Variant Perception

Figures converted from CNY at historical FX rates — see data/company.json.fx_rates. Ratios, margins, and multiples are unitless and unchanged.

Where We Disagree With the Market

The sharpest disagreement is on duration: the market is paying for a 60% three-year revenue CAGR and a FY2027 breakeven simultaneously, but the evidence in the report says those two things cannot both happen — R-and-D is the input to growth, and management's own roadmap (HSD 18x scaling to 400k units in 2026, Riemann BPU, Journey 7 tape-out, Q3 2026 robotaxi pilot) keeps R-and-D growing alongside revenue through FY2027. The 23-analyst Strong Buy with a HK$11.58 mean target reads as an underwriting of "growth AND leverage" on the same calendar. The honest read is "growth OR leverage by FY2027." On a slipping-breakeven path, the current 25x P/S compresses even if the bull is right about the trajectory, and the asymmetry is much narrower than the +85% headline upside implies. The single observable that resolves this is whether H1 2026 R-and-D outgrows revenue for a third consecutive half — the late-September 2026 interim is the decision date.

Native reporting currency is CNY ($537M FY2025 in USD equivalents below). Trading happens in HKD (HK$) on HKEX. Currency symbols are used throughout as appropriate; ratios, multiples, and percentages are unitless.

Variant Perception Scorecard

The score is deliberately bounded. Variant strength is 68, not higher, because consensus is not uniformly wrong — the market correctly prices Horizon as the dominant Chinese ADAS merchant with a real Tier-1 distribution moat and a credible long-term TAM. Where consensus is wrong is on timing and on which financial variable resolves the thesis. Consensus clarity is 72 because the spread is observable: 23 analysts, HK$8.29 low / HK$15.30 high (1.85x range), Goldman raising target to HK$15.30 while removing the name from its APAC Conviction List, and Morningstar fair value HK$4.12 — sell-side is bullish on direction, fragmented on magnitude, and quietly de-rating in conviction. Evidence strength is 70 because the disagreement is anchored to data in the report (H1 2025 R-and-D-to-revenue stepping up to 147% from 132%, license growth decelerating to +7% YoY, the CFO inflection mechanically explained by interest income and subsidies) rather than to a contrarian narrative.

Consensus Map

The consensus is not uniform — it is anchored by 23 analysts at Strong Buy with a HK$11.58 mean but already shows three observable cracks: Goldman removing the name from its APAC Conviction List in May 2026 while raising the target, Morningstar's fair value at HK$4.12 (the stock 79% over fair value), and Alpha Spread at HK$5.01 (20% overvalued). The disagreement that follows lives in the gap between sell-side conviction and third-party DCF anchors.

The Disagreement Ledger

Disagreement #1 — Breakeven slips to FY2028+ on the math of 60% growth meeting Riemann/J7/HSD/robotaxi R-and-D scale. Consensus would say the raised 60% three-year CAGR proves operating leverage is in line because higher revenue absorbs the fixed R-and-D base. Our evidence disagrees: management raised the growth guide because it is investing harder, and the H1 2025 print showed R-and-D-to-revenue stepping back up to 147% from 132% — at exactly the half product mix supposedly inflected. If we are right, the consensus has to concede that 60% growth and FY2027 breakeven are alternatives, not complements; the HK$13-15.30 high-target band requires both. The cleanest disconfirming signal is H1 2026 R-and-D growth rate versus revenue growth rate: if R-and-D outgrows revenue a third consecutive half, breakeven slips to FY2028+, and sell-side mean compresses toward HK$9.

Disagreement #2 — License & services growth is the actual fragile pillar, not hardware GM. Consensus debates product-solutions gross margin in late-September 2026 because that resolves the bull-bear "mix-shift vs chip-gravity" debate. Our evidence says the better question is whether 92%-GM license revenue resumes growth above 25% YoY — because the entire EV/Sales premium over Mobileye rests on Horizon being treated as a software compounder rather than a chip cyclical. H1 2025 license growth of only +7% YoY against the prior trend of +71% is the leading edge of a structural deceleration: the SOP front-load wave for Journey 5/6 has crested, and a Journey 7 wave is not yet billing. If license growth stays below 15% YoY in H1 2026, the IP/algorithm revenue line reprices to a one-time-program-launch model and Horizon converges to chip-cyclical economics regardless of what hardware GM does.

Disagreement #3 — OEM in-housing is the base rate, already running, not a tail risk. Consensus prices a near-zero probability of a top-10 China OEM publicly committing to captive silicon for a flagship NOA program in 2026-2028 — that is what a 25x P/S on $537M FY2025 revenue means. Our evidence: BYD DiPilot already covers most BYD production, NIO Shenji runs on the main brand (Horizon only ships in Firefly), XPeng Turing AI Chip is rolling out, and Li Auto has a proprietary silicon program — captive programs are already operating across roughly 70% of Horizon's revenue base. Mobileye losing Zeekr in Q3 2024 is the analog precedent in the same industry. If we are right, consensus has to concede that the probability of one more flagship defection inside 2026-2028 is closer to 40-50%, not under 10%. The disconfirming signal is the May-September 2026 OEM model-launch and tech-day cadence; the J7 design-win wave that lands or does not by late 2026.

