Shenzhen HQVT Technology Co., Ltd.(1392) · Intelligent Sensing

Shenzhen HQVT Technology In-Depth Research

Other languages
Quick ReadPlain-language overview · read this first

HQVT (01392.HK) enters use cases through multispectral perception hardware, which uses visible light, infrared, and other imaging bands at the same time, then layers on its self-developed large model services. The report rates it Watch. The company is expected to list only on June 22, 2026, so there is no secondary-market price yet; all price judgments are anchored to the offer price of HK$7.20. In 2025, large model services accounted for 53.1% of revenue, with a gross margin clearly higher than modules and perception terminals, making the structural upgrade real.

The fundamental picture is fast revenue growth, thin profit, and tight cash: in 2025, revenue rose 27.9% year on year, while net profit was only RMB 29.354 million and declined as listing and R&D expenses increased; operating cash flow showed a net outflow of RMB 129.9 million, diverging from net profit, while receivables, prepayments, inventory, and borrowings all rose. The report sums it up this way: the revenue story has already emerged, profitability is halfway there, and cash flow has not yet begun.

The company ranks first by revenue in China's multispectral AI large model services niche market, but its share of the overall multispectral AI industry is only 3.3%. Its lead is concentrated in the smallest and newest sub-segment, which is not yet enough to qualify as a moat; customer and supplier concentration is relatively high, and pricing power still needs to be tested after listing.

The offer price implies about 164 times 2025 earnings, already a growth-stock valuation. The report's conclusion on margin of safety is that there is none: the ideal buy zone is HK$4.0 to HK$4.8, the offer price sits outside the three-tier price signals, and the report argues it is worth waiting for a price pullback and post-listing cash flow verification. The biggest risk is the short commercialization history of large model services; any slowdown in acceptance would pressure revenue and gross margin at the same time. If working capital and borrowings continue to deteriorate, the fundraising will become money used to fill a gap.

Overall, the business upgrade is real, but cash flow and valuation have not yet offered enough probability of success; the company is worth continued monitoring, while the current offer price does not leave enough payoff potential. The above is a summary of the report's views and does not constitute investment advice. The stock market carries risk; invest with caution.

Lead

An intelligent-sensing supplier that uses multispectral sensing hardware to secure scenario entry points and lifts gross margin through large-model services, with its Hong Kong IPO launched in June 2026 at an offer price of HKD 7.20. In 2025, large-model services already contributed more than half of revenue and pushed gross margin to 22.3%, but operating cash outflow widened to RMB 129.9 million, while the IPO price implies about 7.2x sales and about 164x earnings. Research rating Watch: the business upgrade is real, the current price offers no margin of safety, and the ideal buy zone is HKD 4.0-4.8.

Full report

Metadata

  • Ticker: 01392.HK

  • Full company name: Shenzhen HQVT Technology Co., Ltd.

  • Current price and market capitalization: HKD 7.20 per share; about HKD 5.574 billion (as of 2026-06-11, based on the offer price/pre-listing basis, not a secondary-market closing price)

  • Currency: HKD; RMB-to-HKD conversions in this report use the 2026-06-11 central parity rate of 1 HKD = 0.8646 CNY, or 1 CNY ≈ 1.1566 HKD

  • Report date: 2026-06-11

  • Industry classification: 智能感知

  • One-sentence positioning: An intelligent-sensing supplier that enters through multispectral sensing hardware and raises gross margin through large-model services.

This is a pre-listing research report. The research base date coincides with the first day of the offering. The company has not yet listed, so valuation and price judgments are anchored to the offer price of HKD 7.20. The company is expected to list on the Main Board of the Hong Kong Stock Exchange on 2026-06-22, with a global offering of 85.1625 million shares and an implied post-issuance market capitalization of about HKD 5.574 billion at the offer price.

Research Summary

HQVT's real commercial engine is to use multispectral sensing hardware to secure scenario entry points, then layer on higher-margin large-model services. It is neither a traditional "camera company" nor a pure-software large-model company. In 2025, 53.1% of the company's revenue already came from multispectral AI large-model services, and this segment's gross margin reached 30.4%, clearly above the 11.2% for modules and 17.7% for sensing terminals. This means HQVT is no longer merely selling components. It is trying to remake itself into a hybrid intelligent-sensing supplier with "hardware as the foothold and software as the profit pool."

The IPO market's current narrative is also clear and can almost be compressed into three words: multispectral, AI, scarcity. The prospectus cites Frost & Sullivan data showing that by 2025 revenue, the company ranked first among multispectral AI enterprises in China with a 3.3% market share, and ranked first in China's multispectral AI large-model service market with a 23.0% share. Media coverage has already packaged it directly as the "first multispectral AI stock." The appeal of this narrative is that it fits both Hong Kong investors' preference for "scarce AI IPOs" and capital markets' imagination around the commercialization of vertical-scenario large models. The issue is also here: the story is very new, but the verification period is very short.

The company has no listed share-price history, so the "reasons for past rises and falls" cannot be found in a price chart. They can only be found in the uplift of primary-market valuation. In the Series C financing in October 2024, based on the disclosed subscription consideration and new shares, the post-money valuation was about RMB 1.0 billion. The Series C+ financing at the end of 2024 was still roughly at this level. By the Zhide No. 9 capital increase in July 2025, the post-money valuation had jumped to about RMB 3.55 billion. In this IPO, the offer price implies a post-issuance market capitalization of about RMB 4.82 billion. In roughly nine months, the primary-market valuation rose by nearly 4.8x. What drove this valuation jump was the revenue breakout in 2024-2025, the launch of large-model services, and the formation of the "multispectral AI leader" narrative, rather than stable cash flow.

The real bull-bear debate is much more specific than whether the industry is attractive. It comes down to two points. First, whether large-model services are sustainable, high-margin, repeatable business. This business started scaling only in 2024 and quickly exceeded half of revenue in 2025, lifting the company's gross margin from 18.8% to 22.3%. Yet over the same period, operating cash flow deteriorated from a net outflow of RMB 6.491 million in 2024 to a net outflow of RMB 129.9 million in 2025, while receivables, prepayments, and inventory expanded sharply. Second, what the company's number-one ranking actually means. Its share of the overall Chinese multispectral AI market is only 3.3%, showing that the industry remains fragmented. Its 23.0% share in the large-model service submarket shows that it is stronger in the most capital-favored niche, but that market was only about RMB 1.545 billion in 2025. In substance, it is still a small and fast new market.

From a fundamental perspective, HQVT is at a very typical and dangerous stage: the revenue story has emerged, the profit story is only halfway proven, and the cash-flow story has not yet begun. 2025 revenue was RMB 668.5 million, up another 27.9% from 2024. Net profit was RMB 29.354 million, which appears to mark two consecutive profitable years. But 2025 profit declined year-on-year because listing expenses and R&D spending increased, while operating cash flow turned more deeply negative and borrowings rose quickly. As of April 30, 2026, interest-bearing debt and lease liabilities totaled RMB 324.1 million, further higher than at the end of 2025. Although the prospectus says the company has not experienced any material adverse change and that revenue in the first four months of 2026 increased versus the same period last year, it does not provide the exact growth rate.

