One segment disclosure decides whether 25× sales is the right multiple.

• What 25× P/S is paying for. The market underwrites Horizon as a high-margin software compounder converging to Mobileye-shape economics by FY2027–28. The premium over Mobileye's 3.4× sales sits on a 92%-gross-margin license-and-services engine plus a China-policy-protected merchant slot.
• What H1 2025 actually printed. Consolidated gross margin compressed from 79% to 65.4% as low-margin product revenue grew +250% to ¥778M while the 92%-GM license line grew only +7% to ¥739M. R&D-to-revenue stepped up from 132% (FY24) to 147% — operating leverage running in reverse exactly at the half product mix was supposed to inflect.
• The decider is one disclosure away. Bull reads the compression as mix-shift (hardware GM rose 44.7% → 46.4% inside the ramp). Bear reads it as chip-margin gravity pulling the consolidated number toward Mobileye's 47.7%. The FY2025 segment table — hardware GM plus license-growth — resolves it.

Narrow but real — and asymmetric by segment.

The moat lives in three places: a license-and-services line still printing 92% gross margin after eight years, a Tier-1 distribution model no other Chinese independent has replicated, and a revealed-preference effect — Chinese OEMs refuse to depend on Huawei (a rival OEM via AITO) or on captive silicon from competitors, leaving Horizon in the neutral lane. The moat does not extend into premium urban-NOA, where NVIDIA holds 59.4% and Horizon trails Huawei 14.4% to 15.2%.

Revenue is real. Operating leverage is not — yet.

Earnings-power bridge: ¥1.84B of FY2024 gross profit is consumed nearly twice over by ¥3.16B of R&D, plus ¥1.05B of S&A. The problem is fixed-cost absorption at the bottom of the P&L, not unit economics at the top. FY2024 reported net income of ¥2,347M is the IPO preferred-share conversion gain — the underlying operating loss ran ¥-2.14B, in the same band it has held for four years. Operating cash flow turned a sliver positive (¥+18M) in FY2024 for the first time; free cash flow stayed at ¥-894M as capex stepped to ¥912M for the Journey 6 ramp. The balance sheet funds 5+ years at the current burn at zero net debt.

OEM in-housing already has a documented precedent at the closest analog.

• Zeekr left Mobileye mid-cycle in Q3 2024. A top-10 China OEM abandoned the merchant model for captive silicon — observable proof that the design-win-for-life moat is permeable. Both bull and bear agree this is the named thesis-killer.
• Captive programs are already running across ~70% of Horizon's revenue base. BYD's DiPilot covers most of BYD's ~10M annual production. NIO's main brand runs in-house Shenji. Top-5 customers are 71.8% of revenue. NIO's revealed hedging (Onvo on NVIDIA, Firefly on Horizon, Shenji at home) is what the rest of the top-10 looks like in transition.
• A single top-10 defection collapses the math. The 25× P/S sits on a 60% three-year CAGR guide and a 5-to-7-year design-win base. One Zeekr-precedent repeat compresses FY2027–28 units 8–15% and forces the multiple to anchor on Mobileye's 3.4× sales — not on the consensus HK$11.58 target.

A bombed-out tape arguing with a raised guide.

The setup. Death cross confirmed March 27, 2026 — the first in the listing history. Price sits 24% below the 200-day, 45% below the September 2025 high of HK$11.32, RSI at 31.7, within 3% of the 52-week low. Lower highs in November, January, February dominate the chart.

The crosscurrent. Management raised the three-year revenue CAGR guide from 50% to 60% at the March 19, 2026 FY2025 result; HSD shipments are guided from 22k (2025) to 400k (2026). The board commenced its first-ever buyback on April 9, 2026 at HK$6.23–6.28 — the first observable break from the prior issue-do-not-return capital pattern. Sell-side held: 23 analysts at Strong Buy, mean target HK$11.58.

The honest read. The tape is pricing the H1 2025 reversal as base case. The fundamentals are pricing it as transition. Both reads are defensible until the FY2025 segment disclosure lands.