Lightelligence In-Depth Zen Horizon Research

1.2 Six Financing Rounds: From RMB 190 Million to RMB 7.8 Billion, the Private Market Built a Unicorn in Eight Years

Before listing, Lightelligence completed six financing events, raising about RMB 2.3 billion in total. The step-by-step valuations disclosed in the prospectus, all in RMB, clearly trace the curve that lifted the company upward:

Data source: HKEX prospectus. This table also clarifies several apparently conflicting valuation figures in media reports. The widely cited "RMB 6.28 billion" was the post-money valuation of the March 2024 C1-3 rounds, while "RMB 7.8 billion" was the post-money valuation of the April 2025 C4 round, the last round before the IPO. They refer to different points in time and are not contradictory. The investor lineup in the C rounds, which raised more than RMB 1.5 billion cumulatively in 2024-2025, is especially noteworthy: funds under China Mobile, Shanghai State-owned Capital Investment, China Reform Fund, Pudong Venture Capital and other state-backed investors, plus follow-on investment from existing shareholders such as Tencent and Sequoia China Securities Times. The shift in the shareholder base, from early U.S. dollar venture capital to later "national team" investors, itself tells a story of moving from a Silicon Valley startup to a domestic-compute flagbearer.

1.3 The Listing Moment: Hong Kong 18C's First Pre-Commercial Benchmark and an IPO Frenzy

On April 28, 2026, Lightelligence listed on the HKEX Main Board, becoming the first benchmark "pre-commercial" listing under Chapter 18C for Specialist Technology Companies, with CICC and Guotai Haitong as joint sponsors Securities Times. Its bookbuilding process was a textbook frenzy:

• Pricing: the offer range was HK$166.6-183.2, and the final price was set at the top end, HK$183.2. The global offering included about 13.795 million H shares, plus an over-allotment option of about 2.07 million shares, fully exercised on May 11, raising net proceeds of about HK$2.377 billion Sina Finance.

• Cornerstones: about 20 institutions subscribed for a combined roughly US$209.9 million, or about HK$1.644 billion, representing about 65% of offered shares. Alibaba led with US$26 million; GIC subscribed US$20 million; Baillie Gifford, BlackRock, Fidelity, Schroders, and Temasek each subscribed US$15 million Sina Finance.

• Subscription heat: the Hong Kong public offering was oversubscribed by about 5,784.7x; margin financing was oversubscribed by more than 2,100x; and grey-market trading rose about 351% Eastmoney.

• First day: the shares opened at about HK$880, roughly 380% above the offer price, at one point rose more than 400% intraday, and closed at about HK$886, up about 384%, implying a closing market capitalization of about HK$81.5 billion and setting a multi-year record for first-day gains among Hong Kong IPOs Caixin Global.

The structural reason behind the frenzy was an extremely small float plus overheated IPO sentiment. H shares represented only about 15%-17% of total share capital, and cornerstone investors locked up more than 60% of those shares. The actual freely tradable secondary-market supply was tiny, so any sentiment was amplified. This is critical to understanding the stock's later violent swings, discussed in Sections VI and VII.

II. Financial Review Over Time: The Huge Accounting Loss Is an Accounting Illusion, but Operating Losses and Cash Burn Are Real

Lightelligence's financial statements look frightening at first glance, but reading them correctly requires stripping out an accounting illusion.

2.1 Revenue: High Growth, but Still Startup-Scale in Absolute Terms

Revenue during the track record period (2023-2025), in RMB, was 38.23 million -> 60.19 million -> 106 million, with 2025 revenue up 76.7% year on year and a three-year compound growth rate of about 66.9% HKEX prospectus, Sina Finance prospectus analysis. The growth rate looks attractive, but remember the denominator: this is a company worth tens of billions of Hong Kong dollars whose annual revenue has only just exceeded RMB 100 million. In the revenue mix, optical interconnects accounted for about 79.2% in 2025, including about RMB 75.58 million from scale-up products, while optical computing accounted for about 20.8%, or about RMB 22.09 million, up roughly 579% year on year Sina Finance prospectus analysis. In other words, the "optical computing" business that underpins the company's valuation imagination generated only a little over RMB 20 million in one year; the real cash-generating business today is optical interconnect, which is easier to engineer and can be sold first.

2.2 Losses: Of the RMB 800 Million Accounting Loss, RMB 800 Million Will Disappear After Listing

This is the most important distinction in the entire report. The prospectus discloses three full layers of loss measures, in RMB:

Data source: HKEX prospectus. The accounting net loss expanded 131.5% year on year in 2025, and cumulative losses over three years were about RMB 2.491 billion, which looks alarming. But of the RMB 1.342 billion loss in 2025 alone, RMB 843 million came from "fair value changes of financial instruments issued to investors," such as preferred shares. This is a typical phenomenon for loss-making companies listing in Hong Kong: the faster the company's private-market valuation rises, the more the "book value" of these preferred shares rises, and the larger the "loss" recognized. But they automatically convert into ordinary shares at listing, are reclassified from liabilities to equity, and will no longer recur afterward HKEX prospectus.

