Horizon Robotics

Phase A — Understand the business

Company Overview

Horizon Robotics is China's leading domestic supplier of advanced driver-assistance (ADAS) and autonomous-driving (AD) compute platforms for passenger cars. Founded 2015 by ex-Baidu deep-learning chief Yu Kai; IPO'd on the HKEX in October 2024 (largest HK IPO of 2024, raising HK$5.41B / ~$696M). The model is a full-stack "chip + software + reference solution" play: it designs the Journey family of automotive-grade AI SoCs and pairs them with its own perception/planning software (Horizon Mono for front-camera ADAS, Horizon Pilot for highway NOA, and Horizon SuperDrive (HSD) for full-scenario urban AD), selling to Tier-1 suppliers and OEMs.

Revenue splits into two segments:

• Product Solutions (Journey SoC + integrated hardware solutions): RMB 1.62B in 2025, +144.2% YoY, now 43% of revenue (up from 28% in 2024). This is the volume-chip line.
• Licensing & Services (IP licensing, algorithm licensing, engineering/NRE, platform fees): RMB 1.94B, +17.4% YoY, ~52% of revenue. This is the high-margin line (94.5% GM).
• Non-automotive (~5.4% of revenue) rounds out the total.

Total 2025 revenue RMB 3.758B (~$544M), +57.7% YoY; automotive-solutions revenue RMB 3.557B (94.6% of total, +53.9%).

Scale of the franchise: partnered with 40+ automakers, empowering 300+ vehicle models, ~5M end-users cumulatively; Journey 6 series alone slated for 100+ models. Named customers include BYD, Li Auto, NIO, Geely, Audi, Hyundai, Volkswagen, Chery.

Key contract structure: Product Solutions is per-unit hardware (design-win → SOP → volume ramp over a model's life). Licensing & Services is upfront + royalty. There is no meaningful recurring take-or-pay — this is a design-win business where revenue is a lagging function of the customer's vehicle sell-through, which introduces demand risk Horizon doesn't control.

Supply Chain

Horizon is fabless — it designs SoCs and outsources manufacture. Map, upstream → Horizon → end customer ` pending prospectus confirmation]:

• EDA / IP cores → Arm (CPU IP), plus Horizon's proprietary BPU (Brain Processing Unit) NPU architecture (the differentiated core; Journey 6 uses the "Nash" BPU). Chokepoint: Arm licensing is a geopolitical dependency.
• Foundry (wafer fab) → Journey 5 was fabricated on TSMC 16nm; Journey 6 family is understood to use a mix of nodes with TSMC/Samsung as leading-edge candidates and Chinese foundries (SMIC) as the sanctions-hedge for lower tiers. This is the single most important chokepoint: a high-end automotive SoC needs an advanced node, and China's access to sub-7nm leading-edge foundry capacity is exactly what US export controls target.
• Horizon → designs SoC + writes the perception/planning stack + ships a reference solution.
• Tier-1 integrators (Continental via the HCT JV, Desay SV, Joynext, and others) → package Horizon's platform into a domain controller/ECU.
• OEMs (BYD, Li Auto, VW via Carizon, etc.) → install into vehicles.
• End market: Chinese passenger-car buyers (with a nascent Europe push via Carizon + a 2025 European office opening).

Single-source dependencies: (1) advanced-node foundry access; (2) Arm IP; (3) demand concentration in a handful of Chinese NEV OEMs (see Lens 3/13). Horizon partly de-risks node exposure by tiering — basic ADAS (Journey 3/5-class) can run on mature nodes Chinese fabs can supply, but the growth story (Journey 6 / HSD at 560 TOPS) needs the leading edge.

