Horizon Robotics: A Zen Horizon Framework Deep-Dive

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

• QHow high is its market ceiling? Is it enlarging an existing pie, or creating an entirely new market?6/10

  A

  Bottom line first: the ceiling is high enough, and it has both attributes: enlarging an existing pie and creating a new market. This is one of the strongest dimensions in Horizon Robotics' Baillie Gifford narrative. But the honest net judgment is that the height of the ceiling and the share Horizon can reliably capture do not move in lockstep. The richest high-end segment is precisely where it is weakest, so a large TAM does not mean Horizon's addressable and defensible share is large. This question deserves to be counted as a strength, but with a discount.

  Start with the size of the pie itself. Smart-driving compute chips are still in the early stage of a dual lift from penetration and ASP: according to industry estimates cited by Gelonghui / Three Thirds, China's smart-driving SoC market is expected to grow from about USD 1.96 billion in 2024 to about USD 6.18 billion in 2030, with a 2025–2030 CAGR of about 21%. The ASP ladder is extremely steep: historical average prices for low-end chips were only USD 30–50, mid-range chips USD 100–150, while high-end chips supporting urban NOA reach USD 400–450. This means that as smart driving moves from usable to good, value per vehicle rises by orders of magnitude rather than linearly. The driver is the steepening penetration curve: under the report's methodology, urban NOA penetration rises from about 21.6% in 2025 to about 42.6% (Xinhua also confirms that urban NOA entered a scale-breakout phase in 2025 and expects it to become a mainstream function by 2030; a more conservative insured-vehicle methodology gives about 6.7%→17.9%. The methodologies differ, but the direction is the same: several-fold volume expansion). Add Journey-series cumulative shipments passing 10 million in 2025-08, around 400 nominated vehicle models, and 40+ partner automakers, and the picture is clear: Horizon is both enlarging an existing ADAS chip pie and, through autonomous driving and the company's extension into robotics / general AI compute, carrying an element of new-market creation. For Baillie Gifford's question about growth room over the next decade, this is a relatively solid upside imagination.

  But a high ceiling does not mean Horizon will share it proportionally. Three discounts need to be stated plainly. First, structural mismatch: Horizon's dominance is concentrated in high-volume but low-ASP mid-to-low-end ADAS, where it ranks first with a 47.7% share of domestic-brand ADAS compute solutions. Yet in the richest and highest-ASP high-end urban NOA segment, it ranks only third (14.4%, behind Huawei's 15.2%). The third-party high-end solution supplier market is even more concentrated, with Momenta (about 60%) and Huawei (about 20%) together taking more than 80%. In other words, the thickest layer of the pie is expanding quickly, but Horizon's position there is relatively weak. Second, in overall NEV installations, Nvidia has already overtaken it to rank second with about 42.4% (Horizon about 36.6%). Nvidia Thor's roughly 2000 TOPS compute ceiling directly targets high-end incremental demand, while Huawei ADS 4.0 attacks from the other side. Horizon has no assured ability to hold share rather than be diluted in the most valuable incremental segment. Third, price wars and customer in-house chip development: core automaker customers developing their own chips, alongside industry price wars, will pressure both its addressable share and ASP, further decoupling the larger pie from the size of its own slice.

  Net judgment: on Q1, Horizon deserves a moderately strong Neutral rating. The market ceiling is genuinely high and includes real option value from new markets, meeting Baillie Gifford's prerequisite of a large opportunity. But the share it can address and defend is constrained by weak high-end positioning, overall overtaking by Nvidia, customer in-house development, and price-war pressure. For a company already priced on an optimistic case (P/S about 17–18×) while still posting large losses (FY2025 adjusted net loss of RMB 2.811 billion, widening by about 67%), a high ceiling is necessary but not sufficient. Space is not the issue. Whether it can convert that space into share and profit in the richest segment is what Q2–Q9 must really press on.

  Jun 4, 2026
• QCan its revenue at least double over the next five years? Will growth mainly be driven by volume, price, or new businesses?7/10

  A

  Bottom line first: revenue doubling over the next five years is almost the most certain item in Horizon's ten Baillie Gifford questions. Based on the current +57.7% FY2025 growth rate and management's 2026 guidance of around 60% growth while maintaining 60%+ gross margin, revenue could double within one or two years and is highly likely to reach several times its current scale within five years. This is a strong dimension that contrasts with many mature growth stocks that are slowing. But the quality of that doubling must be discounted: high growth is currently being purchased with large losses and financing support, not generated by self-funded cash flow.