After removing this accounting illusion, the company's own "adjusted net loss," adding back share-based payments and listing expenses, was RMB 264 million / 296 million / 271 million over the three years. It was almost flat, and even narrowed slightly in 2025. This stands in sharp contrast to the "131.5% expansion in accounting losses" and is the only useful measure for assessing operating fundamentals.

There is one footnote that cannot be ignored: adjusted losses were able to stay flat largely because "other income," mainly government subsidies, offset them. Other income jumped from RMB 13.90 million in 2023 to RMB 192 million in 2025, about 1.8x that year's revenue HKEX prospectus. Excluding subsidies, operating cash leakage would look worse than the headline accounts. This is the other side of a structure in which the "national team" can be customer, shareholder, and subsidy provider at the same time.

2.3 Gross Margin, Cash Burn, and Survival Capacity

Gross margin fell in one direction over three years: 60.7% -> 53.5% -> 39.0% HKEX prospectus. The reason was not accidental. It was structural: low-margin optical interconnect products became a larger share of the mix, and the company gave price concessions to a single major customer. The prospectus discloses that the gross margin of one overlapping major customer in optical interconnect was only 36.4% in 2024 and 23.4% in 2025, while optical-computing customer gross margin remained above 70%. The shift in revenue mix toward low-margin optical interconnects for a single major customer is the root cause of the gross-margin step-down, which also means the current volume ramp has been achieved to a meaningful extent by trading margin for revenue, with weak pricing power.

R&D is the main battlefield of cash burn: 2025 R&D expenses were RMB 479 million, equal to 450% of that year's revenue, and were fully expensed with no capitalization to flatter profit HKEX prospectus. Net cash flow from operating activities over the three years was RMB -176 million / -182 million / -361 million, while average monthly cash burn rose from RMB 19.30 million to RMB 34.20 million and accelerated clearly in 2025, which the company attributed to front-end R&D for PACE 3.

Survival capacity is this company's lifeline, and the IPO addressed an immediate funding need. Based on the company's own estimate, using the year-end 2025 cash balance and assuming future average monthly burn of about RMB 39.30 million, cash would have supported only about 16 months. After including IPO proceeds, it can support about 64 months, or roughly 5.3 years HKEX prospectus. In one sentence: this IPO essentially extended the cash runway from about 1.3 years to about 5 years. That buffer is critical to whether the company can reach scaled monetization, and it also explains why management needed to complete the offering at the highest possible price while sentiment was hottest.

2.4 A Misreading That Must Be Clarified: The 473% Liability-to-Asset Ratio Does Not Mean Insolvency

Many bearish articles cite a "473% liability-to-asset ratio and 0.16 current ratio" to argue that Lightelligence was "technically insolvent." This is a misreading. Of the company's RMB 5.386 billion in total liabilities at the end of 2025, about RMB 4.9 billion, or about 91%, was the aforementioned "financial instruments issued to investors," namely preferred shares that would convert into ordinary shares at listing HKEX prospectus. After listing, these instruments became equity, net liabilities turned into net assets, and the 473% figure became distorted and disappeared. Likewise, the valuation-adjustment clause under financing agreements that would have triggered redemption if a qualified IPO had not been completed by the end of 2028 was released by the successful listing in April 2026. A research report should not treat the pre-listing accounting illusion as a real solvency crisis. Lightelligence's real financial risk is not "insolvency," but "continuous cash burn, no visible profitability timetable, and eventual need for refinancing."

2.5 Customer Concentration: 40% of Revenue Depends on One Customer

There is a clear flaw in revenue quality: the largest single customer rose from 31.1% of revenue in 2023 to 40.6% in 2025, while the top five customers accounted for 78.9% HKEX prospectus. The total number of customers increased from 12 to 44 over three years, still an early-stage profile. More subtly, the largest customer in 2025 came from the low-margin optical interconnect business. In other words, the customer contributing the most revenue was precisely the order with the weakest bargaining position and thinnest margin. In addition, trade receivables surged about 355% year on year in 2025, far above the 76.7% revenue growth rate. Whether collection pace matches revenue recognition is a cash-flow quality issue that requires ongoing monitoring.

III. Business Model and Moat: The "Picks-and-Shovels" Seller in Optical Interconnect, the "Option" in Optical Computing

3.1 Two Legs: One Makes Money, One Carries the Story

Lightelligence's business engine has two wheels, built on three self-developed chip capabilities: oNOC (optical network-on-chip), oNET (optical network between chips), and oMAC (optical matrix computing) Sohu reposting prospectus.

