Lightelligence soars 400% after IPO on a bet on optical interconnect for AI

Lightelligence, a Shanghai-based photonic chip developer, gained almost 400% on its first day of trading after its IPO in Hong Kong. Investors effectively bet that the next bottleneck in AI infrastructure won't be the GPUs themselves, but the data links between them.

Why the stock took off

On paper, the story looks almost paradoxical: a company with just $15.5 million in annual revenue briefly reached a market valuation of about $10 billion. Lightelligence priced its shares at the top of the range, HK$183.2, and raised HK$2.4 billion. At the open, the price jumped to HK$880. The retail tranche was oversubscribed nearly 5,785 times. This wasn't a reaction to current financial results. It was a bet that the AI infrastructure market is willing to pay a premium for technology that could remove the next system-level constraint.

The company's status added to the effect. Lightelligence became the first mainland Chinese photonic chipmaker to list in Hong Kong. For investors, that makes it a rare public play at the intersection of AI, photonics, and data centers. In essence, the market is buying more than revenue. It's buying the company's potential place in the supply chain for ever-larger GPU clusters, where the speed of data exchange between accelerators is starting to matter for overall performance as much as the raw power of the chips themselves.

Why light

Modern models, including large language models and image generators, are trained and run on arrays of accelerators linked into a single cluster. The problem is that as these systems grow, copper electrical links start to hold back scaling: they run hotter, consume more power, and hit bandwidth limits. When thousands of GPUs need to exchange data constantly, even a small gain at the interconnect level adds up to a noticeable difference in the speed, cost, and efficiency of the whole cluster.

• Lower latency for chip-to-chip data transfer
• Higher bandwidth than copper links
• Lower power consumption and heat output
• Easier scaling of connectivity within a server and between servers in a cluster

This is where Lightelligence is pushing optical interconnect, which transmits signals with light instead of electricity. The company describes its flagship product, LightSphere X, as the first distributed optical switching solution for connecting GPU supernodes. According to Lightelligence itself, the architecture can raise model FLOPS utilization by more than 50% and lower the total cost of ownership of compute infrastructure. If those numbers hold up at scale, the technology's value to AI compute providers becomes very clear.

Strengths and risks

The company already has arguments that go beyond pitch decks. According to the article, Lightelligence had accumulated 410 patents by March 2026, and more than half of them apply to both optical interconnect and optical computing. By the end of 2025, it had 44 commercial customers and supported GPU clusters of several thousand cards. In China's scale-up optical interconnect segment, meaning links within a single high-performance node, the company ranked first among independent vendors by revenue, with an 88.3% share.

But the weak spots are visible too. Huawei dominates the overall market with a 98.4% share, so 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, which is very fast growth. However, the net loss widened to 1.34 billion yuan in 2025, the liabilities-to-assets ratio reached 473%, and a single customer accounts for 40.6% of total revenue.

In other words, the public market has so far valued mainly the promise of future demand. Neither founder Yichen Shen's strong scientific background nor a forecast that the global AI computing and interconnect market will grow 27% a year through 2031 resolves the central question: can the business catch up with its own market cap?

What it means

AI infrastructure is entering a phase where what matters is no longer just the accelerators themselves, but how quickly and cheaply they can exchange data. For the market, it's a signal that optical interconnect is becoming a major bet in its own right. For Lightelligence, it means investors now expect large-scale commercialization, not a compelling science story.