  Breaking down the growth sources: it is mainly driven by volume, while price is actually a headwind. In FY2025, the real surge came from the products and solutions segment (chips / modules), with revenue of RMB 1.622 billion, +144.2% YoY, and its share jumping from 28% to 43%. The engine was Journey 6 volume ramp: full-year Journey-series shipments reached 4.01 million sets (+38.8%), including 1.8 million mid-to-high-end NOA-capable solutions, nearly 5 times the prior year. Add SuperDrive (HSD) exceeding 22,000 sets slightly more than one month after mass production in 2025-11, with a 2026 shipment target of about 400,000 sets, plus the absorption of share after Mobileye's retreat in China, and the visibility of volume upside is high. Price, by contrast, is a headwind. Smart-driving democratization is being fought through price wars: gross margin slid from 77.3% in 2024 to 64.5% in 2025 (about 13 percentage points), a classic volume-price tradeoff in which ASP is exchanged for shipments. Management's narrative says future ASP recovery will depend on a rising mix of high-end solutions such as HSD, but that still needs to be delivered. The other half of revenue, licensing and services at RMB 1.935 billion (+17.4%), is growing modestly and is highly dependent on licensing fees from the Volkswagen joint venture CARIZON, so it is not the main force behind doubling.

  One honest compression: doubling is almost certain, but the durability and quality of growth carry real risk, so Baillie Gifford-style blue-sky thinking should not be copied over mechanically. First, the 2026 60% guidance is strictly for the automotive business (2025 automotive segment growth was 54%), not a company-wide commitment. Second, the sustainability of high-end volume is still unproven: the shipment mix of mid-to-high-end chips in 2025 was high at the start and lower later, around 50% in the first half and about 40% in the second half; the main volume driver remains low-priced basic ADAS. Third, demand is concentrated and unstable: the top five customers contributed as much as 77.9% of revenue in the prior-year period, while Li Auto's M100 self-developed chip already returned samples in Q1 2025, and NIO and XPeng are also developing AI chips in-house. Core customers pulling orders through in-house development is the structural threat hanging over volume growth. Overall, the net judgment for Q2 is: doubling is almost locked in, but growth quality and sustainability deserve a clear discount. Several-fold revenue growth over the next five years is highly probable, but this is a high-speed curve supported by losses and financing, by volume gains exchanged for price declines, and by dependence on a few customers not moving in-house. It is not yet validated high-quality compounding.

  Jun 4, 2026
• QFive years from now, what will take over as the next growth engine? Does this second curve exist today?3/10

  A

  Bottom line first: Horizon's real growth engine today is still the first curve: smart-driving chips themselves, not robotics. The so-called second curve (D-Robotics / embodied AI, Robotaxi) does exist today, but only as a tiny-base seedling and a long-term option. It is far from ready to take over as the main driver. The company has not even proven profitability on the first curve, so it is too early to talk about a second-curve handoff. This question requires unpacking the wording. Baillie Gifford asks who takes over five years later, implicitly assuming the first curve matures or slows and another curve steps in. Horizon's first curve, smart-driving ADAS/AD chips, is far from mature. Report data shows the automotive business still accounted for 94.6% of FY2025 revenue. High-end SuperDrive (Journey 6P end-to-end) only entered mass production in 2025, shipped about 22,000 sets, and is only targeting about 400,000 sets in 2026, while Journey 7 is still in planning. In other words, the first curve itself is still climbing steeply and will be the main engine over the next three to five years, not a mature business waiting to be replaced.

  Take the second curve out and weigh it honestly: non-automotive solutions revenue in the report was only RMB 201 million in FY2025, or 5.4% of revenue. Although it was +179.9% YoY, that 179.9% growth was built on a nearly negligible base. The incremental contribution was mainly about RMB 94 million more non-automotive processing hardware sold. Using this to argue that a second growth pole has formed confuses base-effect illusion with growth quality. The most imaginative candidate is D-Robotics, spun out from Horizon at the beginning of 2024. It focuses on non-automotive / embodied intelligence, completed a USD 100 million Series A in 2025-05 led by Hillhouse, 5Y Capital and others, launched the hundred-TOPS-class RDK S100 platform, positions itself as a picks-and-shovels supplier for the robotics era, and has landed customers including Unitree G1. But two things need to be clear. First, D-Robotics has raised financing independently and operates independently. Strictly speaking, it is an affiliated ecosystem asset held by Horizon, not a consolidated second-curve revenue stream. Second, even its own CEO admits that general embodied intelligence is at least 5 years away. This is an option with a long realization period, not a current engine. Robotaxi is similar: the company plans to start L4 pilots in the second half of 2026, but the story remains helping ecosystem partners commercialize operations over the next 2–3 years, with zero contribution today.