Optical computing, the oMAC route, is the company's "signboard" and "option." The principle is to use photonic circuits to perform matrix-vector multiplication directly. In theory, optical transmission has no ohmic heat, low latency, and high bandwidth. The company's PACE chip, first launched in 2021 with results published in the main journal of Nature in 2025, integrates more than 16,000 photonic devices on a single chip and performs 64x64 optical matrix multiplication. In solving a class of combinatorial optimization problems, including Ising / max-cut problems, it achieved about two orders of magnitude acceleration over an NVIDIA A10 GPU Nature. In 2026, PACE 2 "Xizhi Tianshu" further expanded the matrix size to 128x128 and used more than 40,000 devices. These are genuine frontier results and the source of the company's technical credibility.

But optical computing has unavoidable physical weaknesses, and the Nature paper itself lists them clearly: there is no optical memory, so data must be converted repeatedly between digital and analog domains; at high bandwidth / high precision, the overhead of photoelectric conversion through ADC/DAC can exceed the energy consumption of optical computing itself; the analog domain is inherently imprecise, with phase noise and cascading losses causing data-dependent errors; and the chip is extremely temperature-sensitive, with a +/-5°C fluctuation degrading precision by about 1 effective bit Nature. These are the root causes of why "commercial optical computing" has been discussed for twenty years yet has struggled to leave the laboratory. They are also why Lightelligence itself, as well as Lightmatter, Ayar Labs, Celestial AI, and almost every other photonics player, has pragmatically shifted focus in recent years from "optical computing" to "optical interconnect."

Optical interconnect, the oNET route, is the business that actually generates cash today, contributing about 79% of 2025 revenue. The logic is straightforward: optical interconnect only needs to "transport" signals with low loss and low latency. It does not need to bear the precision cost of analog computing, so the engineering threshold is much lower and it can be sold first. Its flagship product "LightSphere X" is the first domestic solution to put "optical interconnect + optical switching" into a GPU supernode. Released in July 2025 by Shanghai INESA, Lightelligence, Biren Technology, and ZTE, it can tightly couple tens to thousands of GPUs into one supernode, reconfigure topology in real time, and scale elastically on demand. The company says it can improve model FLOPs utilization (MFU) by more than 50% GeekPark, IT Home. In March 2026, it launched a 128-card commercial version and said it had been adapted to mainstream large models including StepFun, DeepSeek, MiniMax, Kimi, and GLM.

3.2 Moat: Real but Thin, and Not Yet Validated by Cash Flow

Looking one by one at the sources of Lightelligence's moat, two and a half are truly defensible:

• Technology and talent (valid): an MIT-lineage photonics team + foundational optical-computing IP + Nature-level results. This is a real technical barrier and hard to replicate in the short term.

• Ecosystem position and switching costs (partly valid): Lightelligence deliberately positions itself as a neutral interconnect layer that is protocol-agnostic and compatible with GPUs from different vendors, solving the interconnection problem of domestic GPUs with "ten vendors and ten protocols." This differentiates it from NVIDIA's proprietary NVLink and has tied it to three generations of cooperation with domestic GPU vendors such as Biren Sina Tech. But whether this binding can become high switching costs depends on whether the supernode solution can truly ramp at scale.

• No dependence on advanced process nodes (half a point): silicon photonics chips can use mature processes, avoiding the supply-chain choke points below 7nm. This is a plus under U.S.-China controls. But it also means the technical threshold is not as high as advanced logic chips.

Its largest weakness is that the moat has not yet been validated by cash flow. The strength of a company's moat should be judged by whether it can maintain pricing power and returns under adverse conditions. Lightelligence's gross margin is falling, it is giving concessions to its largest customer, and free cash flow is negative. These all suggest that it is currently more like an early-stage player leading a niche segment but not yet converting that lead into profit, rather than a proven cash-flow machine.

3.3 Management and Governance: Strong Technical Credibility, Highly Concentrated Control

Yichen Shen's technical credibility is one of the company's core assets. Investors should note two points in governance.

First, control is highly concentrated through a special mechanism. Shen directly holds only about 6.43% of voting rights, but through concert-party agreements with several equity incentive platforms, he controls about 34.76% of voting rights in total and is the single largest shareholder Securities Times. A common rumor needs clarification: Lightelligence has not adopted a weighted voting rights (WVR / dual-class) structure. The prospectus contains no weighted voting or Class B shares. Control is achieved through "concert-party agreements + aggregation of shareholdings across multiple incentive platforms." This arrangement, in which control is exercised with economic ownership far below control rights, is a governance discount factor, but it is less severe than a statutory dual-class structure.

Second, a "change of control matter" appeared soon after listing and triggered a brief trading suspension, after which trading resumed NetEase. The details of such disclosures need continued tracking. In addition, total director remuneration in 2025 was RMB 107 million, including about RMB 41.88 million for Shen, most of which was share-based compensation. Share incentive expenses increased from RMB 55 million to RMB 205 million over three years. For a company with just over RMB 100 million in annual revenue, this is a meaningful cash / dilution burden.

