Bopu Semiconductor Raises Series B to Bring Silicon Photonics and Optical Switching to AI Data Centres

Bopu Semiconductor, a Shanghai startup that designs silicon photonics chips, has closed a Series B round worth “several hundred million yuan.” Huaying Capital led the round. The company didn’t give an exact figure, but a raise of that size likely puts it in the tens of millions of US dollars.

Kailian Capital, an intellectual property fund under Shanghai Science and Technology Innovation Group, listed construction company Southeast Space Frame and Shunyi Investment also joined. The deal was announced on November 6, 2026, about a year after Bopu’s Series A, which was also several hundred million yuan and was led by Shanghai Guohe Investment in October 2025.

Two large rounds in a year is a lot for a company founded in 2022. It also says something about where investor money in AI hardware is flowing right now.

The problem Bopu is working on

When people talk about AI infrastructure, they mostly talk about GPUs. But the chips are only part of the story. Training and running large models means moving huge amounts of data between thousands of processors, constantly. The connections between them are turning into a bottleneck.

Traditional electrical connections waste power and struggle at very high speeds. Light doesn’t have the same problems. That’s why optical interconnects, which send data as light instead of electrical signals, are becoming central to how AI data centres are built.

Silicon photonics is the technology that makes this practical at scale.

It builds optical components such as modulators, detectors and switches directly onto silicon chips, using manufacturing processes similar to those used for regular semiconductors. That makes them cheaper to produce in volume.

Yao Ningbo, a hard-tech partner at Kailian Capital, put it bluntly. Growth in AI computing, he said, is turning optical interconnect “from optional to mandatory” (quote translated from Chinese).

What Bopu actually makes

Bopu has a few lines of business, and they’re further along than you might expect from a four-year-old company.

Its main product today is high-speed optical transceiver chips, the parts that convert electrical signals into light and back again. According to Chinese media reports, its 400G and 800G transmit chips are already in volume delivery, built on 100G-per-lane designs. The company says it has started on 200G-per-lane products too, and earlier this year it reported sampling a 1.6T chip with a key customer.

It also licenses its designs. Customers can use Bopu’s modulators, detectors and couplers, or entire chip designs, and have them manufactured at partner foundries. On top of that, Bopu sells wafer-level test equipment, which it originally built to test its own chips.

The newest bet, and the one this round is really about, is optical circuit switching, or OCS.

Why optical switching matters

In most data centres today, network switches work electronically. Data arrives as light, gets converted to an electrical signal, gets routed, then gets converted back into light. Each of those conversions costs power and adds delay.

An optical circuit switch skips that. It routes light directly, reconfiguring paths between servers as needed. For giant AI clusters, that can mean lower power use and a network that’s easier to scale. Some investors describe it as taking the electronics out of the data centre network.

Bopu has built its own reconfigurable silicon photonic OCS. An 8x8 switch chip is in production, according to the company’s website, and a 32-port switch chip and full system prototype are being tested with customers in China.

Where the money is going

The company has laid out three priorities for the new funding.

First, scaling up OCS production. Bopu wants to move from supplying switches on demand to steady batch delivery within 12 months. That means expanding wafer capacity, refining packaging and putting a proper quality system in place. It’s also developing much larger switch matrices, at 64, 72, 128 and 144 ports.

Second, growing its licensing and custom chip business. That includes optical engines for near-packaged and co-packaged optics, where optical components sit right next to the main processor or switch chip.

Third, winning more customers. Bopu sells to optical module makers, switch manufacturers and cloud providers, though it hasn’t named any of them publicly.

The founder

Bopu’s co-founder and general manager, Xing Yufei, has spent most of his career in silicon photonics. He studied optoelectronics at Zhejiang University, did doctoral work in Singapore and finished a PhD in integrated silicon photonics at Ghent University in Belgium, where he researched manufacturing yield at imec.

He later worked at Xanadu, the Canadian quantum photonics company, before returning to China and co-founding Bopu in 2022.

He told 36Kr that he did his PhD with commercialisation in mind from the start. As for what comes next, he was realistic: “The next challenge is still the speed of commercial adoption” (translated).

A crowded and fast-moving field

Bopu isn’t raising money in isolation. Chinese silicon photonics startups have been on a funding run this autumn. Xinguangjie raised close to 1 billion yuan across two rounds for optical engines in AI clusters, also with Huaying Capital involved. Guokeguangxin raised a B+ round of several hundred million yuan. Jiuwei Photonics raised nearly 100 million yuan for phase-change optical switch chips.

There’s a strategic angle here too. Investors in this round repeatedly pointed to domestic supply chains and “self-controllable” core technology. With access to advanced chips and equipment restricted, China is pushing hard to build its own capabilities across the AI hardware stack.

Optical interconnects are part of that.

Globally, the direction of travel is the same. Nvidia announced plans in October to expand optical engine capacity for co-packaged optics, a sign that optics is becoming a core part of AI system design rather than an add-on.

What to keep an eye on

The big question for Bopu is whether its optical switches move from testing to repeat orders. Customers judge these products on how the whole system performs, not just the chip, so Bopu’s switches need to work smoothly with existing network hardware and software.

Yield and consistency at higher volumes will matter too. Bopu has reported strong yields on its transmit chips in test production, but scaling a new product line is a different challenge.

If it gets both right, Bopu could become one of the key suppliers in China’s AI data centre build-out.