  Net judgment: Horizon's second curve at this moment is a real seedling and option. The long-term imagination space of embodied AI should not be dismissed, and the company name Robotics indeed points to an edge-AI / embodied vision. But it has neither taken the lead role nor constitutes quantifiable evidence of a handoff. For a Watch-rated company with FY2025 revenue of RMB 3.758 billion but still large losses, and with a full valuation (P/S ~17×, market cap about HK$75.68B / current price about HK$5.17), the more practical view is that the main battleground over the next five years remains the first curve: whether high-end smart-driving chips can scale and prove unit economics and profitability. Embodied AI / Robotaxi is a call option that can only be exercised after winning the first curve, not an already existing second engine that can be priced into valuation today. Treating the robotics vision as a second curve in advance and paying a premium for it is the narrative inflation investors most need to guard against in this stock.

  Jun 4, 2026
• QWhat is its core competitive advantage? Will this moat widen or narrow over the next three to five years?5/10

  A

  Bottom line first: the moat is real, but its quality and direction diverge: wide in mid-to-low-end, thin in high-end. Horizon's moat is not narrative packaging; it has hard support: ① a software-hardware co-designed proprietary BPU architecture (Bernoulli→Bayes→Nash [Journey 6, native large-parameter Transformer/BEV]→Riemann [Journey 7]) paired with the OpenExplorer toolchain (benchmarked against CUDA, lowering automaker migration costs), creating coupling among chips, toolchains, and reference algorithms; ② a scale / data flywheel: cumulative Journey shipments passed 10 million sets in 2025-08, the first domestic smart-driving company to do so, with FY2025 shipments of 4.01 million, nominations for nearly 400 vehicle models, and 40+ partner automakers; ③ share barrier: first in domestic-brand ADAS with 47.7% share; ④ high switching costs: automotive-grade certification, nomination cycles measured in years, and high midstream replacement cost after mass production (Li Auto's Schumacher and XPeng's Turing self-developed chips both need years of independent tape-out validation); ⑤ open ecosystem: ties with Tier1s such as Bosch, Zhuoyu, and Denso. Together these form a fairly solid moat in mid-to-low-end assisted driving, far more than volume bought by subsidies.

  But this moat has two fatal quality discounts, which mean that over the next three to five years it will likely widen in some segments while being eroded in the richest high-end and third-party markets. First, the moat has not yet translated into pricing power. A real franchise should show volume and price rising together and gross margin expanding. Horizon shows the opposite: while FY2025 revenue was RMB 3.758 billion (+57.7%), adjusted net loss was still RMB 2.811 billion. To defend share, it actively cut prices for Journey 3/5 in 2025, pushing adjusted gross margin for product solutions down to about 42.5% (down about 13 percentage points YoY). In other words, today it is exchanging lower prices for scale. That can widen the volume barrier, but it also proves that at the hardware layer it does not yet have a moat that allows it to raise prices to customers. The pricing-power attribute of the moat has not been validated by profitability.

  Second, the richest high-end segment over the next three to five years is being squeezed by three forces at the same time, so directionally it is more likely to narrow than widen. ① Not the high-end leader, with an upper ceiling: in urban NOA, the market with the thickest profit pool, Nvidia still leads and Huawei ranks second. Horizon is among the fastest growers and has narrowed the gap with Huawei to about 0.8 percentage points (about 14.4% vs 15.2%), but it remains third, while Nvidia Thor looms overhead and next-generation compute continues to pressure it. ② Core customers' in-house development strikes at the foundation. This is the most fundamental long-term threat to the third-party chip business model: NIO, Geely, and XPeng have completed self-developed tape-outs; Li Auto's Schumacher (about 40 billion transistors, 2026 mass production) and BYD are in development. Once leading automakers ramp volume, Horizon's high-end addressable market will be structurally compressed, leaving more mid-tier customers unwilling or unable to develop chips in-house. ③ Price wars continue to pressure hardware gross margin, and the dependence on CARIZON, both a major customer and shareholder-linked joint venture, further limits bargaining room in the rich segment.

  Net judgment: the most honest evaluation is that Horizon has a real but not thick moat. In mid-to-low-end assisted driving, it has genuine and still-widening scale / ecosystem / switching-cost barriers, enough to keep it a core domestic beneficiary of the smart-driving democratization wave. But in the high-end and third-party addressable markets that determine long-term excess returns, the moat is thin, and over the next three to five years the probability of erosion from Nvidia / Huawei pressure, core-customer in-house development, and price-war margin compression is higher than the probability of widening. It looks like a moat that has won scale but not yet won pricing power, far from the level of a mature franchise. This is exactly the source of tension between its Watch rating and its high-growth, high-loss, P/S ~17× valuation.