If I had to summarize the company in one sentence, I would classify it as a "transforming company undergoing valuation reset," not a "high-quality compound grower." The reason is simple. It has indeed shifted from low-margin module sales to a three-layer structure of modules, terminals, and large-model services. It has also proved that it can generate commercial revenue in a very new submarket. But it has not yet proved that it can steadily convert that growth into cash, maintain bargaining power before customer and supply-chain concentration falls, or sustain its valuation after listing in an environment of higher transparency and lower IPO scarcity premium.

Company Longitudinal Development History

HQVT's starting point is clear. Its predecessor, Shenzhen Haiqing Video Technology Co., Ltd., was established in China on April 3, 2013. On November 8, 2022, it was converted as a whole into a joint-stock company and renamed Shenzhen HQVT Technology Co., Ltd. This timeline matters. It shows that the company's real long-term accumulation came from more than a decade of technology entrepreneurship, not a temporary assembly during one AI boom. It also shows that its capital-market transformation came late: there were only a little over three years between joint-stock reform and IPO, and pre-listing governance, financing, and customer structure are still rapidly taking shape.

In its early years, the company looked more like a supplier combining sensing hardware and visual algorithms. The prospectus describes its current capabilities as a full-chain "optics-sensor-imaging-computing" architecture, with today's three main lines being multispectral AI modules, multispectral AI sensing terminals, and multispectral AI large-model services. The real turning point, however, came after 2023, not in 2013. The reason is that in 2023 the company first went through a clear dip. In the first application version, management explicitly explained that 2023 revenue fell 47.9% from 2022, mainly due to active adjustment of the customer mix, customers digesting inventory in 2022, fading pandemic-related demand, and a weakening macro economy. Many companies would package such a year as a "strategic upgrade." HQVT's documents are relatively direct: it was a painful period when old customers and old demand faded before new business and new customers could fill the gap.

The real reversal happened in 2024. The prospectus shows that revenue surged from RMB 117.1 million in 2023 to RMB 522.6 million in 2024, an increase of 346.4%. Net profit turned from a loss of RMB 18.413 million to a profit of RMB 40.412 million. Three factors clearly drove this breakout: strong demand in safety-related scenarios, new customers that contributed large orders, and the commercialization of multispectral AI large-model services. In other words, the customer and scenario switch the company made in 2023 finally showed up in the financial statements in 2024.

But 2025 added another layer of complexity to the story. Revenue continued to grow to RMB 668.5 million, and gross margin rose from 18.8% to 22.3%, better on the surface than 2024. Yet net profit fell back to RMB 29.354 million. The reason is direct: listing expenses, R&D expenses, and administrative expenses rose. More importantly, in 2025 large-model services rose to 53.1% of revenue, becoming the company's largest business and largest source of gross profit. The company proved with facts that the business-structure upgrade is real, but it also exposed a problem: once R&D, marketing, and compliance expenses scale up, profit leverage can be quickly consumed.

The listing path itself also shows how capital markets understand the company. On August 6, 2025, the company first submitted an application to the Hong Kong Stock Exchange. The first version's financial base mainly covered 2022-2024 and unaudited first-quarter 2025 data. On March 16, 2026, it resubmitted the application, with disclosure updated to complete audited 2023-2025 data. By the formal offering on June 11, 2026, the market was seeing an IPO candidate with "revenue already scaled up, profit just turned positive, and attractive ranking in a niche track," rather than a small manufacturing company still in validation. The documents do not separately state the "reason for resubmission," but the version differences suggest that the most important change was completion of full FY2025 data and a more concrete presentation that large-model services had become the core revenue source.

Looking one layer deeper into the capital actions reveals the market temperature more clearly. After the Series C financing in October 2024, the company's post-money valuation was about RMB 1.0 billion based on disclosed consideration. The Series C+ in December 2024 remained roughly in that range. When Zhide No. 9 invested in July 2025, the post-money valuation had risen to about RMB 3.55 billion. This IPO further raises it to about RMB 4.82 billion. The primary market repeatedly marked up the company's valuation over a very short period, reflecting a rapid reassessment of expectations for "hardware entry point + vertical large model," not the disappearance of operating risk.

The best way to summarize HQVT's story over the past decade-plus is in four phases. The first phase was the technology accumulation period from 2013 to 2022, when the company started from vision and sensing hardware, built platform capabilities, and completed joint-stock reform. The second phase was the trough and restructuring period in 2023, when old demand receded, new customers were being switched in, and revenue contracted sharply. The third phase was the structural upgrade period in 2024-2025, when large-model services launched, revenue broke out, profit turned positive, but cash flow deteriorated. The fourth phase is now: the 2026 period of front-loaded capital expenditure and IPO financing, in which the company plans to use capital-market funding to push R&D, capacity, and globalization one step forward. Whether it becomes a high-quality growth stock or remains in the intermediate state of "an appealing story with fragile cash flow" will become clear over the next two years.

Business Model and Financial Quality

HQVT's current business model is already a three-layer stack, not a single-product sale. The bottom layer is multispectral AI modules, sold to system integrators and equipment makers, functioning like a "sensing base." The middle layer is multispectral AI sensing terminals, integrating sensors, algorithms, and standard hardware into finished or semi-finished products. The top layer is multispectral AI large-model services, connecting its self-developed "Origin of Intelligent Vision Large Model" to specific scenarios, including safety, IDC, and intelligent-driving R&D vertical solutions. In 2025, the three accounted for 31.3%, 13.9%, and 53.1% of revenue, with gross margins of 11.2%, 17.7%, and 30.4%, respectively. The most important fact for the company is this: large-model services have become the core business determining the revenue mix and gross-margin center, not an ornament.

Business 2025 Revenue Share 2025 Gross Margin My View
Multispectral AI modules 31.3% 11.2% Volume entry point with thin profit
Multispectral AI sensing terminals 13.9% 17.7% Scenario extension, highly affected by project timing
Multispectral AI large-model services 53.1% 30.4% Valuation core and the key point to verify

The table above is compiled from the prospectus's 2025 segment revenue and gross-margin disclosure.

The benefit of this model is that the company's growth logic does not have to depend entirely on hardware ASP. Gross margin improved in 2024 because large-model services began contributing. Gross margin continued rising to 22.3% in 2025 because this high-margin business rose to more than half of revenue. The downside is equally clear: once acceptance timing for large-model services slows, or customers shift it from "procurement projects" to "joint-development projects," the profit center will immediately fall, because modules and terminals alone are not enough to support a high valuation.

The company's customer interface is mainly held by system integrators, not by end brands or a large base of dispersed users. The prospectus discloses that in 2025, the top five customers accounted for 46.8% of total revenue, with Customer K alone contributing 14.1% and Customer F contributing 10.7%. In 2024, Customer F's contribution once reached 35.5%. This kind of concentration is not uncommon in project-based and integration-based industries, but it means HQVT's reliance on downstream integrators is meaningful, and customer switching plus acceptance timing will strongly affect revenue recognition. The prospectus also discloses that the vast majority of 2025 revenue came from private-sector related customers, with direct government procurement relatively limited. This suggests it is not a typical company that "feeds on fiscal projects," but it also means its bargaining power among commercial customers still needs further validation.

