Congratulations to FengRui's early-stage portfolio company Lightelligence on its successful IPO: From a Boston Dining Table to Photonic Infrastructure for the AI Era

峰瑞资本峰瑞资本·April 28, 2026

The World's First Publicly Traded AI Silicon Photonics Chip Company

On April 28, Lightelligence (01879.HK) officially listed on the main board of Hong Kong Exchanges and Clearing Limited, marking the birth of the "world's first AI silicon photonics chip stock" and becoming the first company in the global optoelectronic hybrid computing track to enter the capital markets.

FreeS Fund was an early investor in Lightelligence, providing continuous support from the Series A in 2017 through the A+ round in 2020, walking alongside the company for nearly a decade. Lightelligence and its founder Dr. Yichen Shen started from an MIT doctoral thesis and have since grown to 410 patents, 44 commercial customers, and the world's highest shipment volume of photonic computing chips — a path few have traveled.

/ 01 / It Started with a Dinner Invitation in Boston

Eleven years ago, Li Feng first met Yichen Shen in Boston in 2015. At the time, Shen had just finished his doctoral thesis, the company didn't exist yet, and FreeS Fund had itself just been founded that same year.

"Dr. Shen was quite well-known and accomplished in school, so we pulled him aside to chat — the main message being, you need to start a company," Li Feng recalled. During one dinner at a Boston restaurant, he happened to run into Shen dining with friends and simply sat down to talk.

For Shen, the encouragement from investors became the starting point of his entrepreneurial journey. "Right after finishing my thesis, I hadn't thought things through clearly — whether to stick with research or start a company," Shen said in a conversation with Li Feng. "A large part of it was influenced by investors. For getting me into this, Uncle Feng bears significant responsibility."

In 2017, Shen completed his PhD and that same year published a cover article in Nature Photonics titled "Deep Learning with Coherent Nanophotonic Circuits" — the first experimental validation of using light for deep learning computations, proposing a scheme for large-scale linear computing through silicon photonics technology and cascaded interferometers for artificial intelligence. This paper remains one of the most cited foundational works in photonic computing, and Lightelligence's entire technical path grew from it.

Also in 2017, Lightelligence was founded in Boston, with FreeS Fund becoming one of its early investors.

A decade ago in Boston, FreeS had a clear strategic footprint in frontier technology projects around MIT — another MIT physics PhD that Li Feng took an interest in during the same period was Shuhao Wen, founder of XtalPi, today's "China's first AI pharmaceutical stock." "These kinds of frontier projects all carry some controversy," Li Feng remarked with a smile. "Just like when we first looked at XtalPi, biotech investors in the Boston area looked at us with pity for backing it."

Back then, photonic AI chips were even more controversial. "AI chips" was still a new concept, to say nothing of how ahead of its time "photonic AI chips" was. In Li Feng's view, Dr. Shen's decision to start a company took considerable courage — while he possessed cutting-edge research, the industry widely doubted the technical feasibility and commercial potential of photonic AI chips. During due diligence, FreeS consulted numerous senior chip technology experts, most of whom held reserved or pessimistic views.

But time has provided the answer. Nine years ago, there were no more than two companies globally working on photonic computing; today there are dozens. In March 2026, Jensen Huang explicitly added optical lateral and vertical scaling to NVIDIA's roadmap at GTC, leading the industry to call 2026 the "first year of commercial silicon photonics." The bottleneck in AI computing power is shifting from single-card performance to interconnectivity, and optical interconnects are seen as the core direction for next-generation computing infrastructure — precisely the direction Lightelligence bet on from day one.

Yang Yongcheng, a FreeS partner who walked this journey with Lightelligence, puts it all in a longer time frame:

"From first principles of physics, the speed of light is the fastest in the universe, and light has inherent advantages in anti-interference and low power consumption. Using light technology for large-scale computation, especially AI acceleration, is most likely the optimal choice in terms of speed and efficiency."

"Lightelligence's vision is optical computing plus novel optical communications, comprehensively empowering AI computing centers — aiming toward the vast ocean of stars in the AI world."

/ 02 / Why This Direction? Why This Timing?

"When making investment judgments, we typically ask three questions: Why this direction? Why this timing? Why this solution?" Li Feng said. "At its core, early-stage investing comes down to two things — getting the direction right, and finding the person with the highest combined capability and caliber in that industry."