  Jun 4, 2026
• QIf the core business is disrupted, does it have the genes for self-reinvention? How does it handle mistakes and bad news?5/10

  A

  Bottom line first: Horizon has an early form of self-reinvention genes, but the sample is short and untested by a major crisis. Its handling of mistakes and bad news is mixed: Yu Kai's willingness to admit that Horizon is not first in high-end is half a point of candor, but giving only an optimistic narrative and no hard metrics on the path to profitability is a clear avoidance. Net judgment: young and resilient, but its reinvention ability has not been proven under the real pressure of a paradigm disruption, so this is a positive factor, not a source of certainty.

  Start with reinvention genes: there are positive examples. The company was founded in 2015 by Yu Kai, former deputy head of Baidu's Institute of Deep Learning (IDL), and was initially positioned as a general edge AI chip company (dual lines of Journey [automotive] + Sunrise [security / IoT]). But after launching China's first automotive-grade AI chip, Journey 2, in 2019-08, it decisively shifted resources toward smart driving, completing a successful pivot from edge AI to automotive-grade chips. At the beginning of 2024, it also spun off the non-automotive / AIoT business into independent D-Robotics, focusing on the main smart-driving path. The underlying architecture is also iterating quickly (Bernoulli→Bayes→Nash [Journey 6]→planned Riemann [Journey 7]), and the company is now extending into cockpit chips + software. Yu Kai has explicitly said he wants to be a player covering all four quadrants: smart-driving chips, smart-driving software, cockpit chips, and cockpit software, and has previewed a cockpit-driving integrated SoC in 2026. A company less than 11 years old has pivoted, kept changing architectures, and actively narrowed focus, showing it is not a rigid one-trick player. But clarity is essential: this adaptation sample is too short. Horizon is not a 40-year veteran that has survived multiple technology cycles. The disruption scenario implicit in Q5, core customers all developing in-house (Li Auto, XPeng, BYD, and Tesla are all pushing self-developed smart-driving chips) plus Nvidia Thor (single chip ~2,000 TOPS) crushing dedicated BPUs from above with general high compute, is exactly a test it has never truly withstood. The report already notes that in high-end urban NOA it lags Huawei and faces pressure from Nvidia Thor; by the end of 2025 the pattern remained: Nvidia leads, Huawei is second, Horizon is third, with the top three together taking about 90%. In other words, paradigm disruption is not a hypothetical question for Horizon. It is an active stress test. The reinvention genes exist, but there is still no track record of defending or overtaking under this force.

  Now look at how it treats mistakes and bad news; both sides are obvious. The candid side: Yu Kai does not avoid the embarrassing fact that Horizon is not the high-end leader, publicly acknowledging that in urban NOA it is second only to Nvidia and Huawei, meaning third. Report data also confirms that FY2025 gross margin fell from a peak of 77.3% to 64.5% and adjusted net loss widened to RMB 2.81 billion (+67%). It also clearly explains the difference in methodology around IFRS moving from profit to loss, with the huge RMB 10.469 billion loss mainly due to non-cash accounting items such as fair-value changes in preferred shares (about RMB 6.66 billion), rather than using accounting illusions to hide the fact that the operating business remains loss-making. The evasive side: there is still no clear timetable for breakeven. Management frames huge R&D spending (137% of revenue) as converting current profit into a future technology moat and confidently says 2026 can deliver around 60% growth and maintain gross margin above 60%, but it still refuses to give hard indicators for when profitability arrives or when the loss ratio inflects. This is a classic case of using a higher-growth narrative to cover profitability questions: the pleasant numbers on growth and share are given in full, while the commitment most in need of being nailed down, the path to profit, is left suspended.

  Putting this together for Q5: Horizon is young, has successfully pivoted, and keeps adjusting architecture and business direction; its reinvention-resilience genes are real. But the evidence base is short and has not been tested in real combat under the dual disruption of customer in-house development and general high compute, so this resilience is more a potential to withstand pressure than proven ability to do so. On bad news, Yu Kai's willingness to admit that Horizon is not first in high-end and its transparent disclosure of IFRS versus adjusted metrics deserve credit; but the lack of a breakeven timetable and the optimistic path to profitability are real avoidance and must be penalized. In Baillie Gifford LTGG's question of why the market has not realized this yet, the market has actually already realized that Horizon has good young-company qualities, but reinvention remains untested and bad-news handling is mixed. That is exactly the tension behind the stock falling by half from its high while still carrying a P/S ~17× valuation. This item is not a short thesis, but it is far from a blind downside anchor.