IV. Industry and Market: The Direction Is Real, but the Prospectus May Have Pulled the Timetable Forward

4.1 Scale-Up and Scale-Out: Where Lightelligence Sits in AI Compute

To understand Lightelligence, start with two terms. Scale-out connects thousands or tens of thousands of GPUs across servers and racks into large clusters, usually over InfiniBand or Ethernet. Scale-up tightly couples multiple GPUs inside one supernode into a "logical single large chip," emphasizing ultra-low latency and shared-memory semantics. The bandwidth gap is an order of magnitude. On NVIDIA GB200, the port bandwidth ratio of NVLink (scale-up) to InfiniBand to Ethernet is about 36:4:1 Introl.

Lightelligence is focused precisely on "the internal scale-up segment that moves from copper to optics." It does not sell GPUs or general-purpose compute. It sells optical interconnect systems, modules, and optical switching chips that connect GPUs. In essence, it is a "picks-and-shovels" seller inside AI supernodes, monetizing the migration from copper to optics.

4.2 "Optics In, Copper Out" Is Real, but the Timetable Is the Key Dispute

The demand driver for optical interconnect is solid: LLM training / inference requires exponentially higher interconnect bandwidth, while copper has hard physical walls. When SerDes speed doubles from 112Gbps to 224Gbps, copper cable reach shrinks from about 2.5 meters to about 1 meter Synopsys. This is the physical basis of "optics in, copper out," and the direction is not in dispute.

But there is a timing dispute that the market underestimates. NVIDIA CEO Jensen Huang explicitly tempered short-term expectations at GTC. He emphasized that short-reach scale-up interconnects inside racks should "use copper where copper can be used, and use optics only when necessary," with copper inside the rack and optics across racks, because copper is passive, consumes zero power, and is low cost. NVIDIA will continue using copper inside racks through 2027-2028, and optics will only see large-scale in-rack adoption on the 2028 Feynman platform 36Kr. Third-party research firms such as LightCounting and Yole also generally place mass deployment of CPO, or co-packaged optics, in 2028-2030. This means Lightelligence's main "scale-up optical interconnect supernode" offering conflicts directly with the current engineering judgment of the industry leader, which believes copper should still be used inside racks for the next few years.

4.3 TAM: The RMB 5.7 Billion to RMB 180.5 Billion Story Needs a Discount

The entire valuation imagination for Lightelligence is anchored in a Frost & Sullivan forecast cited in the prospectus, not China Insights Consultancy as some reports mistakenly wrote: China's scale-up optical interconnect market will grow from RMB 5.7 billion in 2025 to RMB 180.5 billion in 2030, a five-year compound growth rate of about 99.6% Securities Times. This number requires a high degree of caution, for three reasons:

• The definition is narrowed three times: the 99.6% growth rate applies to a narrowed segment of "China x scale-up-specific x optical interconnect," and the 2025 base is only RMB 5.7 billion. Small bases are the easiest place to produce extremely high growth rates.

• Lack of independent third-party corroboration: when the definition is broadened globally, third-party growth rates are far more moderate. LightCounting's global AI cluster optical interconnect definition points to roughly 30%-35% compound growth in recent years and explicitly says it will fall back to 15%-20% after 2027. The broader global optical module market CAGR for 2024-2029 is also only about 22% LightCounting. Under any of these definitions, the rates are far below the prospectus's 99.6%. For now, the 99.6% can only be traced to "Frost & Sullivan + prospectus" as one original source, with multiple media outlets reposting the same source. It differs materially from the independent third-party figures because the segment is narrower and the base is lower, so it should not be placed directly alongside global definitions.

• The timing dispute compounds the issue: as discussed above, NVIDIA places large-scale in-rack optics in 2028, while this TAM curve requires most of the penetration to be completed from 2025 to 2030. The two timetables are in visible tension.

Founder Yichen Shen's more aggressive view is that "the share of optical chips in intelligent computing centers will rise from less than 1% to above 30% within five years" 36Kr. But this is a founder's forward-looking statement during an IPO window, with an interested position, and it contradicts the company's own prospectus data that China's optical-computing product market will be only RMB 1.46 billion in 2030, unless the "30%" refers to optical interconnect penetration rather than optical computing as a share of compute. I believe in the market direction, but I would apply a significant discount to the prospectus growth curve.

4.4 Cycle Positioning

Looking at the two businesses separately: optical interconnect, especially scale-up, is in the early stage of transition from introduction to growth. It has real orders, mass production, and fast growth, but the base is small, margins are falling, customers are concentrated, and the biggest "copper to optics" volume window, the 2028 CPO ramp, has not yet arrived. Optical computing remains in the introduction stage from proof of concept to commercial threshold, carrying the heaviest historical baggage. This is a market with the right direction, but a realization pace much slower than what the share price implies.