Why this direction? From the supply side, Moore's Law for electronic chips has hit a hard bottleneck — at certain transistor scales, quantum effects prevent further power reduction, and thinner copper wires mean higher resistance and more heat. From the demand side, big data, artificial intelligence, autonomous driving — each is exponentially driving up computing power needs. When supply can't keep up with demand, someone will inevitably seek breakthrough technologies at the foundational level.

However, pushing light technology toward large-scale computing is harder than it sounds. Yang Yongcheng explains: the challenge isn't just "light" itself — "basic capabilities that have a hundred implementation methods in electronics are still open problems in photonics"; the more fundamental challenge is that the entire information world remains fundamentally an electrical ecosystem. "Existing data lives in the electrical domain, so you need to build a bridge." This means Lightelligence had to solve not only fundamental technical and process challenges in photonics itself, but also the interconnection between the electrical and optical computing worlds — with difficulty, depth, and breadth beyond the ordinary. For this reason, Yang Yongcheng defines this as "not import substitution, but the world's most leading-edge direction."

Why silicon photonics as the solution? Photonic chips don't compete on advanced process nodes like 7nm/5nm/3nm — they leverage existing electrical chip manufacturing and packaging systems with mature supply chains that can scale to mass production. This avoids the production bottlenecks that stall many novel non-CMOS approaches. More importantly, light propagating in waveguides generates no heat, while linear operations and data movement account for over 80% of total power consumption in existing computing chips.

Why this timing? In 2017, companies capable of photonic computing were few and far between, and Lightelligence happened to emerge from that MIT paper to begin productization. "From research to industry, there's a 5-to-10-year intermediate period where everyone debates when it will explode, until that explosion point actually arrives," Shen once judged. Looking at it today, this judgment has proven correct, and the pace is accelerating.

The rapid growth of the AI computing market is forcing upgrades in optical communications: CPO (Co-Packaged Optics), NPO (Near-Packaged Optics), OCS all-optical switching, Scale-up optical links and other technical concepts are emerging in quick succession, rewriting the industry's interconnect standards. The multiple technical threads Lightelligence has accumulated over a decade — silicon photonics, high-speed electronics, optoelectronic hybrid integration, chip packaging — are "right on time" for these new demands, even carrying something of a "dimensional reduction attack" quality. Yang Yongcheng summarizes this dynamic as "laying eggs along the road" — applying new technologies developed for optical computing to AI optical communications as a commercial leverage play and a forward battle for optical computing's comprehensive entry into AI data centers.

But timing is never clear in hindsight. When FreeS placed its bet in 2017, the industry was far from today's "accelerating rhythm." Li Feng once offered a sober summary — "Team factors, environmental factors, industry factors, technical factors, market factors all change and influence how things unfold; today's static judgment doesn't necessarily directly correlate with tomorrow's dynamic results." The teams that truly navigate this uncertainty are those who go first and try first, stumbling along the way: "Because they go first and try first, they have the most chance of being the first to figure it out, regardless of whether the result and technical solution are the most optimal."

It is this underlying logic that led FreeS to make systematic bets in the silicon photonics direction. Many of its hard tech investments relate directly to silicon photonics, covering optical computing, optical interconnect chips, LiDAR chips, optical modules and optical engines, and more. Lightelligence, as one of the more foundational and disruptive players in this portfolio — building large-scale optoelectronic hybrid computing chips targeting over 90% of linear operations in AI computing — naturally sits at the core of this map.

| Optical computing and optical interconnect chips:

  • Lightelligence: A global leader in optoelectronic hybrid computing power, building next-generation high-performance optical chip products based on original optoelectronic hybrid computing architecture and large-scale optoelectronic integration technology.

  • LightStandard: With the mission to "build a new paradigm for AI computing with light as the standard," developing optoelectronic fusion computing cards and systems for large models and intelligent computing centers based on silicon photonics + phase-change material heterogeneous integration and Crossbar photonic matrix structures.

  • Anpai Xin Yan: Focused on thin-film lithium niobate photonic chips and optical engines, with products covering optical communications, data centers, consumer electronics, security, and automotive electronics.

| LiDAR chips:

  • Nanjing VisionICs: One of the global leaders in single-photon direct ToF (SPAD dToF) technology R&D and commercialization.