  Jun 4, 2026
• QDoes management, especially the founder, have a long-term view and deeply aligned interests with the company? Is it willing to sacrifice current profits for five to ten years from now?6/10

  A

  Bottom line first: on Baillie Gifford LTGG's dimension of whether management has a long-term view, deep alignment, and willingness to sacrifice current profits for five to ten years from now, Horizon is positive and relatively additive, but with clear reservations. Net judgment = long-termism and founder control are real positives and fit the Baillie Gifford pattern well, but the weak protection under the W-share structure, related-party dependence on Volkswagen, and unproven profit discipline mean interest alignment looks more like strong control rights and medium economic alignment rather than textbook high skin-in-the-game.

  Start with the part that most fits Baillie Gifford: Horizon is thoroughly sacrificing current profits for the long term. The report states that FY2025 R&D expense was RMB 5.154 billion, consuming 137% of revenue (STCN figure 137.1%), with R&D staff accounting for about 71.5% of employees. This has produced repeated large losses, with adjusted net loss widening to RMB 2.81 billion (YoY +67%). Management is putting heavy money into proprietary BPU architecture (Bernoulli→Nash→planned Riemann / Journey 7) and the CUDA-benchmarked OpenExplorer toolchain. This is a classic long-cycle, platform-style heavy bet, exactly the aggressive reinvestment company Baillie Gifford likes, sacrificing current profits for a decade later and very different from a mature cash cow squeezing dividends. The long-term view is also supported by the deep cooperation with Volkswagen / CARIAD of about €2.4 billion (D-round investment + joint venture, Freshfields). A bet to co-build an underlying compute platform with a multinational automaker cannot be supported without a five-to-ten-year horizon. Founder Yu Kai himself is a technical long-termist (former deputy head of Baidu IDL, led Baidu Brain / PaddlePaddle, techwalker interview). The narrative of putting neural networks into chips and making the brain for the robotics era has been consistent for a decade. On this point, Horizon qualifies as growth-company management.

  Now look at interest alignment, which must be split honestly into two sides. The positive is firm control, allowing it to withstand long cycles. Sections 3.4/8.4 of the report state that Yu Kai, through Everest Robotics, holds 16.95% economic interest but 55.95% voting rights (W-shares with unequal voting rights, Class A shares carrying 10 votes each), while the three founders together hold about 22.4% economic interest and about 74% voting rights (Sina shareholding structure). This means the company will not be forced into submission by short-term capital, and Yu Kai can keep pouring profits into R&D under the pressure of three financings in one year and continuing losses. For a platform chip company that needs long-term investment, this founder lock-in is exactly the governance premise Baillie Gifford values for executing a ten-year roadmap. But the reservation is also here: economic interest is only 16.95%, and alignment depends on dual voting rights rather than cash equity ownership, so it is strong control with medium economic alignment. The other side of the W-share structure is weak minority-shareholder voice and poor protection (listed in the report as a governance deduction, and on the first large lock-up expiry day in 2025-04 the stock fell about 16% intraday). In other words, Yu Kai's downside exposure is not fully aligned with public shareholders. This is a real gap from the ideal skin-in-the-game where the founder has most of his net worth in the same stock.

  There are two more governance deductions that cannot be omitted. First is related-party dependence: CARIZON, the Volkswagen joint venture (Horizon 40%, CARIAD 60%, Smart Auto Club), is both Horizon's largest customer (related-party transactions were about 40.4% of revenue in 2023) and shareholder-linked through its parent company. This relationship, where a major customer is also tied to the shareholder base, is both a binding mechanism and a risk. If Volkswagen's strategy changes or it reduces holdings, both revenue and governance would be affected. Second is profit discipline remains unproven: although management (Yu Kai) has called for about 60% growth in 2026 and maintaining 60%+ gross margin (National Business Daily), it has still not provided a clear breakeven timetable. The line between sacrificing current profit for the long term and never making money is thin. With losses continuing to widen and financing support still needed, the market has reason to demand that management eventually prove it can move from burning cash for share to scaled profitability. Overall: long-termism, willingness to compress current profit, and founder control that can withstand long cycles are real positives for Horizon on Q6 relative to the Baillie Gifford pattern. But medium economic alignment, weak W-share minority protection, related-party dependence on Volkswagen, and unproven profit discipline all discount this positive. The net assessment is positive with reservations.

  Jun 4, 2026
• QIf it disappeared tomorrow, how much would customers miss it? Is its growth model sustainable and not dependent on harming society or regulation?5/10

  A

  Bottom line first: Horizon is important to customers but not unique. If it disappeared tomorrow, domestic-brand smart driving would lose a key pillar and customers would indeed miss it, but substitutes are plentiful (Nvidia / Huawei / Qualcomm / Mobileye and even customer in-house solutions can all step in), far from the irreplaceability of TSMC or Dassault. Its growth model has a hard flaw: it does not generate its own cash flow and relies on financing support through three fundraisings in one year, so the sustainability of cash-burning growth is questionable. The clean side is that it sits on the tailwind of domestic substitution and does not rely on harming society or regulation. Overall, the net judgment is medium-high indispensability but substitutable, questionable growth sustainability, and clean ethics / regulation, not enough for the high score of a mature leader.