Financial quality is the part of this company that most needs to be examined under a microscope. In 2023, 2024, and 2025, net profit was RMB -18.413 million, RMB 40.412 million, and RMB 29.354 million, respectively. Operating cash flow was RMB 69.571 million, RMB -6.491 million, and RMB -129.9 million, respectively. In other words, cumulative net profit over the past three years was positive, but cumulative operating cash flow was negative. Looking only at the past two years, the divergence between operating cash flow and net profit is even clearer. Management's main explanation for the deterioration in 2025 operating cash flow was increases in prepayments, inventory, and receivables, with the large increase in prepayments used to lock in equipment such as high-performance computing servers to support delivery of large-model services and sensing terminals. This explanation makes some sense, but it still means the cash content of earnings is not high at this stage.

As of April 30, 2026, this pressure had not eased materially. Inventory rose to RMB 270.8 million, trade and bills receivables rose to RMB 257.6 million, prepayments and other receivables remained at RMB 150.7 million, current borrowings increased to RMB 245.7 million, and contract liabilities also rose to RMB 262.1 million. At the same time, the company disclosed that debt and lease liabilities under total-liabilities disclosure totaled RMB 324.1 million, and after 30 April 2026 it obtained another roughly RMB 15 million in bank loans. The statements do not look like a liquidity crisis, and management also states that working capital will be sufficient for the next twelve months. But the simultaneous presence of "high growth + high working-capital occupation + rising borrowings" is itself a combination that requires close tracking.

R&D is another area where the company must keep spending. In 2025, R&D expenses were RMB 105.3 million on an "including capitalization" basis, equal to 15.7% of revenue. Of this, 57.2% came from outsourcing to third parties and 42.8% from internal R&D. By the end of 2025, the company had 156 R&D employees, about 43.1% of total staff. This shows HQVT is not a sales-team-driven channel company. To sustain competitiveness, it must keep investing in model training, chip design, algorithm iteration, and scenario adaptation. For investors, this is both a barrier and an expense ceiling. If R&D stops, the story collapses. If R&D continues, margins will find it hard to expand sharply and quickly.

At the governance level, control is relatively clear, but the company also retains typical traces of a growing private enterprise. After listing, the founder surnamed Zhou will directly and through controlled platforms control about 43.5% of the shares, while Chen Yonggang personally holds about 3.07%. The prospectus also discloses that a substantial portion of past bank loans was backed by joint and several guarantees provided by Mr. Zhou. As of the latest practicable date, the relevant guaranteed loans amounted to about RMB 181.7 million and are expected to be released after listing. This arrangement, where the founder's personal credit supports corporate financing, is common among Chinese private enterprises, but it also shows that the company's pre-listing bank credit was built more on the founder than on independent financing capability. Another detail is that this IPO has no cornerstone investors, new offer shares account for only 11% of total share capital, and use of proceeds is planned as 50% for R&D, 25% for capacity expansion, 15% for global sales expansion, and 10% for working capital. The use of funds is well targeted, but it also shows the company faces a long capital-expenditure list.

Industry and Horizontal Peer Analysis

The hardest part of researching HQVT is that there is no fully symmetric peer set. It spans four layers: infrared/multispectral sensing, machine-vision hardware, vertical AI solutions, and large-model services. Comparing it directly with pure hardware companies would understate its software narrative. Comparing it directly with pure large-model companies would overstate its platform nature. The most appropriate view is to treat it as a "scenario-based multispectral AI supplier" and then find references from the two dimensions of sensing hardware and AI services.

Start with industry position. Based on the prospectus, China's multispectral AI market grew from RMB 6.3 billion in 2020 to RMB 20.0 billion in 2025 and is expected to reach RMB 79.4 billion in 2030. Within that, the multispectral AI large-model service market was about RMB 1.545 billion in 2025 and is expected to reach about RMB 8.4 billion in 2030. HQVT ranked first among Chinese multispectral AI enterprises by revenue in 2025, but its share of the overall market was only 3.3%. It ranked fourth in the multispectral AI module market with a 3.8% share and fourth in the multispectral AI sensing terminal market with a 0.7% share. Only in the multispectral AI large-model service market did it rank first, with a 23.0% share. This structure is crucial. It shows the company's advantage is concentrated in the smallest, newest, fastest-growing submarket, not in dominance of the entire industry.

Among international references, Teledyne carries the most weight. Teledyne's digital imaging business covers visible light, X-ray, long-wave/mid-wave/short-wave infrared, and other sensing, camera cores, and system products. In essence, it is a more mature, more industrialized, more global multispectral imaging platform. Its 2025 annual report clearly states that it provides a range of cooled and uncooled infrared/thermal imaging products. As of June 2026, Teledyne's market capitalization was about USD 28.1 billion, with a price-to-sales ratio of about 4.7x and a price-to-earnings ratio of about 30x. Compared with Teledyne, HQVT is much smaller, more concentrated in applications, and more growth-story-oriented in capital markets, while its earnings stability and cash-generation capability are clearly in a different league.

In China's hardware chain, Raytron Technology and Guide Infrared are more solid references. Raytron's official website emphasizes that it can independently develop CMOS readout circuits, MEMS infrared sensor wafers, infrared detectors, thermal imaging modules, and complete products. As of June 2026, its price-to-sales ratio was about 7.87x and its price-to-earnings ratio about 38.38x. Guide Infrared's official materials show that it is one of China's earlier companies focused on infrared thermal imaging detectors, modules, and integrated optoelectronic systems. As of June 2026, its market capitalization was about RMB 51.97 billion and its price-to-earnings ratio about 53.66x. Compared with these two, HQVT's advantage is stronger software and scenario integration, allowing it to tell a "large-model services" story. Its weakness is that its moat at the underlying device level has not been as fully verified as Raytron's, and its scale effects are far weaker than Guide Infrared's.

On the AI services side, SenseTime is the closest "valuation-sentiment reference" in Hong Kong stocks. SenseTime officially disclosed that 2025 revenue exceeded RMB 5.0 billion and that EBITDA turned positive in the second half. Yahoo Finance pages showed its price-to-sales ratio was about 9.84x in early June 2026. HQVT and SenseTime are of course not on the same scale, but both benefit from the "computer vision/multimodal/generative AI" narrative. The difference is that SenseTime follows the route of an underlying model platform plus general AI ecosystem, while HQVT is more like a small, specialized player led by specific sensing scenarios. SenseTime's valuation can tell you how much the market is willing to pay for an AI narrative, but it cannot prove that HQVT should enjoy the same premium.

Company Main Positioning Valuation Signal from the Market in Mid-2026
HQVT Multispectral sensing hardware + vertical large-model services Offer price implies about 7.2x 2025 sales and about 164x 2025 earnings
Teledyne Global multispectral imaging and sensing platform About 4.7x sales and about 30x earnings
Raytron Technology Infrared detectors and thermal-imaging underlying device chain About 7.9x sales and about 38x earnings
Guide Infrared Infrared thermal imaging and integrated optoelectronic systems About 53.7x earnings
SenseTime Vision AI / generative AI platform About 9.8x sales, with earnings still weak

HQVT's valuation in the table is calculated in this report based on the prospectus and exchange rate. The other companies use public valuation information from official IR, Reuters, Yahoo Finance, Google Finance, and similar sources.