  • LuminWave: A global leader in LiDAR and 3D sensing technology, providing FMCW LiDAR, 3D industrial cameras, and perception solutions based on self-developed silicon photonics chips.

  • Leishen Intelligent System: Using an IDM model to provide commercial navigation and obstacle-avoidance LiDAR products for service robots, industrial automation, and intelligent vehicles, with product lines covering OPA solid-state, single-line, and multi-line LiDAR.

| Optical modules and optical engines:

  • Alpha Optoelectronics: Deeply rooted in optical communications technology, providing optical modules and optical engines for ultra-high-definition video, split TVs, mobile phones, VR devices, industrial cameras, medical imaging, and automotive electronics.

/ 03 / From a "Point" to a "Surface": A Decade's Journey in Hard Tech Entrepreneurship

As early as 2022, Shen shared with Li Feng the lessons he learned from his first five years of entrepreneurship. Four years later, these insights still hold — and explain even better why Lightelligence has reached where it is today.

The first is the understanding of "point" versus "surface." "When doing research, you might see only a breakthrough within a larger system, and easily mistake that point for the whole world. But once you actually build products, you realize that point is just one of many necessary links," Shen said. To build a truly usable optoelectronic hybrid computing chip, Lightelligence had to simultaneously get optical chips, electrical chips, and software right — "just like electric vehicles still need steering wheels and tires."

The second is decision speed. "Early on, we spent a long time pursuing 100-point solutions. Over time, we increasingly realized that what's more important in entrepreneurship is finding 60-point or 70-point solutions in the shortest time possible," Shen reflected. "This may actually become a shackle for many scientist-entrepreneurs — having scored high all their lives, the pursuit of perfection反而 becomes costly in terms of time." Li Feng deeply resonates with this: "In research you pursue 100 points; in industry you need pragmatism — solve the problem first, then adjust as you go."

The third is the deadlock between "major customers versus general ecosystem." Major customers want "plug-and-play," expecting new products to seamlessly adapt to thousands of existing application scenarios; while NVIDIA's software ecosystem built over decades is something no startup can directly challenge in the short term. Lightelligence's breakthrough strategy was to first find that sharpest "needle point" — in a few of the most painful scenarios, using light's efficiency advantage to outweigh generality needs, then gradually expand influence.

Li Feng once offered a precise summary of this logic: "The biggest challenge for general-purpose underlying chips is that only the largest customers can fully utilize them, yet startups struggle to build ecosystems in the short term — these two constraints reinforce each other. If there is genuinely clear technical advantage, finding a sufficiently 'sharp' entry point where efficiency outweighs generality needs can pierce through that window paper."

The fourth, and the most fundamental thing that has carried Lightelligence to today — "firm, even fanatical belief."

"The biggest challenge in hard tech is the long time horizon; many teams change direction halfway through. But often the most important breakthroughs come from a team that believes in a major direction with near-religious fervor over a long period," Shen said. Lightelligence has not changed direction in ten years; that itself is the answer.

Today, four years later, that "needle point" has pierced through the window paper. Lightelligence's second-generation optoelectronic hybrid computing product "Tianshu" carries a 128×128 programmable optical matrix with over 40,000 photonic devices integrated; the jointly launched LightSphere X in 2025 became the world's first distributed optical circuit switching solution for GPU super-node interconnects, improving model floating-point operation utilization by over 50%. By the end of 2025, Lightelligence had achieved commercial deployment with 44 customers, cumulative global shipments ranking first in photonic computing chips, and ranked first in China's independent Scale-up optical interconnect solutions market with an 88.3% market share.

The needle found its mark; the window was pierced.

Closing Words

From that dinner table conversation in Boston in 2015 to the bell-ringing ceremony at the Hong Kong Stock Exchange in 2026 — eleven years in total. Lightelligence has walked to the threshold of becoming the "world's first AI optical computing stock," and FreeS has been fortunate to accompany it for nearly a decade — a "light-chasing" journey that has been both long and fleeting.

Congratulations to Lightelligence on its successful listing! We hope that standing at this new starting point, Lightelligence will continue to do what is right rather than what is easy — carrying forward its unwavering original mission of optical computing toward the true vast ocean of stars in the AI world.

When Instruments and AI Rush Toward Each Other, How Should We Define Performance in the Next Era?

Billions of Years Ago, Living Things Built the First "World Model"

Global Capital Goes Risk-Off: Where Did the Money Go? | Macro Musings