  Start with how much customers would miss it if it disappeared tomorrow. Horizon is indeed an important pole in domestic smart driving: cumulative Journey-series shipments have passed 10 million, the first domestic smart-driving company to reach that scale. As of mid-2025, it had won nominations for about 400 vehicle models, with partner automakers covering China's top ten OEMs. Under the report's methodology, its share of domestic-brand ADAS was about 47.7%, first for two consecutive years. The breadth of this chip + algorithm + ecosystem offering means that if it disappeared, customers such as BYD and Li Auto would lose speed in popularizing low-to-mid-end smart driving in the short term. Customers would miss it. But missing it does not mean being unable to live without it: in the higher-end urban NOA compute chip segment that better determines life or death, Nvidia, Huawei, and Horizon together hold about 90% share, with Nvidia first, Huawei second, and Horizon third, and Horizon is still chasing in high-end. Horizontally, Qualcomm and Mobileye are also present. More importantly, its two largest customers, BYD and Li Auto, are themselves the main drivers of its high-end shipments, and these leading automakers (Li Auto, XPeng, BYD, plus Tesla as the global pattern) are mostly pushing in-house development. If customers lost Horizon, they would have more than one place to go. Conclusion: indispensability is important but substitutable, not monopoly-grade positioning.

  Now the real hard flaw in this question: the sustainability of the growth model. Horizon's FY2025 revenue was RMB 3.758 billion, up 57.7% YoY, an impressive growth rate, but this growth does not self-fund. Adjusted net loss was about RMB 2.81 billion (statutory loss was even at the tens-of-billions scale), and R&D spending was RMB 5.154 billion, 137% of revenue, meaning it spends RMB 1.3 on R&D for every RMB 1 of sales. Gross margin also fell from 77.3% to 64.5%. Its growth relies on financing support rather than operating cash receipts: after the Hong Kong IPO, it raised another RMB 4.674 billion through a top-up placement in 2025-06 and about HK$6.34 billion (about RMB 5.8 billion) through another placement in 2025-09, for total fundraising of more than RMB 10 billion within one year. The company also acknowledged in its interim report that it is mainly funded by financing and operating proceeds, and the market has described it as trapped in a financing–investment–loss–refinancing cycle. Cash on hand is thick (report figure RMB 20.188 billion), but it is lifted by financing, not operating cash generation, and there is still no clear breakeven timetable. Under LTGG's question of whether it can self-roll without external funding, this burn-cash-for-share model must carry a question mark.

  Finally, cleanliness is a positive. Horizon sits in the policy tailwind of domestic substitution in smart-driving chips. The business itself does not harm society, touch regulatory red lines, or rely on gray monetization, so it is clean on ethics and compliance. One objective risk to note is that its advanced-process foundry depends on TSMC, creating geopolitical supply-chain risk, but that is an external risk rather than unclean behavior. In sum, the honest net judgment for Q7 is that Horizon is an important but substitutable pole in domestic smart driving, with medium-high indispensability; its growth model is questionable because it depends heavily on financing support; and its ethics / regulation profile is clean. It does not deserve the high score of a mature leader and fits its Watch rating.

  Jun 4, 2026
• QWhat are the unit economics of this business (gross margin, incremental returns)? Do they improve or deteriorate as scale grows? Where does the money it earns go?3/10

  A

  Bottom line first: unit economics are one of Horizon's hardest wounds. Consolidated gross margin of 64.5% appears decent, but it is falling quickly (2024 peak 77.3%→64.5%, down about 13 percentage points YoY). Core hardware solution margin is thinner: products and solutions had only about 42.5% adjusted gross margin for the full year, and even fell to 24.3% in the second half. Adjusted net loss (RMB 2.811 billion, widening by about 67% YoY) expanded faster than revenue (+57.7%). In other words, at current evidence, unit economics are not improving as scale grows; they are deteriorating, and operating leverage has not yet arrived. High-margin software / IP licensing supports the reported 64.5% consolidated figure, but more and more incremental revenue comes from automotive hardware solutions in a price war, with active price cuts on Journey 3/5. The 13-point gross-margin decline is the direct cost of grabbing volume in smart-driving democratization. For a company trading at about 17× P/S, this trend of selling more while unit returns get thinner is the negative signal that Q8 most needs to confront.