After horizontal comparison, HQVT's ecological position becomes clear. It is more like a "niche full-stack player": using modules to occupy positions, terminals to expand projects, and large-model services to capture higher gross margin and higher narrative optionality. It is neither a device leader nor a general model platform or pure system integrator. The advantage of this position is that it can quickly assemble solutions in new scenarios. The disadvantage is that it can be squeezed simultaneously by stronger upstream and downstream companies. If stronger infrared/sensing device leaders move downstream into systems, HQVT will come under pressure. If larger AI platform companies move downstream into vertical-industry solutions, HQVT will also come under pressure. What it really needs to defend is productization speed and delivery depth in several key safety scenarios, not an abstract notion of "technological leadership."

Current Fundamentals and Bull-Bear Debate

From the "current point in time" visible in the prospectus, the company remains on a growth track. Management explicitly discloses that revenue in the first four months of 2026 increased versus the same period in 2025, mainly because multispectral AI large-model services recognized more revenue in the first four months of 2026, with higher demand and acceptance work than in the prior-year period. The directors also confirm that since the end of 2025, the company's sales performance, trading conditions, and prospects have not experienced any material adverse change. For a company currently conducting an IPO, this is an adequate recent update, but the information density is limited because it does not disclose exact year-on-year growth or profit data.

What the market is currently trading is really whether it can become the first listed template to commercialize a closed loop in the multispectral AI track, not how much money it can earn this year. The evidence supporting this narrative has three layers. First, the company's share of the multispectral AI large-model service market reached 23.0% in 2025, ranking first in its submarket. Second, large-model service revenue already accounts for more than half of total revenue and has lifted overall gross margin. Third, the company plans to allocate 50% of IPO proceeds to R&D, including 25.3% for large-model capabilities, 8.2% for MEMS chip R&D, and 5.9% for next-generation intelligent-sensing products. This shows management intends to deepen the "hardware entry point + model service" model rather than dress up profits before listing.

The real bull case is concentrated around these points. First, the business structure has changed substantially and is no longer walking on one low-margin module leg. Second, the company has already secured the number-one position in China's multispectral AI large-model service market, a market that is new enough, and the prospectus expects this market to maintain a 41.1% CAGR from 2025 to 2030. Third, the 2023 revenue trough shows the company reaccelerated after switching customers and scenarios, not through a simple linear extrapolation. Such an experience usually carries more information than a company that only grows in an upcycle. Fourth, revenue in the first four months of 2026 continued to increase year-on-year, indicating that 2025's high growth did not suddenly brake before the IPO.

The bear-side evidence is equally solid and closer to the financial statements. First, profit is thin. The 2025 net margin was only 4.4%, while administrative and R&D expenses are rising, showing that the company is still far from the high-profit model of a mature software company. Second, cash-flow quality is poor. 2025 operating cash outflow was RMB 129.9 million, far worse than net profit of RMB 29.354 million. Third, customer and supplier concentration remains high. The top five customers and top five suppliers accounted for 46.8% and 62.2% in 2025, respectively, and to lock in resources such as HPC servers, the company made a one-off large prepayment of about RMB 146.9 million to four suppliers. Fourth, financing reliance is rising, with total debt and lease liabilities reaching RMB 324.1 million by the end of April 2026. Fifth, the IPO valuation is not cheap, and there is no secondary-market history. In other words, investors are buying a new stock that has already priced in a substantial amount of optimism, not an old stock mispriced by the market.

I think the market's easiest mistake is to equate "number one in a submarket" directly with "a moat has already been built." HQVT's share of the overall Chinese multispectral AI industry is only 3.3%, showing that no absolute leader has emerged in this industry. It looks stronger because it moved fastest in the new subsegment of large-model services. First-mover advantage in a new market can bring a revenue breakout, but it may also be only a timing lead created by concentrated acceptance of early projects. To disprove this risk, two things must be seen after listing: first, the revenue share and gross margin of large-model services can hold steady; second, receivables, prepayments, and inventory as a percentage of revenue begin to fall. Without these two signals, the story is only half proven.

Valuation, Risks, Catalysts, and Horizontal-Vertical Summary

Start with the hardest valuation anchor. Based on the offer price of HKD 7.20 and 774.2 million post-issuance shares, the company has a pre-listing-basis market capitalization of about HKD 5.574 billion. Converting 2025 revenue of RMB 668.5 million and net profit of RMB 29.354 million into HKD at the 2026-06-11 exchange rate gives 2025 revenue of about HKD 773 million and net profit of about HKD 33.95 million. Therefore, HQVT currently trades at about 7.2x 2025 sales and about 164x 2025 earnings. Even using the company's adjusted net profit of RMB 55.245 million after excluding listing expenses and share-based payments, the adjusted P/E is still about 87x. The prospectus gives unaudited pro forma adjusted net tangible assets per share after issuance of about HKD 1.09, implying a P/B ratio of about 6.6x. The market is valuing it as a "growing AI/sensing platform stock." This pricing is neither a cheap IPO nor a manufacturing-company multiple.

Next, look at valuation history. The company has no secondary-market history, so only equity-valuation migration can be observed. Based on rough calculations from the financing rounds disclosed in the prospectus, the Series C post-money valuation in October 2024 was about RMB 1.0 billion, the Series C+ in December 2024 remained near about RMB 1.01 billion, Zhide No. 9's investment in July 2025 implied a post-money valuation of about RMB 3.55 billion, and this IPO implies about RMB 4.82 billion. Valuation rose much faster than profit and was driven almost entirely by the shift in revenue structure and track imagination. For investors, this does not necessarily mean the price is wrong, but it at least means the "margin of safety" cannot be built on a historically cheap valuation, because there is no cheap history here.

After looking through cash flow, the problem becomes sharper. Over the past three years, cumulative operating cash flow was negative while cumulative net profit was positive. In 2025 alone, operating cash outflow reached RMB 129.9 million, far below net profit. The prospectus does not split maintenance capex from expansion capex, so owner earnings cannot be calculated precisely. But from the large 2025 prepayments for HPC servers and the large fundraising plans for capacity expansion and model R&D, recent capital investment is clearly expansion-oriented. For this reason, assigning a high valuation directly to accounting profit would create a large margin of error. For this company, price-to-sales and scenario penetration currently explain value better than headline P/E.

The three valuation scenarios below mainly use "2026 revenue and the price-to-sales ratio the market is willing to assign," with improving profitability as a check. They put the current price back into a more disciplined framework and do not constitute investment advice.

