as such systems grow, copper electrical connections begin to limit scaling: they heat up more, consume more energy, and hit bandwidth limitations. When thousands of GPUs must constantly exchange data, even a small gain at the interconnect level becomes a noticeable difference in the speed, cost, and efficiency of the entire cluster.

• Lower latency in data transmission between chips
• Higher bandwidth compared to copper connections
• Lower power consumption and heat dissipation
• Easier scaling of connections within a server and between servers in a cluster

This is where Lightelligence is advancing optical interconnect — transmitting signals with light instead of electricity. The company's flagship product LightSphere X is described as the first distributed optical switching solution for connecting GPU supercomputing nodes. According to Lightelligence itself, such an architecture can increase FLOPS utilization in models by more than 50% and reduce the total cost of ownership of computing infrastructure. If these metrics are confirmed at scale, the value of the technology for AI computing providers becomes very clear.

Strengths and risks

The company already has arguments beyond presentations. According to the article, by March 2026 Lightelligence had accumulated 410 patents, with more than half applicable to optical interconnect and optical computing segments. By the end of 2025, it had 44 commercial customers and support for GPU clusters of several thousand cards.

In the Chinese segment of scale-up optical interconnect, that is, connections within a single high-performance node, the company ranked first among independent suppliers by revenue with a 88.3% share. But weaknesses are also visible.

Huawei dominates the entire market with a 98.4% share, while Lightelligence remains the largest third-party supplier, not the industry leader overall. Revenue grew from 38 million yuan in 2023 to 60 million in 2024 and 106 million in 2025, representing very rapid growth.

However, the net loss in 2025 expanded to 1.34 billion yuan, the ratio of liabilities to assets reached 473%, and one customer brings in 40.6% of all revenue.

In other words, the public market has primarily valued the promise of future demand. Even the founder Yichen Shen's strong scientific background and the forecast of global AI computing and interconnect market growth at 27% annually until 2031 do not remove the main question: will the business be able to catch up with its own capitalization.

What this means

AI infrastructure is entering a phase where not only the accelerators themselves matter, but also how quickly and cheaply they can exchange data. For the market, this is a signal that optical connections are becoming a separate major bet. For Lightelligence — that from now on they are expected not a beautiful scientific story, but large-scale commercialization.