  The money it earns, and the money it raises, is almost all poured into R&D. FY2025 R&D expense was RMB 5.154 billion, up 63.3% YoY and equal to 137% of revenue (132% in 2024), with R&D staff accounting for more than 70% of employees. This is a typical early-stage heavy-investment pattern for a platform chip company, building walls for next-generation high-end solutions (HSD) and a software-hardware integrated platform. But it also means that as long as gross margin keeps falling and R&D keeps scaling, the loss scissors will keep opening. It is important to be honest: IFRS net loss reached RMB 10.469 billion, but the majority of that (about RMB 6.66 billion) was non-cash, non-operating items such as fair-value changes in preferred shares, not real operating cash loss, so losses should not be exaggerated on that basis. The operating-relevant metric is the adjusted net loss of about RMB 2.8 billion. Even so, the fact that adjusted loss is still widening is enough to show the problem.

  On cash generation, Horizon still cannot self-fund and depends on external financing support. Operating cash flow returned to net outflow during the reporting period. The RMB about 20.188 billion cash balance mainly came from listing and multiple refinancing rounds (IPO proceeds of RMB 6 billion, another RMB 4.7 billion 8 months later, plus discounted placement net proceeds of about HK$6.34 billion in 2025-09), not from its own operating cash generation. Horizontally, its unit economics are still leading among domestic smart-driving chip peers: hardware-focused Black Sesame Technologies' gross margin in the first half of 2025 plunged from 50% to about 24.8%, far below Horizon, while also suffering large losses. Horizon leads loss-making peers in both gross margin and scale, but a leading loss is still a loss. Overall judgment: Horizon's current unit economics are weak and still deteriorating; scale leverage has not been proven. The whole outcome for Q8 rests on one scissors-gap question: can revenue keep expanding in 2026–2027, pull R&D as a share of revenue down from 137% to a profitable range, and stop the gross-margin decline? Until that inflection is confirmed by data, unit economics deserve an honest score of unproven but clearly weak today, and should not be beautified just because the consolidated 64.5% gross-margin number looks acceptable.

  Jun 4, 2026
• QWhat conditions must all hold for it to rise fivefold in ten years? Are those conditions realistic? What expectations are implied in today's share price?3/10

  A

  Bottom line first: for Horizon to rise fivefold in ten years, five difficult conditions must all hold at the same time: revenue must expand several-fold, large losses must turn into meaningful profit, high-end share must withstand pressure from both sides, valuation must not be crushed by the combination of no profit and a high P/S, and the embodied / Robotaxi second curve must open up. None can be missing. Each condition on its own is possible but not easy; multiplied together, a fivefold outcome is a high-variance, low-probability lottery ticket, not a base-case expectation. More importantly, today's ~17× P/S already embeds optimistic expectations of continued high-growth delivery. This is not a Baillie Gifford-style mispricing where the market has not recognized a cheap great company. It is the opposite: the market has already paid for the optimistic case.

  Put the math clearly first. A fivefold rise means market cap moving from HK$75.68B to about HK$378B and share price from HK$5.17 to about HK$26, equivalent to about 17.5% annualized for ten consecutive years. Assessing realism item by item: ① Revenue must grow from FY2025's RMB 3.758 billion (+57.7% YoY, three-year CAGR ~82%) to the tens-of-billions RMB scale. Given the industry β of concentrated Journey 6 adoption, HSD having nominations for more than 20 vehicle models, and ADAS share near 50%, several more years of high growth is the most likely of the five conditions. ② Profitability is the hardest gate: R&D is 137.1% of revenue, adjusted net loss is RMB 2.811 billion and still widening, and there is no breakeven timetable. Bringing R&D as a share of revenue back into a profitable range without falling behind requires both scale leverage and price / margin delivery, which is highly uncertain. ③ Share: urban NOA, the real high-end battlefield, is currently a one-superpower, two-strong-player market, with Nvidia leading, Huawei second, and Horizon third. It must withstand Nvidia Thor moving downward and core customers pulling orders through in-house development. Holding share is already hard; moving upward is harder. ④ Valuation: even after the stock halved, the market is still paying ~17× P/S for an unprofitable business (the high was estimated around 40×). If growth slows at the margin or losses persist too long, the downside from re-rating is far greater than for mature companies. This is the most realistic risk that can break the fivefold story. ⑤ Second curve (embodied intelligence / Robotaxi) remains narrative-stage, upside optionality rather than cash flow that can be included in the base case.

  What today's share price implies is the core of an honest assessment. Sell-side consensus target price is ~HK$11 (22–23 firms, mostly Strong Buy, implying about +110%), but this 2x target rests on 2027 forward revenue delivery and a buy-side-skewed sample. In other words, even the sell-side consensus for a double is optimistic up front; a fivefold outcome then requires profit turning positive and valuation not being cut on top of that. The market, at ~17× P/S, has already paid for sustained high growth. This contrasts sharply with the ideal Baillie Gifford target, a great company the market has not yet recognized and bought at a mispriced trough. Horizon's payoff does not come from cheapness, but from the real upside possibility of strong growth β, a large TAM, and domestic substitution.