Dimension Bear Base Bull
2026 revenue / margin assumptions Revenue up 10%–16% YoY, large-model service share struggles to rise further, net margin about 3%–4% Revenue up 24%–32% YoY, large-model services maintain roughly half of revenue, net margin about 5%–6% Revenue up 38%–48% YoY, large-model services continue scaling, net margin about 7%–8%
Cash-flow assumption Operating cash flow remains weak or only near breakeven Operating cash flow improves but does not fully match profit Operating cash flow turns positive, working-capital occupation peaks
Valuation multiple assumption 3.7–4.2x PS 4.7–5.2x PS 5.8–6.2x PS
Key catalyst Large-model services maintain growth without an obvious repeat-purchase flywheel 2026 revenue continues growing and receivables turnover improves Scenario replication, overseas customer expansion, model-service validation
Key risk Customer concentration and price competition compress gross margin High R&D investment continues to consume profit leverage Competitors enter, model commercialization falls short
Implied return range -44% to -33% -19% to -6% +11% to +28%
Permanent capital-loss risk If large-model service share falls and receivables plus prepayments keep rising, valuation can quickly compress back to hardware-stock territory If cash-flow improvement does not materialize, the base valuation will also move lower If high-growth expectations miss, the bull multiple will be hard to sustain

The ranges above are based on 2025 revenue, the business structure disclosed in the prospectus, and three scenario assumptions using public valuations of comparable companies such as Teledyne, Raytron Technology, and SenseTime.

The margin-of-safety review leads to a clear conclusion. First, the current HKD 7.20 price is materially above the conservative-scenario value, so the margin of safety is zero. Second, the most fragile variable across the three scenarios is acceptance intensity and revenue share for large-model services, not total revenue. If this assumption is cut by 30%, the base-scenario per-share value moves closer to HKD 5 rather than a little above HKD 6. Third, if profit does not grow over the next three years, the earnings yield based on 2025 net profit is only about 0.6%, and even based on adjusted net profit it is only about 1.1%, both meaningfully below the Hong Kong 10-year government bond yield of about 3.279%. This means that at the current offer price, this is a "possibly good company without a good price." My margin-of-safety conclusion is: none.

For risks that could truly create permanent capital loss, I would pick only five. First, business risk. Large-model services are the high-margin center, but their commercialization history is short. If acceptance timing in several core scenarios is delayed, the company's gross margin and revenue recognition will immediately come under pressure. Probability: medium; impact: high. Second, financial risk. Operating cash flow turned sharply negative in 2025, and borrowings continued rising at the end of April 2026. If working-capital occupation keeps increasing, IPO proceeds will quickly become money used to "fill balance-sheet gaps" rather than money used to "buy future growth." Probability: medium-high; impact: high. Third, concentration risk. A single major customer accounted for 35.5% in 2024, while the top five customers and suppliers accounted for 46.8% and 62.2% in 2025. Any key customer shifting orders or key supplier tightening resources would transmit through revenue, gross margin, and cash flow. Probability: medium; impact: high. Fourth, valuation risk. The company's current pricing is essentially built on the AI narrative and growth expectations. If the post-listing market style shifts toward low valuations or cash-flow preference, valuation compression can occur quickly. Probability: high; impact: medium-high. Fifth, geopolitical and supply-chain risk. The company discloses that its products are usually not directly subject to the EAR, but its cloud computing services involve servers containing U.S.-origin integrated circuits, so the export-control environment remains a long-term background variable. Probability: medium; impact: medium.

Catalysts are concentrated in variables that can directly change market expectations. Positive catalysts include: the first periodic report after listing proves that large-model services remain the main growth source; operating cash flow turns positive or at least improves significantly; the top-five customer share falls and the customer structure diversifies; overseas certification, overseas sales, and new-scenario implementation begin to materialize. Negative catalysts include: receivables and prepayments continue to deteriorate; borrowings continue to increase; large-model service revenue share falls; gross margin drops below 20%; and if primary-market funding enthusiasm fades after listing, insufficient trading and liquidity would also amplify share-price volatility.

Tracking Indicator Normal Range Warning Threshold
Large-model service revenue share >50% <40%
Comprehensive gross margin >22% <20% for two consecutive periods
Operating cash flow / net profit >0.8x <0 for two consecutive periods
Receivable turnover days <80 days >100 days
Prepayments + inventory / annual revenue <35% >50%
Top-five customer revenue share <45% >55%
Liability ratio and leverage Gearing ratio <0.6 >0.8
Post-2026 new financing Mainly internally generated cash flow Refinancing or large new debt

These thresholds are tracking disciplines I set based on the company's current statement structure and are not from the prospectus. What really matters is the direction, not the threshold itself: HQVT must prove that it can turn growth from "project-recognition growth" into "cash-realizable growth."

Putting the horizontal and vertical views together, HQVT has truly proved two capabilities along the way. First, it can turn multispectral sensing technology into real revenue, so it is not a laboratory company with no customers or scenarios. Second, it can push itself a step upward from hardware supplier toward higher-margin service provider when the industry is just beginning to scale. But it has not yet proved the third and most important point: whether it can truly hold this position. Past success came partly from management making the right customer and scenario switches in 2023-2024, and partly from the era dividend of the industry and AI narrative. Today, the era dividend is still present, but the quality of delivery is not enough. The market's most likely mistake is to misread "growth exists" as "the business model is mature," not to doubt whether the company has growth. I do not doubt this company deserves continuous tracking. I doubt whether HKD 7.20 gives IPO investors enough odds.

I would compress the bull case into four points. First, multispectral AI large-model services became the largest source of revenue and gross profit in 2025, so the business-structure upgrade is real. Second, the company has a 23.0% share in China's multispectral AI large-model service market and ranks near the front of the niche track. Third, revenue continued growing year-on-year in the first four months of 2026, at least showing that momentum did not go out before the IPO. Fourth, IPO proceeds are directed at the company's current gaps, with clear uses in R&D, capacity, and sales expansion.

I would also compress the bear case into four points. First, profit is thin and cash flow is weak, with serious divergence between 2025 operating cash flow and net profit. Second, customer and supplier concentration remain high, and key resources require large prepayments to lock in. Third, the current IPO valuation is already priced as a growth stock rather than a manufacturing stock, while the company has not yet produced matching cash quality. Fourth, the overall industry remains fragmented, and the company's share of the overall multispectral AI market is only 3.3%; its leadership is more apparent in a new submarket than in the entire industry.

If this investment loses 50% three years from now, two scripts are most likely. The first script occurs in 2027: several larger infrared/AI sensing companies begin packaging multispectral capabilities into more complete system solutions, downstream integrators shift toward larger suppliers, HQVT is forced to cut prices to win orders, comprehensive gross margin falls from around 22% to 15%–16%, and the market compresses its valuation from 7x sales to around 3.5x. The second script occurs in 2027-2028: large-model services continue growing, but acceptance is slow, cash collection is even slower, receivables and prepayments keep rolling up, operating cash flow remains negative for consecutive periods, and borrowings are forced to keep expanding or even lead to refinancing. The market realizes it is more like a project-based company with heavy capital occupation than an asset-light AI platform, and the valuation center moves lower. Both scripts rely only on the continued deterioration of fragilities already visible in the current statements and require no black-swan event.