Competitive Advantages (moats)

What Horizon actually has:

1. Domination of the low end = a data + design-win flywheel. #1 in China's domestic-brand basic ADAS at 47.7% share. Every basic-ADAS design win seeds a relationship, a software footprint, and fleet data that can be upsold to higher tiers. Cumulative 4.01M Journey units shipped in 2025 alone (+38.8% YoY).
2. Vertical full-stack integration. Owning both the BPU silicon and the algorithm stack lets Horizon co-optimize TOPS-per-watt and offer OEMs a turnkey solution — the same architectural bet Mobileye and (differently) Huawei make. Chinese OEMs increasingly want a domestic full-stack alternative to Nvidia for supply-security and cost reasons — that is the secular tailwind ("autonomous-driving independence").
3. Switching costs at the platform level. Once an OEM ports its intelligent-driving software to the Journey/BPU toolchain for a vehicle program, re-platforming mid-cycle is expensive — real but program-scoped, not company-locked (they can and do multi-source across models).
4. Blue-chip validation: the Volkswagen relationship (VW invested ~$2.3B to form the Carizon JV in 2022; 500+ engineers; HSD to roll out on VW China models) is a genuine moat-signal — a global OEM outsourcing its China AD brain to Horizon.

Bargaining power: Weak-to-moderate over customers (design-win business, concentrated buyers who can dual-source — see 37.6% top-customer share). Weak over the critical supplier (foundry). Software licensing (94.5% GM) is where its pricing power is real; commodity hardware (34.5% GM) is where it is thin.

The honest moat verdict: Horizon's moat is strong at the low end and unproven at the high end. Basic ADAS is a near-monopoly; the mid/high NOA tier — where the profit and the narrative live — is a genuine three-way brawl it is currently losing on share to Huawei (see Lens 4/7).

Segments

By product line, 2025:

The trend that matters: the mix is shifting hard toward hardware (Product Solutions +144% vs Licensing +17%), and hardware carries 34.5% GM vs software's 94.5%. That mechanically compressed overall gross margin by 12.8pp (auto GM down 11.7pp to 67.2%). This is the central financial tension: the growth engine (Journey 6 volume) is dilutive to margin. Bulls read this as "climbing the value chain, volume before margin"; bears read it as "the high-margin licensing model is being replaced by a low-margin chip-vendor model."

By tier: mid-to-high-end NOA solutions grew ~5x YoY and reached 45% of Journey shipments (up from ~9% in 2024) — the up-tiering is real and fast, even if margin lags.

Geography: overwhelmingly China. Europe is a 2025-seeded option (Carizon/VW, European office), not yet a revenue segment.

---

Phase B — Measure performance

Earnings Result (FY2025, reported 2026-03-19)

The print:

• Revenue RMB 3.758B, +57.7% YoY — a clean beat on the growth narrative; H1'25 had already run +67.6%.
• Gross margin 64.5%, −12.8pp YoY — the mix-driven compression (above).
• R&D RMB 5.154B, +63.3% YoY = 137.1% of revenue — management (Yu Kai) framed this explicitly as a strategic choice to build foundation models + a technical moat.
• IFRS net loss RMB 10.469B (vs a profit in 2024) — but this is dominated by non-cash / non-operating items: RMB 6.664B of fair-value losses on convertible preferred shares & financial liabilities. Perversely, that loss grew because the market cap rose (the convertible/preferred instruments are marked up as equity value rises) — a "success-penalty" accounting artifact common to HK/PRC pre-IPO structures. Plus SBC and the share of Carizon/HCT JV losses.
• Adjusted (non-IFRS) net loss RMB 2.812B — adds back SBC, one-off capital-raising costs, and the fair-value swings. This RMB 2.8B is the number that matters for burn — not the RMB 10.5B headline.

Balance sheet — the decisive fact: cash & equivalents RMB 20.188B (~$2.93B), +31.3% YoY. Against an adjusted burn of ~RMB 2.8B/yr, that is a ~7-year runway. Solvency is not the risk here — Horizon is one of the best-capitalized names in its peer set. Cash actually grew, implying financing/working-capital inflows offset the operating burn.

Market reaction: the stock has de-rated ~53% YTD 2026 and trades around HK$3.64–4.08 (2026-06), off a 52-week high of HK$11.32. The de-rate is not about the print's growth — it is about post-IPO-lockup supply, margin compression, and the market repricing the high-end-NOA competitive risk (Lens 8).

Earnings Calls (sentiment trend)

No research-layer transcripts exist; drawing on reported call commentary.