V. Horizontal Competition: An "Independent Third Party" Squeezed from Both Sides

5.1 One-Sentence Industry Map

Lightelligence is not "another NVIDIA." It positions itself as an independent supplier of "scale-up optical interconnect + optical switching" for domestic GPU supernodes, deliberately avoiding direct confrontation with NVIDIA in general compute. The awkwardness of this niche is that it is squeezed from both ends by leaders and giants.

5.2 Competitor Profiles

NVIDIA (NVDA.US) - ceiling and substitute. NVIDIA itself is pushing scale-up from copper toward optics. It is still mainly copper today, using copper backplanes inside NVL72 racks, but the 2028 Feynman generation will integrate CPO directly into NVLink switches HPCwire. Once NVIDIA's native optical interconnect matures, the incremental story of "optics replacing copper" inside the NVIDIA ecosystem will be absorbed. Lightelligence's real room for survival lies in the domestic heterogeneous GPU market that NVIDIA cannot enter or does not prioritize.

Broadcom (AVGO.US) and Marvell (MRVL.US) - system-level U.S. leaders. Broadcom's CPO switch, Tomahawk 6 / Bailly, has already entered volume shipments Broadcom IR. Marvell moved even more directly: in February 2026, it acquired Celestial AI for about US$3.25 billion, obtaining Photonic Fabric all-optical scale-up interconnect technology Marvell Newsroom. This means the U.S. camp has already filled in the "all-optical scale-up" gap, creating the most direct system-level opponents if Lightelligence wants to expand overseas.

U.S. silicon photonics startups - peers with capital firepower. Ayar Labs completed a US$500 million Series E in March 2026 at a US$3.75 billion valuation, with NVIDIA, AMD, and Intel all as strategic investors Optica. Lightmatter is valued at US$4.4 billion Lightmatter. These companies pursue chiplet-level in-package optical I/O, at a different layer from Lightelligence's system-level route, but the endgame is the same GPU interconnect profit pool. Notably, the most expensive private-market photonics unicorns are valued at only US$2.5 billion to US$4.4 billion, far below Lightelligence's post-listing market capitalization. This is an important external anchor for judging whether the public-private valuation gap is excessive.

Chinese photonics peers and optical module makers. Domestic optical-computing startups such as Guangbenwei and Qisuan Guangqi are mostly earlier-stage. Optical module leader Zhongji Innolight (300308.SHE) had RMB 25.0 billion of revenue and RMB 7.1 billion of net profit in the first three quarters of 2025, making it a profitable giant at scale Sina Finance. Eoptolink (300502.SHE), InnoLight affiliate Yuanjie Semiconductor (688498.SHG), and others are industry-chain collaborators. If these optical module leaders move upward into "optical interconnect systems / CPO full machines," they could squeeze Lightelligence's system integration margins from downstream.

5.3 The Truth Behind 88.3% Market Share: A Carefully Narrowed Denominator

Lightelligence's most frequently cited selling point is that "it has about 88.3% share in China's independent scale-up optical interconnect market." This number is literally valid, backed by the prospectus and Frost & Sullivan, but it is highly misleading unless all qualifiers are stated together:

• "China" means only mainland China;

• "scale-up optical interconnect" means only the niche of tightly coupled GPU interconnects inside supernodes;

• "independent / third-party" excludes self-use solutions such as Huawei's.

The key contrast is this: in the overall China scale-up optical interconnect market, Huawei has 98.4% share, mainly for self-use, while Lightelligence has only about 1.4% Sina Finance. The 88.3% measures share in "the small third-party market left after Huawei is excluded." Back-calculating from that, the third-party segment is less than RMB 100 million a year. High share does not mean a large business. More importantly, the biggest variable in this market-share narrative is that if Huawei turns to external supply, the small denominator that lets Lightelligence claim "number one" could collapse at any time.

VI. Current Fundamentals and Bull-Bear Debate: What the Market Is Trading Five Weeks After Listing

6.1 Post-Listing Share Price: Not a One-Way Rally, and Not Below IPO Price

Using the base date, the timeline is: April 28 first-day close at about 886 (+384%) -> wide-range volatility afterward -> June 3 jump of 16.83% to HK$708 on favorable news, including trading resumption after the control matter -> June 4 one-day pullback of about 19.6%, closing at about HK$569.5 (market data) Sina Finance. In other words, by the base date, it had neither continued a one-way rise nor fallen anywhere near the IPO price. The June 4 price was still about 3.1x the offer price of 183.2. All claims about "breaking issue price / halving / lock-up selling pressure" remain forward-looking risks, not realized facts.

6.2 What the Market Is Trading

The current share price is trading almost entirely on the scarcity narrative of being the "world's first AI optical computing stock" + supply-demand distortion from an extremely small float + continuation of IPO sentiment, rather than fundamentals. This is obvious from the valuation multiple in the next section. The gulf between real fundamentals, RMB 100 million revenue, cash burn, and falling gross margin, and the market narrative, hundreds of billions of Hong Kong dollars in market capitalization and an industry trend flagbearer, is the stock's defining feature.