The final conclusion can be direct. HQVT is a company with real business, real growth, and real technical-route value. Its most attractive feature is that it has already lifted itself out of the low-margin hardware-company category through large-model services. But the same point is what makes me hesitate most: the new high-margin engine has appeared too quickly, while cash-flow verification has arrived too slowly. At the current offer price, the market has already paid a meaningful prepayment for "first in a niche track, scarce AI IPO, successful structural upgrade," but it has not yet received the receipt for "cash flow moving in the same direction, customer diversification, and a stable competitive position." For this type of company, the biggest danger is often odds being consumed too early, not the disappearance of growth.

【Company Profile Scores】

  • Fundamental quality: Medium

  • Growth: High

  • Moat: Medium

  • Financial robustness: Weak

  • Management credibility: Medium

  • Valuation attractiveness: Low

  • Risk level: High

  • Suitable investor type: High-risk speculation / event-driven; unsuitable for investors treating it as a low-risk long-term core holding

【Investment Rating】

  • Rating: Watch

  • One-sentence investment thesis: The business upgrade is real, but cash flow and valuation do not yet offer sufficient odds.

  • Three-tier price signals: 【Ideal/Fair Buy Price】4.0–4.8 HKD Rationale: Corresponds to the conservative scenario of 3.7–4.2x sales and leaves enough discount versus base-case valuation.

  • Acceptable holding price: 5.8–6.8 HKD

  • Clearly overvalued price: Above 8.0 HKD

  • Current price classification: Outside the three tiers

  • Is it worth waiting for a better price: Yes. Better buy triggers are, first, price returning below HKD 4.8; second, post-listing evidence of improved operating cash flow and lower receivables and prepayments as a percentage of revenue. The opportunity cost of waiting is missing short-term IPO sentiment, but the trade-off is clearer cash-flow verification.

  • Target holding period: 3–5 years

  • Expected annualized return: Bear about -38%; base about -12%; bull about +19%

  • Maximum loss risk: 50%–60%. Triggers would be large-model service scaling below expectations, declining gross margin, continued working-capital deterioration, and valuation multiples falling back to hardware-stock territory.

  • Signals that trigger reassessment: Large-model service revenue share falls below 40%

  • Comprehensive gross margin stays below 20% for two consecutive periods

  • Operating cash flow is negative for two consecutive periods and materially weaker than net profit

  • Top-five customer revenue share rises back above 55%

  • Borrowings and lease liabilities continue rising and trigger refinancing or obvious dilution

【Valuation Range】

  • current: 7.20 (as of 2026-06-11, offer price/pre-listing basis)

  • bear (conservative · ideal buy zone): [4.0, 4.8]

  • base (reasonable · acceptable holding zone): [5.8, 6.8]

  • bull (optimistic · above the clearly overvalued line): [8.0, 9.2]

The main uncertainties in this research are fourfold. First, the prospectus only provides a directional growth description for the first four months of 2026 and does not give precise profit or cash-flow data. Second, maintenance capex and expansion capex are not separately disclosed, so only directional judgments can be made. Third, multispectral AI large-model services are a new submarket, and industry statistics plus comparable companies are still immature. Fourth, the company has not yet listed, so secondary-market liquidity, first-day pricing, and subsequent institutional coverage have no historical sample.

Main reference sources include: HQVT's formal prospectus dated 2026-06-11, the first application version dated 2025-08-06, and the updated application version dated 2026-03-16; official IR or official websites of Teledyne, SenseTime, Guide Infrared, and other companies; and public market-data pages such as Yahoo Finance, Reuters, Google Finance, and AsianBondsOnline.

Other Securities Mentioned in This Report

  • TDY.US — Teledyne/FLIR is a mature global reference for multispectral imaging and infrared sensing, used to calibrate HQVT's "hardware + sensing platform" valuation.

  • 688002.SHG — Raytron Technology is an important Chinese reference for the infrared detector and thermal-imaging underlying device chain, used to compare HQVT's device moat and earnings quality.

  • 002414.SHE — Guide Infrared is a peer in infrared thermal imaging and integrated optoelectronic systems, used to observe HQVT's scale gap and valuation mapping on the hardware side.

  • 00020.HK — SenseTime is a Hong Kong-listed valuation reference for computer vision and generative AI, used to judge whether HQVT's "large-model service premium" has been overdrawn.

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

TDY6880020024140020

Multispectral AIIntelligent SensingLarge-Model ServicesHong Kong IPOSecurity
Reader Q&A10

Baillie Framework · Ten Questions for Growth Investing

10

Hunting ten-year five-baggers among great growth stocks — pressing the upside question: "Can it get much bigger?"

Baillie Framework · Ten Questions for Growth Investing — score profile: 45/100 total Ceiling 5/10 · Revenue 2x 7/10 · Next engine 5/10 · Moat 4/10 · Reinvention 5/10 · Management 7/10 · Customer need 4/10 · Unit economics 3/10 · 5x path 2/10 · Blind spot 3/10 0510 How high is its market ceiling? Is it expanding an existing pie, or creating an entirely new market? — 5/10 Ceiling 5 Can its revenue at least double over the next five years? Will growth be driven mainly by volume, price, or new businesses? — 7/10 Revenue 2x 7 Five years from now, what will take over as the next growth engine? Does this "second curve" exist today? — 5/10 Next engine 5 What is its core competitive advantage? Will this moat widen or narrow over the next three to five years? — 4/10 Moat 4 If its core business is disrupted, does it have the genes for self-reinvention? How does it treat mistakes and bad news? — 5/10 Reinvention 5 Does management, especially the founder, have a long-term view and deep alignment with the company? Is it willing to sacrifice current profit for five to ten years later? — 7/10 Management 7 If it disappeared tomorrow, how much would customers miss it? Is its growth model sustainable and not dependent on harming society or regulation? — 4/10 Customer need 4 What are the unit economics of this business, including gross margin and incremental returns? Do they improve or worsen as scale grows? Where does the money earned go? — 3/10 Unit economics 3 What conditions must hold simultaneously for it to rise fivefold in ten years? Are those conditions realistic? What expectations are implied by today's share price? — 2/10 5x path 2 Why has the market not realized all this yet? Is it because the market cannot understand it, looks down on it, or cannot look far enough? What will become the "narrative inflection point"? — 3/10 Blind spot 3
  • How high is its market ceiling? Is it expanding an existing pie, or creating an entirely new market?5/10

    The market ceiling has two layers. The existing pie the company operates in is medium-sized; its real bet is on a new sub-segment it is helping to create, and that segment is still very small today. According to the Frost & Sullivan figures cited in the prospectus, China's multispectral AI market grew from RMB 6.3 billion in 2020 to RMB 20.0 billion in 2025, and is expected to reach RMB 79.4 billion in 2030. The company's overall share is only 3.3%; its modules and perception terminals rank fourth, with shares of 3.8% and 0.7%, respectively. That is a case of expanding an existing pie in a highly fragmented industry. The real new market is multispectral AI large-model services, where the company ranks first with a 23.0% share (this segment leadership had already been disclosed in its second filing). But this sub-segment was only about RMB 1.545 billion in 2025 and is expected to be about RMB 8.4 billion in 2030. The honest arithmetic is this: even if the market reaches RMB 8.4 billion as expected and the company keeps its 23% share, that corresponds to only about RMB 1.9 billion in revenue. That can support revenue multiplying several times, but not a giant-company narrative of fivefold growth over ten years. The real upside would come if multispectral perception layers are adopted as standard components by autonomous-driving R&D and by more AI systems beyond IDC, rewriting the definition of the track. At the prospectus level, however, this remains a direction, with no order evidence yet.

