NUPIAO reporter |
NUPIAO editor | Liu Fangyuan
On June 30, the Shanghai Stock Exchange officially accepted the STAR Market IPO application of Shanghai Xihe Technology Co., Ltd. (Xihe Technology for short), with Guotai Haitong as the sponsor.
Xihe Technology was founded in May 2021. Its main products are silicon photonics integrated chips. It’s an independent third-party chip design house running a fabless model, selling standardized silicon photonics chips to optical module makers worldwide. The company’s lineup covers 400G, 800G, and 1.6T speeds, and its chips are already shipping in volume to major domestic and international optical module vendors, powering AI clusters and hyperscale data centers.
Barely five years old, Xihe has already been named a 2025 national-level “little giant” specialized in advanced manufacturing. In 2025, its revenue hit 461 million yuan, a staggering 549% year-on-year jump, while its global market share in silicon photonics integrated chips reached around 13%. That’s one hell of a growth curve.
Riding the AI boom: 400G chips bring in over 90% of revenue
“Most silicon photonics startups founded around the same time are still struggling to get out of negative revenue territory, so Xihe’s ability to ramp up real sales is a huge deal. The team’s technical chops and industry connections run deep,” a fellow silicon photonics entrepreneur told NUPIAO.
Xihe’s core team has nearly three decades of silicon photonics R&D and mass production experience. Founder Wu Aimin has a Fudan academic background and spent over 15 years at the Shanghai Institute of Microsystem and Information Technology, leading advances in silicon-based on-chip light sources and large-scale integration that won a Shanghai Natural Science First Prize. CTO Dr. Feng Dazeng previously worked at Kotura and Mellanox and was part of the team that developed the world’s first commercial silicon photonics variable optical attenuator way back in 2006.
“Academia-born startups” aren’t rare in hard tech. By the end of 2025, after its latest external funding round, Xihe’s post-money valuation had reached 3.701 billion yuan. Its investor list reads like a who’s who of top-tier VCs and industry money: Oriza Seed, Oriza Holdings, a fund backed by Sunny Optical, SMIC Capital, Jinpu Investment, Fortune Capital, Xinding Capital, Lamon Peak Capital, Guangzhou Science City Venture Capital, Puhua Capital, and more.
Let’s rewind to 2021, when Xihe was founded. Back then, global AI compute investment hadn’t exploded yet, and silicon photonics was stuck in a rut with two big technical bottlenecks—“high-performance modulation” and “high-efficiency coupling.” Commercialization was painfully slow, and the whole industry was just biding its time. The tipping point came in 2023, when those bottlenecks finally cracked, and generative AI burst onto the scene, sending compute demand through the roof.
“They nailed every single inflection point,” the industry source told NUPIAO. “Just when silicon photonics took off, the technology breakthrough and market blast collided perfectly.”
The IPO prospectus tells the story: In 2023, Xihe’s revenue was a mere 6.23 million yuan, with a 28.34 million yuan loss. In 2024, revenue jumped to 70.95 million yuan. Then came 2025: revenue rocketed to 461 million yuan, with a net profit of 176 million yuan. That’s a three-year revenue compound annual growth rate of 760.34%.
But the revenue mix is what really catches the eye. In 2025, a whopping 92.47% of Xihe’s revenue came from 400G silicon photonics integrated chips. 800G contributed just 5.49%, and 1.6T a slim 2.04%.
400G products are hardly cutting-edge by today’s AI data center standards. In the North American AI data center market—the bellwether for optical module demand—800G is already mainstream, and surging demand has made DSPs and other components tight, with shortages becoming the buzzword of the year. Meanwhile, 1.6T products are shifting from small-batch to mid-volume production, and demand is just as hot.
But NUPIAO has learned from industry insiders that domestic data centers still lean heavily on 400G. The reason? China’s AI GPU deployment scale is capped by chip supply, so downstream optical module makers’ main shipping products remain 400G. Xihe precisely exploited that “time gap.”
That same silicon photonics industry veteran noted that 400G and 800G chips today don’t require sky-high technical heroics; what they really test is a company’s supply chain integration muscle.
Right now, the silicon photonics industry has evolved into a finely sliced division of labor: upstream, SOI substrate materials and wafer foundry services; midstream, chip design, packaging, and testing; downstream, optical module integration. Xihe runs a fabless model, focusing only on front-end chip design and final testing while outsourcing the manufacturing steps in the middle.
According to the prospectus, in 2025 Xihe sold 4.7543 million silicon photonics integrated chips with an 88% sell-through rate, yet still managed a gross margin of 62.82%—well above comparable peers like Youxun Co. (43.37%), Yuanjie Technology (58.15%), and Changguang Huaxin (32.44%).
