1. Summary
This brief is for European companies, research groups and investors who want to understand China's photonics industry for AI. Large AI models are trained and run on clusters of thousands of GPUs, and the links between those chips increasingly use light instead of copper. The lasers, modulators, chips and modules that turn electrical signals into light and back make up the "optical link". This brief maps who makes those parts in China, how mature each technology is, where the companies are, how they are funded and who they sell to. It is based on public sources (government documents, company filings, funding announcements and industry research) as of November 2026. The full report has the details and the sources.
A few terms used below: InP (indium phosphide) and TFLN (thin-film lithium niobate) are materials for lasers and modulators; CPO and NPO (co-packaged and near-packaged optics) move the optical parts right next to the processor; optical I/O connects chips directly with light; OCS (optical circuit switching) routes signals as light without converting them; electrical ICs are the chips that drive and read the optical parts.
Currently, China has suppliers at every stage of the optical link used in AI data centres. The large module makers grew on orders from overseas cloud companies. Government policy, state funds and private venture capital pay for the newer chip-level work, and IPOs in Shanghai and Hong Kong provides early investors the exit. Among the technologies, the parts nearest the processor are still at an early stage, and most of the companies working on them are startups.
| Area | Summary |
|---|---|
| Policy | AI is already a national priority for a long time. In June 2026 a ministry document named optical interconnect for the first time, Shanghai and Suzhou subsequently have incorporated optical interconnect in their strategies and run their own silicon photonics programmes. |
| Technology | China already makes fibre and pluggable modules at scale for years and sells them worldwide. The current technology that starts to ship in volume are InP lasers, TFLN modulators and electrical ICs. CPO, optical I/O, OCS and optical computing are still at early stage. |
| Companies | Innolight and Eoptolink are the two largest module makers in the world. Most upstream chip makers and frontier companies are smaller listed firms or startups. |
| Geography | 28 of the 51 companies mentioned in this document are in the Yangtze River Delta, mostly in Suzhou and Shanghai. Wuhan and Beijing are the next clusters. |
| Business | For companies, the leaders sell mainly to overseas cloud companies. For investors, the current main exit is through IPOs, especially after Shanghai STAR Market and Hong Kong opened to loss-making tech companies in 2025 and 2026. |
Relevance to Europe
China is still building up high-speed InP lasers and chip-level photonic integration, two areas where Europe has long experience, for example the InP integrated photonics platform in Eindhoven. Chinese suppliers mostly win on scale, speed and cost, while Europe's integrated photonics is sold on performance and IP. The two sides are strong in different places, which leaves room for collaboration.
The startups are already exploring these links. A co-founder of Zhilian, a Beijing optical switching startup, holds a PhD from TU/e in Eindhoven. The founder of Bopu in Shanghai worked at imec in Belgium. Xilian in Suzhou bought the German company Sicoya in 2021, and Robotechnik, also in Suzhou, bought ficonTEC, a German maker of packaging and test equipment for silicon photonics. Several TFLN, InP epitaxy and OCS startups are adding capacity or are in engineering validation, and may need European equipment, process know-how and IP.
2. Technology matrix
As the figure clearly shows - the layers nearest the chip are the least mature. This is generally true globally. The figure below lists out the major players in China.
3. Geography
The map below shows the geographical location of the companies listed above, with 7 excluded as there is no public information about their HQ location.
28 of the 51 companies are in the Yangtze River Delta. Suzhou alone has six listed companies, covering lasers, devices, modules and test equipment.
| City | Profile | Main companies |
|---|---|---|
| Suzhou | Has a local supply chain from lasers and optical devices to modules, packaging equipment and test instruments. The city opened a silicon photonics cluster in October 2026, and an 8-inch silicon photonics line is due to start by the end of 2026. | Everbright, Innolight, Source Photonics, TFC Optical, Robotechnik/ficonTEC, Stelight |
| Shanghai | Has more frontier startups than any other city, in optical I/O, OCS and optical computing. Published China's first silicon photonics plan in March 2025 and set up a cluster in Zhangjiang. | Lightelligence, NovelSi, Bopu, Xinguangjie, Xinjie Guanghe |
| Wuhan | Home to the older optical communications and fibre companies. Optoelectronics output was about RMB 660bn in 2025. | HGTECH, Accelink, YOFC |
| Beijing | Has no listed interconnect company but five startups in optical I/O, fast optical switching and optical computing. | Guanglian Xinke, Lingxi Photonics, Zhilian |
| Other cities | Xi'an has InP laser makers and a silicon photonics pilot line, Xiamen has electrical ICs, Kunming has InP substrates, Chengdu has Eoptolink and Shenzhen has Huawei. CIOE takes place in Shenzhen. | Yunnan Germanium, Yuanjie, UXIC, Eoptolink, Huawei |
4. Funding
The closer a company is to chips and servers, the larger the share of state money. Photonics startups raise from a mix of state funds, corporate investors and VCs, and the mix changes from one stage to the next. Competitions are one effective way in. Shanghai's Zhangjiang Cup for silicon photonics, part of the city-wide "Innovate in Shanghai" competition, puts winners in touch with the Shanghai Future Industry Fund, CAS Star and Zhangjiang Hi-Tech for financing, incubation and chip tape-out. Startups like Saili and Xili won its top two enterprise prizes in 2026.