    Jun 11, 2026
  • Can its revenue at least double over the next five years? Will growth be driven mainly by volume, price, or new businesses?7/10

    It can double, and the probability is not low. The drivers are volume and new businesses; price increases are basically not something to count on. On the base, 2025 revenue was RMB 668.5 million, up 27.9% year on year (public reports also confirm 2025 revenue of about RMB 670 million and a profit decline of about 27%). Doubling in five years requires only about a 15% compound growth rate. The prospectus expects the multispectral AI large-model services sub-segment to grow at a 41.1% CAGR from 2025 to 2030, while the overall market expands from RMB 20.0 billion to RMB 79.4 billion; the company's revenue was still growing year on year in the first four months of 2026. Structurally, large-model services, commercialized only in 2024, already contributed 53.1% of revenue in 2025. This is classic new-business-driven growth. Modules, with a gross margin of 11.2%, and terminals, at 17.7%, are volume-running entry points and have no pricing power to speak of. Two discounts must be stated. First, project-based revenue is highly dependent on major-customer timing: the top five customers accounted for 46.8% of revenue in 2025, and Customer F once accounted for 35.5% in 2024. Second, the growth history is not linear: revenue plunged 47.9% in 2023. More important is the quality of the doubling. In 2025, operating cash flow had a net outflow of RMB 129.9 million, diverging from net profit of RMB 29.354 million. If revenue doubles by doubling receivables and prepayments at the same time, the significance for shareholder value will be heavily discounted.

    Jun 11, 2026
  • Five years from now, what will take over as the next growth engine? Does this "second curve" exist today?5/10

    Strictly speaking, large-model services, which account for 53.1% of revenue today, are the "second curve" that took over only two years ago. The next engine that would take over five years from now currently exists only in the use-of-proceeds table. One handoff of growth engines has already happened: modules and terminals established scenario entry points, and multispectral AI large-model services, commercialized in 2024, contributed 53.1% of revenue in 2025 with a 30.4% gross margin, becoming the largest source of both revenue and gross profit (when the company filed as a multispectral AI technology leader, this structure was already described plainly). There are three candidates for the next baton, all corresponding to IPO use of proceeds: MEMS chip R&D, accounting for 8.2% of proceeds and, if successful at the device level, a key step from solutions provider toward the underlying layer; next-generation intelligent perception products, at 5.9%; and global sales expansion, at 15%. Today all three are still expenses and plans, and none has generated material revenue. Scored by Baillie Gifford standards: the second curve has only just been established, with a commercialization history of only two years, which the report lists as the No. 1 business risk; the third curve has not yet sprouted. The most generous reading comes from precedent: after revenue plunged 47.9% in 2023, the company switched customers and scenarios, revenue surged 346.4% in 2024, and it incubated large-model services. That shows the company's ability to incubate new engines is real, which is more informative than the list of curves itself.

    Jun 11, 2026
  • What is its core competitive advantage? Will this moat widen or narrow over the next three to five years?4/10

    The moat is narrow and has not yet been tested after listing; over the next three to five years, the pressure for it to narrow is greater than the evidence that it will widen. The foundation: a full-chain "optics-sensors-imaging-computing" architecture plus the self-developed Zhiyuan Origin large model, an R&D team of 156 people accounting for 43.1% of employees, 2025 R&D investment including capitalized R&D of RMB 105.3 million, equal to 15.7% of revenue, and a 23.0% share in the multispectral AI large-model services sub-segment (media therefore called it a multispectral AI leader). But each point needs a discount. 57.2% of R&D spending is outsourced to third parties, raising doubts about the depth of the technology. First place in a sub-segment looks more like point-in-time leadership in a new market: the company's share of the entire China multispectral AI industry is only 3.3%, the industry is highly fragmented, and early project concentration can also produce a nice ranking. Its competitive position is squeezed from both ends: upstream device leaders such as Raytron Technology and Guide Infrared can move downstream into systems, while downstream platform companies such as SenseTime can move into vertical solutions. A niche full-stack player is most vulnerable to this kind of squeeze. There is no evidence of customer lock-in. Customer F, which accounted for 35.5% of revenue in 2024, fell to 10.7% in 2025, showing that order switching can happen quickly. In the next three to five years, there is almost only one realistic path for the moat to widen: turn the MEMS chip R&D that accounts for 8.2% of proceeds and model capabilities into device-level barriers. Until that is delivered, this is a shallow moat maintained by running fast.

    Jun 11, 2026
  • If its core business is disrupted, does it have the genes for self-reinvention? How does it treat mistakes and bad news?5/10

    There is one genuine record of reinvention, and its disclosure of bad news is reasonably candid. This is one of the few dimensions where the company comes close to Baillie Gifford standards, but the sample size is only one. The evidence of reinvention is in the financial statements: revenue plunged 47.9% in 2023 as old customers and old demand faded. The company switched its customer structure and scenarios, revenue surged 346.4% to RMB 522.6 million in 2024 and it turned profitable, while large-model services were commercialized at the same time. In 2025, that business already accounted for 53.1% of revenue. Transforming from a low-margin module vendor into a three-layer structure of "hardware positioning, software capturing profit" is a turn fully recorded in the statements. Its attitude toward mistakes also earns points: the first application version bluntly attributed the 2023 decline to active customer-structure adjustment, customer inventory digestion, retreat of pandemic demand, and macro weakening, without packaging it as a strategic upgrade. Three discounts also need to be recorded. First, this reinvention had a passive component, with old demand fading first. Second, the funding cost of reinvention was not small: operating cash flow had a net outflow of RMB 129.9 million in 2025, and interest-bearing and lease liabilities totaled RMB 324.1 million at the end of April 2026. Another disruption would require the capital market to replenish ammunition. Third, since its founding in 2013, the company has been led by founder Zhou Bo alone as chairman and general manager. The ability to reinvent currently looks more like the founder's personal endowment; evidence that it has become organizational or cultural is still not visible.

    Jun 11, 2026
  • Does management, especially the founder, have a long-term view and deep alignment with the company? Is it willing to sacrifice current profit for five to ten years later?7/10

    The alignment of interests is real, and there is evidence of a long-term investment posture. But on the question of "sacrificing the present for five to ten years later," the IPO pricing itself adds a less generous footnote. Alignment: founder Zhou Bo founded the company in 2013 and, before the IPO, directly and through platforms including Zhongcheng Tianying, controlled a total of 48.87% of equity; after listing, he still controls about 43.5%. The new shares issued this time account for only 11% of total share capital, so dilution is restrained. The harder evidence is personal credit backing: a considerable portion of the company's past bank loans was jointly guaranteed by Mr. Zhou, about RMB 181.7 million as of the latest practicable date, and expected to be released only after listing. Evidence of sacrificing current profit: 2025 net profit fell to RMB 29.354 million, mainly because listing expenses and R&D expenses rose; 50% of proceeds are allocated to R&D, including 25.3% to large models and 8.2% to MEMS chips, 25% to capacity expansion, and only 10% to working capital. There is no trace of dressing up profit before listing. The offsets also have to be put on the table: post-money valuation in the primary market rose from about RMB 1.0 billion to RMB 3.55 billion in nine months, then to about RMB 4.82 billion implied by the offering. The valuation surge on the eve of filing had already drawn public questions about the promoters and financial authenticity. Issuing to new shareholders at about 164 times 2025 P/E shows deep alignment, but very little concession.