6.3 Bull-Bear Debate

Bulls would say: Lightelligence is China's purest listed vehicle for the AI interconnect "copper to optics" trend. It has Nature-level technology, real deployment at thousands-of-GPU scale, top-tier institutional endorsement, and about five years of cash runway. Leadership in a narrow segment at 88.3% plus 99.6% market growth means volume could grow exponentially once adoption begins. Silicon photonics does not depend on advanced process nodes, making it a secure domestic-substitution asset.

Bears would say: revenue is RMB 100 million, losses are more than ten times revenue, the largest customer accounts for 40%, and demand depends on national-team projects. The overall segment in which it leads is dominated by Huawei at 98.4%. On technology route, optical computing has repeatedly missed commercialization timelines, while scale-up risks being bypassed as NVIDIA stays with copper until 2028. The most fatal issue is valuation: 464x P/S has prepaid all optimistic assumptions, while the October lock-up expiry and 18C precedent of post-listing drawdowns create much greater downside than upside.

The evidence for both sides is on the table, but the scale clearly tilts bearish, not because the company is bad, but because the price is too high.

VII. Valuation Analysis: 464x P/S, Isolated from Every Comparable

7.1 What the Current Valuation Means

As of June 4, 2026, Lightelligence's market capitalization was about HK$53.5 billion. Against 2025 revenue of RMB 106 million, or about HK$115 million, that implies a price-to-sales ratio of about 464x. At IPO pricing, the multiple was about 133x. On the first day intraday high, it reached about 730x. Even after the sharp June 4 drop, it remained as high as 464x.

7.2 Peer Comparison: No Comparable Is in the Same Order of Magnitude

The conclusion is straightforward: Lightelligence's P/S ratio is higher than the most emotionally priced A-share AI chip name Cambricon, higher than the most expensive private-market photonics unicorn, about 15x Zhongji Innolight, a profitable leader, and about 23x NVIDIA. It is not merely "expensive"; it is isolated. Even under highly optimistic forward revenue, such as assuming 2027 revenue reaches RMB 400 million, forward P/S would still be above 100x. That requires the market to give a Chinese hard-tech company a globally rare valuation multiple for a long and stable period.

7.3 Three Scenarios (Price Signals, Not Investment Advice)

Because the company has no profit and traditional PE/DCF is hard to apply, this section uses a "forward P/S x forward revenue" framework for three scenarios. Note that this is scenario analysis under the research framework, with a high density of assumptions. It is not a price forecast.

The current price of HK$569.5 is above the upper end of the base scenario (500) and approaching the lower end of the bull scenario (700). In other words, the market's current pricing is already close to the "bull scenario" and has largely bought in a base-to-bull future.

7.4 Margin-of-Safety Review (Independent Discipline)

Under the framework's margin-of-safety discipline:

• The current price is at a large premium to the bear-case implied value of 150-240. The margin of safety is zero, and in fact negative.

• The most fragile assumption across the three scenarios is that "the market will sustain a P/S ratio far above peers over the long term." If that assumption is cut by 30%, meaning the multiple moves back toward peers, the base-case valuation compresses from 350-500 to around 250.

• If revenue continues to roughly double over the next three years at the 66.9% growth rate but P/S reverts toward peers, even using Cambricon's 130x as a reference, the current price would still be overvalued.

• This is a classic "good company, bad price." It is worth waiting for a better price. Specifically, the ideal buy zone would need to return near IPO pricing, about HK$170-240, where the valuation multiple would be back in the "extremely expensive but not absurd" range.

• Margin-of-safety sufficiency conclusion: none.

VIII. Risk Analysis: Turning Risks into Verifiable Variables

Technology risk (medium-high probability / high impact): optical computing carries the heaviest historical baggage of "always five years away," while analog precision and ADC/DAC overhead are recognized hard constraints. There is also a real risk that NVIDIA's engineering route of "staying with copper until 2028" bypasses scale-up optical interconnect. Observable indicators: optical-computing revenue share, PACE 3 tape-out and commercialization progress, and the timing of copper-optics transition in NVIDIA / Broadcom CPO roadmaps.

Commercialization risk (medium-high probability / high impact): revenue base is tiny, the largest customer accounts for 40%, and demand heavily depends on tenders from national-team intelligent computing centers. Observable indicators: customer concentration, whether the company breaks beyond state-owned customers, and delivery / repeat orders for the two scale-up OCS orders totaling RMB 87.30 million in 2025 Q4 Zhidx, which are the hardest order evidence currently available.

Financial risk (high probability / medium impact): losses may accelerate faster than revenue growth; R&D equals 450% of revenue; and a roughly five-year cash runway means IPO proceeds could likely need refinancing after about 1.5-2 years as burn rises, making dilution hard to avoid. Observable indicators: quarterly cash burn rate, cash on hand, and whether the company initiates a placement / convertible bond.

