Xili Photonics Technology (Hangzhou) Co., Ltd., a photonic chip company targeting optical interconnect in AI data centers, has closed a Pre-A funding round. The company did not disclose the exact amount; outlets including PEdaily described the round as being in the hundreds of millions of yuan.

The round was led by Fortune Capital (Futeng Capital), with participation from the Shanghai Future Industry Fund, Cornerstone Capital, the Greater Bay Area Fund, Skyview Capital, Shenzhen Capital Investment and listed company Zhiwei Intelligent. Lighthouse Capital acted as exclusive financial adviser.

Xili says the proceeds will fund R&D on thin-film lithium niobate photonic chips and on-chip system integration, supporting product iteration, wafer-level process coordination and mass-production validation. The company also plans to deepen work with AI data centers, optical module makers, communications equipment vendors and optoelectronic packaging and test firms.

AI clusters widen the need for optical interconnect

As large-model training and inference scale up, AI infrastructure is moving from single servers to clusters of 10,000 and even 100,000 accelerators. Chips must constantly exchange model parameters, intermediate results and cached data, and system performance increasingly hinges on how efficiently compute, storage and networking work together.

Data center optical modules are moving from 400G and 800G to 1.6T and beyond, with some vendors already working on 3.2T products. At the same time, interconnect is stretching from inside servers and racks out to whole data centers, campuses and cross-region networks, raising the bar on bandwidth, power consumption, latency and reach.

Scale Up addresses high-speed links among multiple compute chips inside a single machine or rack; Scale Out covers large-scale expansion between servers and racks; Scale Across involves longer-distance transmission between rooms, campuses and across metro areas.

Each scenario places different demands on optical components, but the shared goal is the same: add bandwidth while keeping energy use, signal loss and system complexity in check.

Thin-film lithium niobate emerges as a next-generation photonic material

Xili has bet on thin-film lithium niobate, or TFLN.

Lithium niobate offers fast electro-optic response, relatively low optical loss and good linearity, and has long been used in optical communications modulators. Advances in thin-film fabrication and micro-nano processing now allow such devices to shrink further and integrate on-chip with other optical components.

Compared with conventional lithium niobate devices, TFLN parts promise higher modulation bandwidth at lower drive voltages, making them suitable for high-speed electro-optic modulation, coherent transmission, multi-wavelength light sources and microwave photonic systems.

Material performance, however, does not automatically translate into commercial advantage. TFLN still has to solve wafer processing consistency, chip yield, packaging and coupling, thermal stability and volume manufacturing cost, and ultimately compete with mature or fast-moving alternatives such as silicon photonics and indium phosphide.

Light sources, modulator chips and photonic engines

Xili has mapped its product lineup to different transmission distances and interconnect types: multi-wavelength comb light sources, high-speed electro-optic modulator chips, coherent IQ modulator chips and photonic engines.

Multi-wavelength comb sources generate several equally spaced wavelengths from a single light source system, aimed mainly at dense parallel interconnect inside the rack. In principle, the approach cuts the number of discrete lasers and reduces the complexity of multi-wavelength systems.

The high-speed electro-optic modulator chips target 400G per lane as well as 1.6T and 3.2T optical interconnect, loading electrical signals onto the optical carrier. A modulator's bandwidth, drive voltage and insertion loss directly shape an optical module's data rate, power draw and cooling requirements.

The coherent IQ modulator chips and integrated photonic solutions are aimed primarily at links between data halls, campuses and metro networks. Coherent transmission uses amplitude, phase and polarization to raise capacity, but the systems are more complex and demand more of device linearity, loss and signal processing.

All of these products draw on the same TFLN material platform, wafer process and chip design capabilities. If the underlying process can be reused reliably, the company can shorten development cycles across products. For now, it has not disclosed sampling volumes, customer validation progress or mass-production timing for any individual product.

A founding team rooted in thin-film lithium niobate research

Xili was founded in January 2025. Founder and chief scientist Wang Cheng earned his undergraduate degree from Tsinghua University's Department of Microelectronics and a PhD in electrical engineering from Harvard University. He is currently with the Department of Electrical Engineering at City University of Hong Kong.

Public materials from City University of Hong Kong and Peking University show that Wang has worked for years on integrated lithium niobate photonic chips, spanning optical communications, microwave photonics, nonlinear optics and terahertz devices. His group has published multiple results on TFLN electro-optic modulators and integrated microwave photonic processing.

Xili says its core team has worked together for more than eight years, with members drawn from Harvard University, City University of Hong Kong and the Chinese Academy of Sciences, covering materials R&D, wafer processing, photonic chip design, device engineering and packaging coordination.

Moving from lab results to data center deployment still requires product reliability validation, process stability work and supply chain build-out. In high-speed optical communications in particular, customer qualification cycles are long, and new devices typically must pass performance, lifetime and environmental testing before entering volume purchasing.

From research results to scale deployment

Xili plans to push TFLN technology first into AI data centers and next-generation optical communications, then explore applications in 6G, optical computing and quantum technology.

Expanding AI computing power clusters have opened new market space for high-speed optical interconnect, but also drawn competition from silicon photonics, indium phosphide, TFLN and other photonic integration approaches. Each material platform has its own strengths in bandwidth, power, cost, manufacturing maturity and integration, and the eventual shakeout will depend on volume manufacturing and customer deployments.

For Xili, the metrics to watch include wafer-level process yield, sampling and qualification progress for core products, the timing of first mass production, actual customer orders, and whether TFLN devices can deliver competitive manufacturing costs on top of better performance.