    Jun 11, 2026
  • If it disappeared tomorrow, how much would customers miss it? Is its growth model sustainable and not dependent on harming society or regulation?4/10

    Its indispensability is weak: customers would be inconvenienced, but they have alternatives. On society and regulation, there is no original sin; the real long tail is export controls. The customer interface is system integrators, and the top five customers accounted for 46.8% of revenue in 2025. Modules have only an 11.2% gross margin and are replaceable perception bases; terminals at 17.7% are not exclusive either. The stickiest large-model services, with a 30.4% gross margin and 23.0% sub-segment share, recognize revenue by project acceptance. Switching suppliers mid-project is costly, but there is no evidence of lock-in between projects. Customer F's share of company revenue fell from 35.5% in 2024 to 10.7% in 2025, showing the dependency is more the company's dependence on customers. The overall industry is fragmented, with the company's total share only 3.3%, and integrators have a long list of alternatives. The sustainability side looks better: in 2025, the vast majority of revenue came from private-sector customers, fiscal-project exposure was limited, and the growth model is not built on harming users or regulatory arbitrage. Geopolitics is the exception: the prospectus discloses that products are generally not directly subject to EAR, but cloud-computing services use servers containing U.S.-origin integrated circuits, so tighter export controls could choke supply. Overall judgment: if it disappeared, customers would run new tenders, and few would write eulogies. That is exactly the part of the current valuation where no premium should be paid.

    Jun 11, 2026
  • What are the unit economics of this business, including gross margin and incremental returns? Do they improve or worsen as scale grows? Where does the money earned go?3/10

    The gross-margin structure is improving, while cash conversion is worsening. At present, the larger this business grows, the heavier its capital occupation becomes; it has not yet proved that bigger is better. On gross margin: modules are 11.2%, terminals 17.7%, and large-model services 30.4%. High-margin services rose to 53.1% of revenue, pushing the overall gross margin from 18.8% in 2024 to 22.3% in 2025. The structural improvement is real. But 30.4% is not high for a business priced with a software story. It depends on heavy resources such as HPC servers, and in 2025 it made one-off prepayments of about RMB 146.9 million to four suppliers to secure computing power and other resources. Incremental returns are the weakness: 2025 net profit was RMB 29.354 million, but operating cash flow had a net outflow of RMB 129.9 million. By the end of April 2026, inventories were RMB 270.8 million, trade receivables RMB 257.6 million, prepayments and other receivables RMB 150.7 million, and current borrowings RMB 245.7 million. Every additional yuan of revenue drags working capital forward with it, so expansion currently consumes cash rather than producing it. Where the money goes: IPO net proceeds are about HKD 537 million, with 50% for R&D, 25% for capacity expansion, 15% for global sales, and 10% for working capital. The direction addresses the problems, but media have already questioned the reasonableness of the capacity-expansion and working-capital fundraising uses, which also confirms that internal cash flow cannot support this spending list. Whether scale effects emerge can be tracked with one signal: when the ratio of receivables plus prepayments to revenue starts falling.

    Jun 11, 2026
  • What conditions must hold simultaneously for it to rise fivefold in ten years? Are those conditions realistic? What expectations are implied by today's share price?2/10

    At least four conditions must hold simultaneously, and the offer price has already prepaid the earlier good news. Start with the arithmetic: post-offering market cap is about HKD 5.574 billion; fivefold means about HKD 27.9 billion, or about RMB 24.1 billion. Supporting that would roughly require: first, revenue rising from RMB 668.5 million in 2025 to RMB 4.0–5.5 billion, a ten-year CAGR of 20%–24%, which means after fully capturing the ceiling of the large-model services sub-segment, expected to be about RMB 8.4 billion in 2030 and corresponding to only about RMB 1.9 billion if the company keeps a 23% share, it still has to create another growth pole in overseas markets, new scenarios, or chips; second, net margin rising from 4.4% to 13%–15%, proving this is software economics rather than project-engineering economics; third, operating cash flow turning from a net outflow of RMB 129.9 million in 2025 to sustainably positive, with the share of receivables and prepayments falling; fourth, the market still being willing to assign a 30–40 times P/E at that time. Each condition is possible on its own; multiplying all four together makes it a low-probability event. What does today's HKD 7.20 imply? About 7.2 times 2025 P/S and about 164 times P/E, or about 87 times adjusted P/E. If profit does not grow for three years, the earnings yield is only about 0.6%, far below Hong Kong's ten-year government bond yield of about 3.279%. In the report's three-scenario valuation, the offer price implies -44% to -33% versus the conservative case, and only +11% to +28% even in the optimistic case. The ideal buying zone is HKD 4.0–4.8. At the filing stage, media were already asking whether it was telling a story about the future or overdrawing the future. A fivefold path exists, but the starting price first owes investors a period of valuation digestion.

    Jun 11, 2026
  • Why has the market not realized all this yet? Is it because the market cannot understand it, looks down on it, or cannot look far enough? What will become the "narrative inflection point"?3/10

    The market has already realized it in advance. The growth narrative was fully priced before listing, and the Baillie-style perception gap currently mainly exists in the opposite direction. Evidence: media had long packaged it as the first multispectral AI stock, and Tencent Finance was already asking at the second filing whether it was "talking about the future or overdrawing the future". Post-money valuation in the primary market rose from about RMB 1.0 billion to RMB 3.55 billion in nine months, then to about RMB 4.82 billion implied by the offering. The offer price translates to about 7.2 times 2025 P/S and about 164 times P/E, higher than Teledyne's roughly 4.7 times P/S and also higher than SenseTime's roughly 9.8 times P/S. Even the short thesis has become public: questions about the valuation surge before filing and the coexistence of high cash and high debt have appeared in financial media. Under the three-way framing in the question, the problem with this new stock is that the market is looking too far and paying too early, folding the year 3–5 story into the offer price first. If there is a positive expectation not yet priced in, there is only one: the market is trading it as an AI-concept new stock, and has not yet modeled it as a full-stack perception platform that could internalize MEMS device capabilities. That layer needs results from the chip R&D that accounts for 8.2% of proceeds. The narrative inflection point therefore lies in verification rather than discovery: if the first two periodic reports after listing show operating cash flow turning positive, the ratio of receivables and prepayments falling, large-model services holding above 50% of revenue, and the share of the top five customers declining, the story moves from plausible to proven, and valuation has a reason to step up again. Conversely, if gross margin falls below 20% or the service share drops below 40%, it will be reclassified as a hardware project company.

    Jun 11, 2026
Ask about this report

Members can ask about this report; once answered it appears under "Reader Q&A" on this page. You can also highlight a passage in the text to ask about it directly.