Disagreement #4 — The FY2024 CFO inflection that bull notes cite was interest income and subsidies, not operating leverage. Consensus reads the +$2.4M FY2024 CFO turn as the cleanest evidence operating leverage has arrived. Our evidence: $52M interest received on the post-IPO cash pile plus $27M government R-and-D subsidies plus $31M customer prepayments delivered the swing; strip the interest line alone and FY2024 CFO is -$50M, not +$2.4M. If we are right, the market has to concede that the operating-leverage narrative is being read off a non-recurring set of inputs and that H1 2026 CFO under further capital injection (no new placement) likely turns negative again. The resolution is mechanical — H1 2026 CFO stripped of finance income, against H1 2026 operating loss.

Evidence That Changes the Odds

The strongest single piece of evidence is item #1 — H1 2025 R-and-D-to-revenue rising to 147% from 132% — because it is observable, recent, and points in the wrong direction for the operating-leverage thesis. The most fragile is item #8 (talent attrition) because it cannot be verified from disclosure. The most important pair is #2 and #6: the license-line deceleration and the FY2025 adjusted operating loss widening tell the same story from two different angles — operating leverage is not yet visible in either the revenue mix or the cost structure.

How This Gets Resolved

The single most decisive resolution is signal #1 — H1 2026 R-and-D growth rate versus revenue growth rate. It is observable, dated, and directly tests the load-bearing assumption of the consensus thesis (operating leverage arriving on the FY2027 calendar). The single signal that resolves the durable thesis variable (not just the near-term print) is #3 — any one of the top-10 China OEMs publicly committing to captive silicon for a flagship NOA program. That signal is binary, observable in the May 2026 through end-2028 OEM model-launch cadence, and converts the underwriting from "narrow but real moat" to "moat not proven" — at which point the 25x P/S compresses regardless of what hardware GM does.

What Would Make Us Wrong

The strongest counter-evidence to the breakeven-slip view is that management raised the three-year revenue CAGR guide to 60% knowing exactly what the R-and-D envelope looks like through FY2027 — Horizon's CFO and CEO both signed off on a guidance number that implies operating leverage arrives, not slips. If management has internal visibility we do not (Tier-1 confirmation that the J7 design-win wave is bigger than externally visible; HSD attach holding above 80% on launch SOPs; subsidy renewals through FY2027), the H1 2026 print could surprise to the upside on revenue and R-and-D simultaneously, in which case the consensus mean is the right anchor and the variant view is just a duration concern that resolves itself inside one half-year. Founder-CEO Dr. Kai Yu has met five of his first six concrete public-company promises (10M cumulative Journey shipments, HSD Q3 2025 mass production with Chery, 4M+ FY2025 unit shipments, license-and-services scaling, related-party transaction caps observed); the one missed promise (CARIZON ramp) was disclosed flatly. That execution track record is the strongest reason to think the consensus operating-leverage call is not naive.

The strongest counter-evidence to the license-deceleration view is that the FY2025 full-year license number recovered to +17.4% YoY ($271M from $226M), which is consistent with H1 2025 being the air-pocket low rather than the new structural level. If the J7 design-win wave lands in H2 2026 with named flagship Chinese OEMs (the disclosure path is the FY2025 H2 results and FY2026 H1 segment notes), license growth re-accelerates above 25% YoY and the software-attach economics survive intact. The bull frame on Mobileye-premium economics holds.

The strongest counter-evidence to the OEM in-housing view is the revealed-preference hedging pattern at NIO (Shenji on main brand, Onvo on NVIDIA, Firefly on Horizon) and the structural coordination problem that keeps Chinese OEMs refusing dependence on Huawei (a rival OEM via AITO) or on rival captive silicon. Captive programs at BYD, NIO, XPeng and Li Auto have been operating for two-plus years without one publicly committing to captive silicon for a flagship NOA program — the absence is itself evidence the merchant lane is structurally defended. If two more reporting cycles pass without a top-10 defection, the base-rate read weakens; absence by end-2028 confirms the bull frame.

The strongest counter-evidence to the CFO-quality view is that interest income on the post-IPO cash pile is real cash and persists while the balance holds — the IFRS-15 classification is conventional, not aggressive. If management uses the buyback authority to consume the cash pile, the interest line shrinks proportionally, but in either direction the cash is the company's. The forensic point is real but its materiality is bounded.

The first thing to watch is the H1 2026 R-and-D growth rate versus revenue growth rate in the late-September 2026 interim — that single comparison resolves whether the consensus 60% CAGR plus FY2027 breakeven is mathematically intact or whether the duration slip we are calling has crystallised.