Management's message has been remarkably consistent and unapologetically growth-first: Yu Kai repeatedly frames the R&D-above-revenue burn as a deliberate land-grab to build "foundation models" and win the high-end before margin matters. Recurring phrases: "technological moat," "full-scenario intelligent driving," "mid-to-high-end NOA," "60% average annual growth." The tone shift over 2024→2025 is from "prove the ADAS franchise" to "we are now fighting for the premium tier" — and the introduction of forward SKUs (a "cabin-driving fusion Agentic CAR SoC," an OS) signals ambition to expand beyond driving into the whole software-defined-vehicle stack.

What they've stopped emphasizing: near-term path to profitability. The 2025 guidance language is "~60% average annual growth going forward" with no margin or breakeven target — a tell that management is asking the market to underwrite growth, not earnings, for years.

Comps

Peer set = the global automotive AD-compute complex.

Read: on the one figure I can anchor, Horizon trades around ~10x sales with no profits — a growth-optionality multiple, not a value one. Its true comp is Mobileye (also a loss-adjacent, margin-compressed AD-SoC pure-play) rather than Nvidia (for whom AD is a rounding error). The multiple is only justifiable if you believe Horizon wins a large share of the high-end NOA tier — which it currently trails Huawei in. Dividend yield: 0% (reinvesting everything). 5-yr avg ROE: n/m — negative/pre-profit history.

Stock-Price Catalysts

What has actually moved 9660.HK (>5%):

• IPO pop, Oct 2024: +16% debut day; Yu Kai became a paper billionaire. Positive.
• Journey 6 adoption / design-win headlines through 2025: Goldman raised its target on Journey 6 traction (Oct 2025). Positive.
• Post-IPO lockup expiration (2025): flagged explicitly by analysts as a bearish overhang; coincides with the start of the multi-quarter de-rate. Negative — a supply, not fundamentals, event.
• 2026 broad China-tech/HK de-rate + margin-compression print: the stock is −53% YTD 2026, from HK$11.32 to ~HK$3.6–4.1. Negative.

Pattern the tape reveals: this name trades on (a) high-end design-win momentum and (b) share-technicals (lockups, HK liquidity, China-tech beta) far more than on the quarterly P&L (the loss is understood to be strategic). The single biggest fundamental swing factor priced by the market is competitive share in the high-end NOA tier vs Huawei/Nvidia — headlines on that axis move it most.

---

Phase C — Judge people & books

Management

Yu Kai (Kai Yu) — Founder, Chairman & CEO. World-class ML pedigree: executive head of Baidu's Institute of Deep Learning (IDL) 2012–2015 (built Baidu's autonomous-driving team + PaddlePaddle); director of NEC Labs America media lab 2006–2012. He is a founder-technologist archetype, not a professional operator — which fits a company whose thesis is "out-engineer the compute+algorithm stack" but is a flag for capital discipline.

• Track record: built Horizon from 2015 into China's #1 basic-ADAS supplier and a credible high-end challenger, landed a $2.3B VW JV, and executed HK's largest 2024 IPO. Genuinely impressive company-building; has never run a profitable business — Horizon has never posted a full operating profit.
• Skin in the game: Yu is the largest shareholder at ~13% via a family trust + wholly-owned vehicle. Founder-controlled with dual-class structure (Class B = the "-W"/weighted-voting listing). Alignment is high on upside; the dual-class means minority holders have limited governance leverage.
• Backers: Hillhouse (Zhang Lei), HongShan (ex-Sequoia China, Neil Shen), EQT Private Capital Asia (Jean Salata), plus strategic VW. A top-tier syndicate — validating, but these are also the pre-IPO holders whose unlocks pressure the stock.
• Capital allocation: the defining choice is R&D at 137% of revenue — a maximal, deliberate reinvestment bet. It has funded real share gains and up-tiering; it has also produced no profits and a margin-compressing mix shift. The Carizon/HCT JVs are strategically smart (customer-funded moat) but are near-term loss centers dragging the consolidated P&L.
• Red flags: the RMB 6.66B fair-value loss is an accounting artifact, not a governance issue — but the complexity of the preferred/convertible/JV structure (Carizon, HCT, D-Robotics) makes the true operating economics harder to read than a clean OEM chip vendor. No related-party or fraud signals surfaced.