Valuation and capital-market risk (high probability / high impact): 464x P/S plus a narrow 18C float, with free float only in the single-digit percentage range, amplifies volatility. Cornerstone shares unlock around October 2026, and pre-IPO shareholders unlock around April 2027, two clear supply-shock gates TMTPost. The pattern of 18C IPOs "peaking at listing and then halving" has appeared repeatedly: XtalPi fell about 74% from its peak, while Black Sesame International broke issue price by about 27% on the first day Jiemian.

Governance and external risk (medium probability / medium impact): the founder concentrates control through concert-party agreements, creating a governance discount; a "change of control matter" has already caused a suspension early after listing; U.S.-China export controls on photonics / AI chips are a double-edged sword, both supporting domestic substitution and creating risk of high-end equipment / tool-chain supply disruption; dependence on government subsidies creates policy-cycle exposure.

IX. Catalysts and Tracking Indicators

Positive catalysts: delivery and new signing of large scale-up OCS orders, successful tape-out and commercialization of PACE 3, overseas expansion or breakthrough beyond state-owned customers, inclusion in Stock Connect / major indices bringing incremental liquidity, and continuation of the AI capex cycle lifting optical interconnect demand.

Negative catalysts: selling pressure from the October 2026 cornerstone lock-up expiry / April 2027 pre-IPO lock-up expiry, first periodic report disclosing larger losses or customer loss, Huawei turning to external supply and hitting the third-party market, early maturation of native CPO from NVIDIA / Broadcom, refinancing dilution, and fading IPO sentiment.

Tracking dashboard investors can monitor continuously:

• Revenue and growth rate, and scale-up product revenue share, to verify commercialization ramp

• Adjusted operating loss, monthly cash burn rate, and cash on hand, to verify survival and profitability path

• Gross margin, to verify pricing power; falling below 35% would be a deterioration signal

• Largest customer / top-five customer share, to verify customer diversification

• Backlog / contract liabilities / receivables turnover, to verify revenue quality

• P/S premium relative to Cambricon and Zhongji Innolight, to verify whether valuation is reverting

• Trading volume, turnover, and cornerstone selling disclosures around unlock dates, about 2026-10 and 2027-04

X. Horizontal-Vertical Synthesis

10.1 What Lightelligence Has Proven Vertically

Viewed vertically, what Lightelligence has truly proven over the past eight years is the engineering and capital-markets capability to turn a frontier MIT paper into a hard-tech company with real products, real customers, and a successful public listing. It crossed the "valley of death" in photonic computing, pragmatically shifted from "optical computing" to "optical interconnect" to find a source of revenue, and during the rise of domestic compute, packaged itself as "China's first AI optical computing stock." These capabilities are real and still exist today.

What it has not proven is the ability to turn technology leadership into sustainable profit. Gross margin is falling, it is giving concessions to major customers, free cash flow is negative, and there is no clear profitability timetable. Its past success benefited substantially from the AI compute boom, domestic-substitution policy tailwinds, and private-market capital enthusiasm. Most of these factors remain in place today, but the capital enthusiasm has already been pushed to an extreme in the secondary market.

10.2 Bull and Bear Cases

Core bull reasons (3): 1) real and scarce photonics technology IP, with Nature-level results, PACE/PACE 2 chips, and LightSphere supernode deployment at thousands-of-GPU scale, unique in China; 2) positioned in the long-term AI interconnect "copper to optics" trend, protocol-agnostic, not dependent on advanced process nodes, and aligned with domestic substitution; 3) about five years of post-IPO cash runway and top-tier cornerstone investors, leaving near-term survival secure.

Core bear reasons (4): 1) absurd valuation, with 464x P/S isolated from all comparables and no margin of safety; 2) tiny and fragile business, with RMB 100 million revenue, 40% from one customer, demand dependent on national-team projects, and the overall segment dominated by Huawei at 98.4%; 3) technology / timing risk, with optical computing repeatedly delayed and scale-up potentially bypassed as NVIDIA stays with copper until 2028; 4) clear downside catalysts, including lock-up gates starting in October 2026 and the historical pattern of 18C IPOs halving after listing.

10.3 Pre-Mortem: If the Stock Loses 50% Three Years Later, What Would the Script Be?

Script one, most likely: lock-up stampede + valuation mean reversion. At the end of October 2026, about US$209.9 million of cornerstone shares held by 20 institutions unlock, while 18C free float is only in the single-digit percentages. If market sentiment fades by then, P/S may revert from several hundred times on the first day toward the center of gravity of peers that have broken issue price, such as XtalPi at about -74% from peak. Even a move back only to an "overvalued but survivable" 80-100x, based on 2025 revenue, would compress market capitalization by an order of magnitude and take the share price back near the IPO price of 183 or below. This is the script that requires the least "company deterioration" to occur.

