Lingdong Xinguang has raised tens of millions of yuan to commercialize silicon photonics, the practice of building chips that move data with light on silicon wafers, into chip to chip links by 2027, though it has yet to name a customer or publish an independent benchmark.
AI training clusters are scaling past the physical limits of copper. Inside a single rack, accelerator-to-accelerator links carry terabits per second across centimeters of wire; across a cluster, the same signals traverse meters. Both distances are starting to cost more energy, generate more heat, and consume more board space than the chips themselves can pay for. The industry has a candidate fix, replacing those electrical traces with optical links that move data as light, and a small set of Chinese companies, including a 2022 spin-out from Tsinghua University, are betting that the inside-the-package version of that fix is theirs to win.
Lingdong Xinguang (灵动芯光), founded in 2022, sits in the silicon-photonics corner of that race. Silicon photonics is the practice of building optical components (waveguides, detectors, modulators, gratings, couplers) onto the same kind of silicon wafers that CMOS logic uses, so they can be manufactured in the foundries the chip industry already runs. The category has been framed by the company's founder as a "foundational technology platform" at a stage analogous to the early node of integrated circuits, with Europe and the United States leading on advanced process and ecosystem work and China closing the gap. That framing is the company's own, rather than third-party validation, per a 36kr Hard氪 exclusive on the company's recent funding round.
The bet that distinguishes Lingdong Xinguang from peers is wavelength density. Optical links between chips usually work by shining different colors of light down the same fiber, with each color carrying its own data lane. Coarse wavelength division multiplexing (CWDM) uses a small number of widely spaced lanes; dense wavelength division multiplexing (DWDM) packs many more, narrower lanes into the same fiber, multiplying the bandwidth each port can carry. The company's stated differentiator is a DWDM multi-wavelength silicon-photonics light source with a 32-to-64 wavelength roadmap. That is the number to watch if you want to compare it with what other optical-I/O players have put on the record. The company has not yet published an independent benchmark for that roadmap.
The team behind the bet is anchored by a Tsinghua University professor of electronic engineering who, per the founder interview in the 36kr piece, began silicon-photonics research during a 2009 visiting scholarship at MIT and has continued the work for more than fifteen years through his research group at the university. His titles, including a reported deputy-directorship at the Tsinghua Information Optoelectronics Research Institute, and the precise year of the MIT visit are founder-attested and have not been independently verified. The co-founder, Wang Hui (王晖), is a Tsinghua graduate whose résumé the company describes as including R&D and general-management roles at O-Net (昂纳), Multiplex, BKtel, and Xinfeng Electronics. Those companies supply optical components and subsystems to telecom and industrial laser buyers.
The product pipeline runs in three layers. The most mature is an ultra-narrow-linewidth laser built as a silicon-photonics chip combined with hybrid-integrated III-V gain material; the company says it is already in volume production with customer orders. Second is an FMCW (frequency-modulated continuous-wave) LiDAR light source, aimed at automotive and industrial sensing. Third, and the focus of the new funding, is an optical-I/O chiplet called SmartPHY, a chip-package-internal device that converts electrical signals to optical and back, so that two adjacent accelerator dies can talk to each other over light instead of copper. The product the company calls SmartComb is the multi-wavelength light source that feeds those links. If the architecture works as the company describes, "tens of thousands of AI chips can behave like one giant chip," in the words of the founder as quoted by 36kr.
The receipts to back that vision are still thin. The company has not named a customer beyond a generic reference to "头部客户" (top-tier customers) in the funding announcement, has not disclosed revenue, and has not put a public benchmark on its 32-to-64 wavelength roadmap or on the energy-per-bit and latency figures that would let a buyer compare SmartPHY against other optical-I/O offerings on the market from established players in the United States and Europe. The 2027–2028 commercialization window for the optical-I/O chiplet is a company-side timeline, not a market consensus.
The capital behind the bet is a tens-of-millions-of-yuan Angel++ round, with Panlin Capital (磐霖资本) as lead investor and Tongfang Investment (同方投资) plus a Huize Tiancheng sub-fund partnered with Shenzhen Angel (汇泽天诚) as co-investors, according to the 36kr Hard氪 exclusive announcing the deal. Use of funds is R&D on chip-to-chip optical interconnect, productization of the SmartComb multi-wavelength source, and development of the SmartPHY optical I/O chiplet. Panlin's track record in early-stage Chinese hardware and biomedical bets is documented in its 36kr Pitchhub profile, with additional public coverage in Sina finance and the 21st Century Business Herald, both of which describe the firm's pattern of leading or co-leading first-round checks into Chinese deep-tech teams.
What to watch next: a public benchmark on the multi-wavelength source; a named hyperscaler, accelerator vendor, or systems integrator as a design partner; a package-level demo of optical chip-to-chip links on a working AI accelerator; and any disclosure on manufacturing yield, since silicon photonics' economics depend on whether the foundry process can hold tolerances across the hundreds of devices each die needs. Until those land, the company's bet is that its Tsinghua research lineage, its DWDM wavelength density, and its hybrid-integration manufacturing plan are the right pieces for the moment copper runs out of room. Whether 2027 arrives with those pieces in production is a question the market, not the founder, will answer.