Forensic Red Flags

Forensic lens, web-only — no filings on the shelf; figures.

1. The headline loss is 64% non-operating. RMB 6.664B of the RMB 10.469B loss is fair-value remeasurement of convertible preferred shares & financial liabilities — non-cash, and it increases as the company succeeds (market-cap-linked). Not a red flag per se, but a trap for the unwary: never quote the IFRS loss as the burn. Adjusted loss RMB 2.812B is the real figure.
2. Gross-margin quality is deteriorating by design. Overall GM −12.8pp; the shift from 94.5%-margin software to 34.5%-margin hardware is the cause. Watch whether software attach recovers as Journey 6/HSD scale, or whether Horizon is structurally becoming a lower-margin chip vendor.
3. Revenue is a lagging, customer-controlled variable. Design-win → SOP → volume depends on the OEM's vehicle sell-through. If a flagship customer's model underperforms, Horizon's revenue misses with no recourse. Combined with top-5 = 77.9% / top-1 = 37.6% concentration (H1'24, prospectus) this is the sharpest accounting-adjacent risk.
4. JV loss consolidation. Increased losses were partly "increased shared loss of CARIZON and HCT... in ramp-up". The equity-method/consolidation treatment of these customer-JVs means reported operating losses embed early-stage JV burn that is arguably investment, not core loss — but it clouds the standalone picture.
5. Cash grew while the company burned (+31.3% to RMB 20.188B). Consistent with financing inflows / working-capital timing, not an operating cash generation story — do not misread rising cash as self-funding.

Regulatory findings (required sub-section).

• SEC (EDGAR EFTS — LR + AAER): Not applicable. Horizon has no CIK and does not file with the SEC (HK-listed). Zero SEC findings — because no SEC jurisdiction, not because it was searched clean.
• Non-SEC / geopolitical (the real regulatory risk): web search "Horizon Robotics" (entity list OR sanctions OR export control) — no evidence Horizon is currently on the US Entity List or DoD 1260H list as of this writing. BUT the sector risk is acute and rising: the 2025 BIS "affiliates/50%-rule" expansion could balloon blacklisted Chinese entities ~15x, and the DoD has been adding Chinese tech names (BYD, Baidu, CATL, etc.) to 1260H. As a domestic-champion AD-compute supplier dependent on advanced-node foundry + Arm IP, Horizon is a plausible future target — an unpriced tail. Its main hedge is foundry-tiering toward domestic (SMIC) capacity for lower SKUs.
• 10-K Item 3 (Legal Proceedings): n/a — no 10-KForm 10-KA company’s audited annual report to the US regulator. The most complete thing it publishes. exists. Prospectus risk factors flagged intense competition, high customer concentration, heavy R&D, supply-chain risk, and "risks related to sanctions and export-control laws" explicitly.
• Verdict: No material litigation or enforcement found (web + prospectus, as of 2026-07-06). The dominant regulatory risk is prospective US export-control exposure, not any current action.

---

Phase D — Project & stress-test

Forward Projection

No research-layer financials; built bottom-up from reported FY2025 actuals + management's "~60% avg annual growth." All outputs, inputs labeled. No our model logged (watchlist rule + web-only base). Reporting currency RMB; Horizon reports on a Dec fiscal year.

Anchor (FY2025 actual): revenue RMB 3.758B, GM 64.5%, adjusted net loss RMB 2.812B, cash RMB 20.188B.

Revenue path:

• Base (~+50%/yr, decelerating from 57.7%): FY26 ~RMB 5.6B · FY27 ~RMB 8.2B · FY28 ~RMB 11.6B. Below mgmt's stated ~60% to be conservative on OEM-demand risk.
• Bull (mgmt's ~60%): FY26 ~RMB 6.0B · FY27 ~RMB 9.6B · FY28 ~RMB 15.4B. Requires winning high-end NOA share back from Huawei and VW/Europe ramping.
• Bear (+30%, share loss at the high end): FY26 ~RMB 4.9B · FY27 ~RMB 6.3B · FY28 ~RMB 8.2B.