Script two: the second growth curve collapses. If customers accept the mainstream route of "scale-up remains copper, scale-out uses native CPO from NVIDIA / Broadcom," Lightelligence's differentiated narrative is bypassed by the engineering route of the leaders. Combined with Huawei's 98.4% dominance in China's optical interconnect market, independent third-party room is squeezed from both sides. Revenue then fails to deliver the exponential growth needed to justify the valuation, while gross margin, already down from 61% to 39%, keeps falling because of concessions. That produces a double hit to earnings expectations and valuation.

Script three: the cash runway is shorter than loss acceleration. IPO net proceeds were about HK$2.377 billion, but 2025 net burn, using operating plus capex measures, was already in the hundreds of millions, and the company itself said 2026 losses would "increase significantly." If it is forced to issue shares after about 1.5-2 years around the time controlling shareholders unlock, and the share price has already fallen because of the first two scripts, financing would be completed at a lower price with larger dilution, forming a negative feedback loop of "share price falls -> financing becomes harder -> more dilution -> further decline."

10.4 Final Research Conclusion

Company profile score: fundamental quality Medium / growth High (tiny base) / moat Medium (not validated by cash flow) / financial resilience Weak / management credibility Medium-high / valuation attractiveness Low / risk level High / suitable investor type: high-risk growth and thematic investors; not suitable for value investors who care about margin of safety.

Investment rating: Avoid.

• One-sentence investment thesis: the technology and market direction are real, but 464x P/S has prepaid far-distant success, while lock-up expiries and fading sentiment create downside catalysts; the current price offers no margin of safety.

• Three price signals (endpoints from the scenarios in 7.3): ideal buy price approximately HK$150-240, the bear scenario and near IPO pricing; reasonable hold price approximately HK$350-500, the base scenario; clearly overvalued price >= HK$700, the lower end of the bull scenario.

• Current price classification: HK$569.5 is "above the upper end of the base scenario and approaching the bull scenario," classified as clearly overvalued.

• Whether it is worth waiting for a better price: yes. Buy conditions would be either a P/S ratio returning to the peer-comparable "extremely expensive but not absurd" range, corresponding to about HK$170-240, or a substantive commercial inflection point, including customer diversification, gross-margin stabilization, and cash-flow improvement. The opportunity cost of waiting is potentially missing the right side of the trend, but relative to the risk of a 50%+ drawdown, that opportunity cost is worth paying.

• Target holding period: if entered at a reasonable price in the future, it should be viewed over a 3-5 year industry-trend realization period.

• Expected annualized return if bought at the current price: under the bear / base / bull scenarios, the outcomes are roughly deeply negative / low single digits / high-beta positive return, respectively. The distribution is extremely right-skewed with a thick left tail.

• Maximum loss risk: referring to the pre-mortem, in the worst case of lock-up stampede plus valuation reversion, the drawdown from the current price could reach 60%-70%.

• Signals that would trigger reassessment: 1) share price falls below HK$240, bringing valuation into a discussable range; 2) gross margin rises and stabilizes above 40% for two consecutive quarters, and largest-customer share falls below 30%; 3) scaled commercial orders appear beyond state-owned customers; 4) Huawei clearly turns to external supply, a major negative; 5) evidence appears that optical interconnect products are being systemically replaced by native NVIDIA / Broadcom solutions.

Re-emphasis: this report is not investment advice. It is research analysis based on public information. This company is a newly listed, unprofitable, narrow-float, extremely volatile 18C Specialist Technology stock. Any investment decision should be based on one's own risk tolerance and checked against the primary prospectus and latest disclosures.

Research Uncertainties (Known Blind Spots)

• The primary prospectus PDF was not read word for word: the core financial figures in this report are mainly cross-verified through authoritative financial media quoting the prospectus, or the post-hearing information pack, with at least two sources. For precise decimal places and the exact definitional boundaries of market-size measures, the original HKEX prospectus should be the final check.

• The base-date share price is single-day market data: HK$569.5, the June 4, 2026 close, is internally consistent with HK$708 on June 3, with the decline matching. But 18C names are extremely volatile, and price timeliness is high, so valuation multiples will change quickly with price.

• Unnamed customers: the prospectus describes customers by category. The identities of the largest customer and the two customers behind the RMB 87.30 million orders are not disclosed. Market speculation, including ZTE / Huawei / Biren / MetaX-related guesses, has no primary-source confirmation and should not be treated as confirmed customers.

• Foundry / OSAT partners, specific shipments, and yield: except for about 10,200 OCS modules in backlog, there is no public data. "Transferred to domestic foundries" and "high yield" are only qualitative statements.

• TAM and the 99.6% CAGR forward assumption: this is a forecast from a single institution, Frost & Sullivan, cited in the issuer's prospectus. It may have self-interest bias, differs materially from third-party definitions, and the underlying assumptions for penetration rate / unit price could not be independently verified.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.