  Net judgment, both sides: do not deny that Horizon does have high-upside potential: the smart-driving TAM is huge, domestic substitution has both policy and industry tailwinds, and growth elasticity is strong. In an optimistic case of several-fold revenue + profitability + rising share, fivefold is not physically impossible. This is its genuine value relative to mature blue chips. But it should also not be exaggerated: starting valuation is already full, the path to profitability is unclear, and the risk of permanent capital loss is materially higher than for mature companies (persistent losses + high-P/S valuation collapse + refinancing dilution combined; the near-52-week low of HK$5.07 is a footnote to fragile market sentiment). Therefore the honest answer to Q9 is that the list of conditions required for a fivefold return has low realism and each condition is independent yet must jointly hold. Today's price is not unrecognized cheapness; it is already priced for optimism. It deserves the position sizing of a high-upside lottery ticket (small, long-dated, able to tolerate zero), but it does not deserve confidence that ten-year fivefold is the central expectation.

  Jun 4, 2026
• QWhy has the market not realized all this yet? Is it because the market does not understand, looks down on it, or cannot look far enough? What will be the narrative inflection point?3/10

  A

  Bottom line first: reading Horizon's current HK$5.17 price and about HK$75.7 billion market cap as the market failing to understand and mispricing a cheap great company is the most dangerous misread of Q10. The truth is simpler and more honest: this is a stock covered by 22+ sell-side firms, unanimously rated Strong Buy, with consensus target price around HK$11. It is not a hidden gem in a cognitive blind spot. It is a well-researched star that was even overly loved at the HK$11.32 52-week high in 2025-09. The halving from the high to the current price (about 53% drawdown from the 52-week high) is not the market turning bearish on its growth. It is a return from extreme optimism that extrapolated growth mechanically (the high implied 35–40× P/S) toward a demand for evidence of profitability. Therefore an honest breakdown of Q10 must have three layers, and none of them can be glossed over with cheap equals market wrong.

  ① Does the market not understand it? Basically no. P/S is still as high as 17–18×, coverage is dense, and Buy ratings are crowded. This shows the market understands its domestic smart-driving chip leader status and its 47.66% first-place share in domestic-brand ADAS chips quite clearly. There is no lack of understanding. ② Does the market look down on it? Partly, but the direction must be set correctly. The market is not rejecting its ceiling; it already priced growth fully, even excessively. The current halving reflects several financially verifiable bad news items coming together: valuation-bubble concerns; a perception shock from metric switching (2024 IFRS net profit recorded RMB 2.35 billion of profit due to non-operating gains, but adjusted net loss in the same period widened to RMB 1.681 billion. The contrast between accounting profit and a core business still deeply loss-making was itself an ignition point for valuation compression); concerns over core-customer in-house development; Hong Kong tech sentiment volatility; and the supply shock from the 2025-04 lock-up expiry day when the stock fell more than 16% intraday. This is a return from excessive optimism and the market beginning to ask for a profit path, not undervaluation. ③ Can the market not look far enough? This is the only layer where underestimation may be real. When the market prices a loss-making company with a relatively short duration, it often underweights long-term options whose realization windows are years away: Robotaxi pilots that only land in the second half of 2026, the embodied-intelligence second curve bet by spun-out D-Robotics, and platform positioning after full smart-driving democratization. This layer of not looking far enough has some validity, but it is option value, not current margin of safety.

  Net judgment: unlike a stock that is cheap and ignored, Horizon is more like a growth story that was fully priced and then halved because profitability was questioned. The imagination space is real, but the margin of safety is insufficient, and the narrative still depends on distant delivery. Because of this, the real narrative inflection point will not be valuation itself becoming cheap, but a set of signals that can be hard-verified by financials and share data: first, a clear narrowing of the adjusted loss ratio, or even hard guidance on breakeven, which directly addresses the market's current biggest concern; second, sustained upward breakthrough in high-end urban NOA share. Horizon currently ranks third in urban NOA behind Nvidia and Huawei. Whether it can hold and lift share amid Nvidia / Huawei pressure and major customers developing in-house is the decisive factor in whether growth can defend its upside; third, embodied (D-Robotics) or Robotaxi truly becoming reportable revenue, turning the second curve from PPT into cash flow. Until these three types of signals light up in the data, the honest answer to Baillie Gifford's question about fivefold upside over the next decade is: the story is large enough, but the market has already once paid too high a price for that story. What investors are waiting for now is not another telling of the narrative, but financial reports proving it real.

  Jun 4, 2026