Margin & profitability: GM likely stays pressured in the low-to-mid 60s near-term (hardware mix) and only recovers if software attach re-accelerates on HSD. With R&D held anywhere near current absolute levels (~RMB 5B, growing slower than revenue), operating breakeven is plausibly FY28–FY30, not before — the base case is no positive EPS in the three-year window. This is a runway-and-share story, not an EPS story — so the operating battery's EPS lens genuinely doesn't bind here; the tracked question is "does high-end NOA share inflect up before the cash and the multiple run out of patience?"

Tracked forecast (not logged to our model per --watchlist rule): the scoreable base call would be "9660.HK FY2027 revenue ≥ RMB 8.0B, p≈0.55" and "Horizon still below Huawei in China high-end-NOA computing-platform share at YE2026, p≈0.60."

Bull vs Bear

Bull case. Horizon is the default domestic full-stack AD-compute platform for a Chinese auto industry that structurally wants to de-risk off Nvidia. It already owns the low end (47.7%), is up-tiering into NOA ~5x/yr, has a $2.3B-validated VW/Europe optionality via Carizon, a 94.5%-margin software line that can re-mix upward as HSD scales, and a RMB 20B cash fortress that lets it out-spend rivals on R&D for years without financing risk. If it closes the 0.8pp gap to Huawei and wins even a third of a fast-growing high-end tier, the ~10x-sales multiple looks cheap in hindsight. The forward "Agentic CAR SoC" + OS ambition extends the TAM from driving to the whole software-defined vehicle.

Bear case (permanent-impairment risks). (1) It loses the high-end war. Huawei (full-stack, better brand, deeper pockets) leads in high-end NOA and Nvidia owns the premium-import mind-share; Horizon could get boxed into the low-margin basic-ADAS commodity tier — winning the least-valuable part of the market while margin compresses toward hardware economics. (2) Customer concentration + China price war. Top-1 = 37.6%, top-5 = 77.9%; the Chinese NEV market is in a brutal price war, and Horizon's design-win revenue rides its customers' volatile sell-through with zero pricing power over them. (3) Export-control tail. A future Entity-List/1260H listing, or loss of leading-edge foundry / Arm-IP access, would cripple the high-end roadmap overnight — unpriced.

Pre-mortem (18 months out, thesis broke): Journey 6 / HSD design wins ramp slower than headlined, Huawei extends its high-end share, the mix stays hardware-heavy so GM keeps sliding, a flagship OEM customer's model flops and revenue misses, more lockup supply hits a thin HK float — and the stock, already −53%, halves again as the market decides Horizon is "the Chinese basic-ADAS commodity vendor," not "the Chinese Mobileye+Nvidia."

Are multiples too high? At ~10x sales with no profits and a margin-compressing mix, yes on a value basis — it is priced as a high-end-NOA winner it has not yet become. The multiple is a bet, not a cushion.

Contrarian view (what the market refuses to see): the market is anchored on the RMB 10.5B "loss" and the −53% chart and treating this as a cash-burning also-ran. What it under-weights: the burn is 73% non-cash, cash grew to ~$2.9B (7-yr runway), the software line prints 94.5% margins, and it owns the on-ramp (basic ADAS) that feeds the high-end funnel. If Chinese-AD-independence is a durable secular force, the incumbent domestic platform with the biggest install base and the deepest war chest is a rational long-shot on the winner — the debate is entirely "does it win the high end," and the price now embeds heavy skepticism.

Devil's Advocate (short-seller)

Dismantling the bull case.