National funds
National funds are large investment funds set up by the central government, with the Ministry of Finance, state-owned banks and state-owned companies as shareholders. They invest over long terms in industries the state treats as strategic, either directly in companies or through other funds. These are the two most relevant to AI hardware:
| Fund | Size | Set up | Scope |
|---|---|---|---|
| National IC Industry Investment Fund, phase III (Big Fund III) | Registered capital RMB 344bn | May 2024 | The whole semiconductor chain |
| National AI Industry Investment Fund | RMB 60.06bn, 13-year term | January 2025 | Public deals include compute chips and large-model companies |
State share of AI equity funding
This measures something wider than the national funds above: the share of money in AI funding rounds that comes from any state-linked investor. So it includes the national funds above, as well as guidance funds set up by provincial and city governments to invest alongside private capital, state-owned investment platforms and state banks. It shows how much of each part of the AI industry is paid for by the state rather than by private VCs.
| Segment | State share |
|---|---|
| AI chips and servers | Government guidance funds and investment platforms 47%, state banks 14% |
| AI cloud | About 30% |
| Models and applications | Mostly private; no share given |
Across all AI equity deals, the state-linked share rose from about 13% before the pandemic to about 30% in 2024 and 2025. It was about 25% from January to August 2026.
Typical fund raising rounds
| Stage | Main investors | Typical path | Examples |
|---|---|---|---|
| Seed and angel, from leading VCs | Leading private VCs | Founders from big tech companies or top labs raise angel rounds directly on their record | Xinguangjie's angel round |
| Seed and angel, from incubators and universities | Incubators from universities or local governments, local seed funds, provincial state VCs | Shenzhen has a RMB 2bn seed fund for lab and academician teams. Shanghai's silicon photonics cluster runs a proof-of-concept platform and a startup competition to find early projects | Xinjie Guanghe was incubated at a Shanghai Jiao Tong University's state lab |
| Seed and angel, from local government | Local governments' angel funds and guidance funds | Often invested by local gov's on city level with the condition that the company sets up local offices/factories locally. Suzhou New District has a RMB 2bn angel fund. Shenzhen's angel fund of funds can reach RMB 10bn and dropped its 1.75x local reinvestment rule in 2026 | Liangyin's angel round was led by Zhuhai city state capital |
| Series A onwards | Provincial state VCs from other regions, private VCs, corporate investors, existing shareholders | Shanghai's future industry fund and eight private funds have put more than RMB 900m into silicon photonics. Module makers buy in through industry funds or direct stakes | Guoke Guangxin's B+ round included two provincial state VCs from outside Zhejiang |
| Growth and pre-IPO | Central state-owned companies and national funds | Large late rounds, followed by a listing on the Shanghai STAR Market or in Hong Kong | Guanglian Xinke's A+ round and Liobate's Series B both brought in central state or national fund money |
Recent photonics rounds
| Company | Segment | Round | Investor types |
|---|---|---|---|
| Liobate | TFLN modulator chips | Series B of several hundred million RMB in August 2026 | Mainly private VCs, with a national fund and a corporate investor |
| Guanglian Xinke | Optical I/O, NPO | Series A of about RMB 500m in June 2026 and A+ of about RMB 1bn in September; post-money valuation above USD 1bn | Mainly leading VCs, the A+ added central state and state-backed money |
| Xinguangjie | NPO, CPO, optical I/O | Registered in April 2026; five rounds and more than RMB 1bn within five months | Mainly leading VCs, with a small corporate stake |
| Guoke Guangxin | Silicon photonics, NPO/CPO | Series B+ of several hundred million RMB in September 2026 | Led by a private VC, with provincial state VCs from Sichuan and Shandong |
| Lightelligence | Supernode interconnect, optical computing | Seed in 2018; Series C in September 2025 | An internet company led the early rounds with VCs alongside; the Series C added central and local state investors |
| Guangbenwei | Phase-change optical computing | Several billion RMB over the past year | Mainly local state capital, with a corporate investor and an incubator |
5. Marketing, sales and exits
| Area | Summary |
|---|---|
| Marketing | Current selling points focus on higher GPU utilisation, lower energy per bit and the ability to deliver at volume. Chinese customers also weigh supply security. |
| Sales | Albeit domestic interest, the leaders depend on exports. 90.6% of Innolight's 2025 revenue came from outside China, 57.3% from the US, and its top five customers made up 76%. Supernodes built in China are a second source of demand. Revenue moves with hyperscaler capex, exchange rates and trade policy. |
| Exits | Mostly through IPOs and the market is very positive. Lightelligence (April), Innolight (July) and Dongshan Precision (October) listed in Hong Kong in 2026. There are also M&A exits, such as Dongshan's purchase of Source Photonics and Robotechnik's purchase of ficonTEC, but they are smaller than the IPOs. |
Full report
China's Photonics-for-AI Ecosystem
The full 57-page report covers strategy and policy, the technology options, funding, every company in the charts, sales and exits, with sources. Leave your email and we'll send you the download link.