By shipping in volume, Xihe has proven its engineering and supply chain integration chops. Even in a fabless model, you need the clout to secure higher production priority at foundries while guaranteeing capacity and developing custom processes. The prospectus reveals that Xihe “assisted the wafer foundry in developing a 12-inch silicon photonics process platform.” That kind of deep involvement in process development demands exceptionally strong industry relationships and a rock-solid reputation.
2026: the “Year of Silicon Photonics” and the next-gen product gauntlet
Xihe’s growth runway is built on the broader silicon photonics substitution wave.
In traditional optical modules, one of the most critical transmitter components is the EML laser. EML chips are pricey, and high-end EML capacity is monopolized by a handful of overseas players like Lumentum and Coherent, with orders already booked out to 2028 and a supply-demand gap of over 30%.
But silicon photonics takes a different path: it uses fewer CW laser sources plus a silicon photonics chip to bypass the EML bottleneck with lower cost and lower power consumption. The domestic CW laser supply chain, with players like Yuanjie Technology, is already shipping at scale, offering better cost advantages and stability.
A Nomura research note points out that silicon photonics’ penetration rate in 800G/1.6T optical modules is expected to hit 50%–70% in 2026, and industry forecasts are calling 2026 the “Year of Silicon Photonics.”
In the silicon photonics integrated chip competitive landscape where Xihe sits, the players fall into three camps.
The first camp is the old-guard integrated platform giants like Intel and Cisco, for whom silicon photonics is just one piece of a much larger business. Intel has been researching silicon photonics for over 30 years, has shipped more than 8 million photonic integrated circuits, and although it spun off its pluggable optical module business in 2023, it still holds onto chip design and manufacturing. Intel commands a 61% share in the datacom silicon photonics market, while Cisco holds nearly 50% in telecom.
The second camp is the optical module heavyweights like Zhongji Innolight and Eoptolink, who are Xihe’s customers today but are increasingly reaching upstream to develop their own silicon photonics chips for in-house modules. For example, Innolight’s subsidiary PaiXi Technology and Eoptolink’s acquisition Alpine are both working on proprietary silicon photonics chips.
The third camp is the “independent third-party” players like Xihe: they don’t make optical modules, they only make chips. The AI boom has spawned a crop of competitors in this space—Silicon Technology, Xiguang Xinchuang, DustPhotonics (acquired by Credo in 2026), Sailuo Technology, and others. Among them, Xihe is the largest and most commercially advanced.
Looking at the market opportunity, the silicon photonics chip pie is big enough for everyone. According to Frost & Sullivan, the global silicon photonics module market was 63.1 billion yuan in 2025 and is forecast to hit 263.3 billion yuan by 2030, growing at a 33% compound annual rate.
But for 1.6T and above high-end products, it’s still a battle of R&D and deployment capability.
Xihe’s 1.6T products contributed only 9.41 million yuan in revenue in 2025, a mere 2% of the total, which suggests they’re still in customer validation or small-scale trial production. The road to mass procurement—where cost, yield, and supply chain management write the score—still lies ahead.
Meanwhile, an even bigger wildcard is the NPO/CPO (near-packaged optics/co-packaged optics) technology shift. The traditional pluggable module is morphing into an optical engine form factor. The global NPO/CPO optical engine market was just 620 million yuan in 2025 but is expected to explode to 204.2 billion yuan by 2030, a terrifying 219% annual growth rate. That’s the next big bet for every silicon photonics company.
Xihe’s IPO aims to raise 2.43 billion yuan, and the biggest chunk—a budget of 952 million yuan—is earmarked for a “next-generation silicon photonics integrated chip R&D and industrialization project.” The plan is to spend five years developing single-channel 400G silicon photonics integrated chips and multi-channel, high-bandwidth, high-density chips for NPO/CPO applications, getting ready for the era to come.
Even on this silicon photonics track, heavyweights like Nvidia, TSMC, and Broadcom are just starting to flex their muscles. The industry insider told NUPIAO that right now, China is doing a great job of industrialization, while the U.S. has stronger early-stage research and 0-to-1 innovation capabilities.
Xihe Technology also faces real risks and challenges. First, customer concentration is high: in 2025, the top five customers accounted for 96.40% of revenue, and Customer A alone contributed 60.40%. Second, downstream customers are also developing their own chips. Optical module giants like Innolight and Eoptolink are moving upstream, and if their in-house chips achieve breakthroughs, Xihe’s customer demand could shift dramatically.
A 13% global market share is a pretty solid starting point for a five-year-old Chinese chip company, but the future depends on whether it can successfully rebuild itself through 1.6T, 3.2T, and eventually CPO products.