• The moat is at the bottom, the money is at the top. Horizon dominates the commoditizing tier (basic ADAS) and trails in the tier with pricing power (high-end NOA, where it's #2 behind Huawei). Bulls conflate "market leader" (true, in units) with "profit leader" (false). The 47.7% basic-ADAS share may be a share of a race to zero margin — hardware GM is already 34.5%.
• Revenue concentration is a loaded gun. 37.6% from one customer, 77.9% from five. A single design-loss at BYD/Li Auto, or one flagship model missing volume, and the growth story cracks — with no contractual protection (no take-or-pay).
• The most dangerous competitor bulls underestimate is Huawei, not Nvidia. Huawei brings a full-stack ADS, a premier consumer brand that sells cars, unlimited balance sheet, and a "buy-China" halo — and it's already ahead in high-end share (15.2% vs 14.4%). Nvidia owns the premium-import tier. Horizon is squeezed from both ends.
• Worst capital-allocation optics: R&D at 137% of revenue with no breakeven guidance asks the market to fund a moat on faith; the Carizon/HCT JV losses and the fair-value-of-preferred complexity make the true operating loss deliberately hard to pin.
• What must hold for today's price: high-end NOA design wins ramp on schedule, GM stops falling, Huawei doesn't extend its lead, no export-control action, and the HK float absorbs remaining lockup supply. That's five things, several outside management's control.
• −20-30% growth-disappointment scenario: in the bear path (FY28 ~RMB 8.2B vs bull ~RMB 15.4B — a ~47% gap), a ~10x-sales multiple on a decelerating, low-margin revenue base rerates toward ~4-5x → the equity halves again from already-depressed levels.
• Single permanent-impairment scenario, and plausibility: US export-control listing that severs leading-edge foundry + Arm-IP access. Plausibility: moderate and rising — not base case, but a real tail given the 2025 BIS affiliate-rule expansion and 1260H additions of Chinese tech champions. It would kneecap the high-end roadmap while leaving Horizon stuck in commodity ADAS.

Management Questions (ordered by information value)

1. On China high-end-NOA computing-platform share, you're at 14.4% vs Huawei's 15.2% — what is your concrete plan and timeline to take the #1 position, and what happens to the thesis if you're still #2 at YE2026?
2. Overall gross margin fell 12.8pp as hardware out-grew software. Where does blended GM bottom, and what specifically re-mixes it back up — HSD software attach, or is Horizon structurally becoming a lower-margin chip vendor?
3. Customer concentration was top-1 37.6% / top-5 77.9% at the last disclosure. What are those figures today, and how do you de-risk a single design-loss at a flagship OEM?
4. What is your exposure to US export controls — which foundry nodes and IP (Arm) are single points of failure for Journey 6/HSD, and what is the domestic-foundry (SMIC) fallback for the high-end tier, not just basic ADAS?
5. R&D ran at 137% of revenue. What is the absolute R&D trajectory and the revenue level at which you reach adjusted operating breakeven — give a year.
6. Walk through the Carizon and HCT JV economics: how much loss are they contributing, when do they turn, and how much of consolidated 2025 operating loss is really early-stage JV investment vs core?
7. The VW/Carizon + Europe push — what revenue is realistically contracted vs aspirational over 2026-2028, and does it carry the same 94.5% software margins or OEM-dictated hardware economics?
8. Journey 6 is slated for 100+ models — what is the SOP-to-volume conversion rate you're actually seeing, and how much of that pipeline is high-tier (HSD) vs basic?
9. How do you defend software pricing as Chinese OEMs vertically integrate their own intelligent-driving stacks (à la BYD's in-housing)?
10. The "Agentic CAR SoC" + OS — is this a genuine TAM expansion into the software-defined-vehicle stack, or scope-creep away from winning the AD-compute war you're in?
11. Cash grew to RMB 20.188B while you burned — how much was operating vs financing/working-capital, and what's the true annual cash burn (not adjusted-loss)?
12. What is the replacement risk from Nvidia Thor at the premium tier as it localizes for China — where do you win, where do you lose?
13. On the dual-class / preferred structure — beyond the fair-value noise, what's the fully-diluted share count trajectory and dilutionDilutionIssuing new shares, so each existing share owns a smaller slice of the same company. to public holders over three years?
14. What KPI should investors judge you on quarterly, given EPS is years out — high-tier design wins? HSD shipment share? Software attach rate?
15. If the high-end AD-compute market consolidates to two winners, why is Horizon one of them rather than Huawei + Nvidia?

---