Xizhi Technology Research Report: Its popularity is second only to Alibaba and Kuaishou. How will the market open today?

At first, I just wanted to see if this IPO was worth participating in, but the more I looked into it, the more I realized it’s not just an IPO issue—
it has brought to light the new Hong Kong stock sector 18C, the industry landscape of domestic AI computing power, the valuation arbitrage logic between A-shares and Hong Kong stocks, and how retail investors can use informational advantages to play these larger themes in Hong Kong stocks.
1. What does this company do
Xizhi Technology does two things: optical interconnects and optical computing. Additionally, there is a concept of optical modules in the market, which can be easily confused, but they occupy completely different positions in the industry chain.
Optical Interconnects: Replacing copper cables with fiber optics
When an AI training cluster expands from a thousand cards to tens of thousands or hundreds of thousands of cards, copper cables hit the physical wall—signal attenuation, soaring power consumption, and heat dissipation issues.
This is not an engineering problem; it’s a hard constraint of electromagnetism.
Xizhi’s solution is to connect GPUs with optical fibers, reducing transmission latency by over 90%, and increasing GPU utilization (MFU) in the cluster by 50%.
This segment accounts for 71% of the company’s revenue in 2025 and is currently a cash flow business.
Specific products include OCS (All-Optical Switch) and oNET (Inter-Module Optical Network).
Optical Computing: Performing matrix multiplication directly with light
This is a completely different concept. It’s not about transmitting data; it’s about using light itself to perform calculations.
How is it achieved? Light is naturally suited for linear operations—light propagation in a medium is itself a mathematical transformation.
By passing two beams of light with specific intensities through a modulator, the total intensity after passing through directly results in matrix multiplication.
This is not a metaphor; it’s literally "light doing the computation."
What does this mean? Latency can be reduced to the 3-nanosecond level (compared to tens or hundreds of microseconds for GPUs), and energy consumption is lower by an order of magnitude.
The optical computing business will account for 19% of Xizhi’s revenue in 2025, a small absolute value (19.1 million RMB), but with a year-over-year growth of 579%.
This is its "second curve" for the secondary market.
Optical Modules: Translators between light and electricity
An optical module = an optoelectronic signal converter. It converts electricity into light and light back into electricity—a pure "translator."
Xizhi’s optical computing uses light itself to perform mathematical operations—completely different product categories.
Optical modules are the "vessels" of data centers, while optical computing is the "brain cells" of data centers.
These are two entirely different markets, different customers, and different players.
In the US stock market’s AI optical communication sector, Lumentum (LITE) has surged nearly 20 times over the past year—
but its story involves lasers + OCS, not traditional optical modules.
The global leader in optical modules is China’s A-share company Zhongji Xuchuang, which will generate over 20 billion RMB in revenue in 2024, supplying Google and Meta, with a PE ratio of 30-40 times.
However, Zhongji Xuchuang and Xizhi are not direct competitors; they occupy completely different positions in the industry chain.
Overseas, Coherent is a major player, with a 7-fold increase in one year.
Aside from Xizhi, there are no listed pure plays in the optical computing track—this is precisely the reason why Xizhi, as the "world’s first AI silicon photonics chip company," is rare in listing on the Hong Kong stock market.
Overseas, Lightmatter and Ayar Labs are still in private funding stages.
Full article available with better formatting.