NVDA budget-alt dream shattered; is Horizon Robotics at risk?

Overall, while$HORIZONROBOT-W(09660.HK) delivered 1H26 revenue and blended GPM at the upper end of guidance, the outperformance was largely pulled by the high-margin 'Licensing & Services' line. The market’s anchor business — Product Solutions (chip sales plus HSD algorithms) — missed meaningfully, with both revenue growth and margin deteriorating. The headline met guidance, but core quality was weak, as detailed below:

1. Revenue met guidance, but heavily reliant on licensing: 1H26 total revenue reached RMB 2.06 bn, up 33% YoY. Growth was almost entirely driven by a beat in Licensing & Services, while the core Product Solutions slowed sharply.

2. Core product business slowed; the 'volume + ASP up' thesis broke: Product Solutions revenue was only RMB 930 mn in 1H26, with YoY growth collapsing from 85% in 2H25 to 15%. It even fell short of the company’s lowered full-year guidance of 85% growth for this segment.

Shipments under pressure: Chip shipments were 2.218 mn units in 1H26, up just 12% YoY. Management cut the full-year shipment guide from 5.42 mn to 5.00 mn units (implying ~25% YoY).

The slowdown stemmed from two constraints. First, China retail demand for passenger vehicles was soft in 1H26 (-20.2% YoY), and second, key customer BYD underdelivered and unveiled its in-house high-compute chip 'Xuanji A3' (three-chip cluster >2,100 TOPS), stoking concerns that Horizon’s AD share could be squeezed.

ASP stalled: 1H26 ASP was ~RMB 417 per unit, up only 2.4% YoY. This indicates the high-ASP flagship J6P has not scaled, with shipments still dominated by the Mono series (J2, J3, J6B) in mid/low-tier chips.The long-expected 'domestic substitution via J6P driving both volume and ASP up' failed to materialize in 1H26.

3. Licensing & services was the lone bright spot: Revenue reached RMB 1.02 bn in 1H26, up 90.4% YoY and far above the market’s ~RMB 720 mn expectation. Beyond ongoing contributions from the JV with Volkswagen Group China (Cool Core Cheng), Horizon won a global top-tier Tier-1, DENSO (servicing Toyota), lifting it into the top-5 customers for 2025.

The Toyota program launched mass production in 1H on GAC Toyota’s highest-volume entry model. Via DENSO’s mainstream platforms, volume contributions are expected to start the year after next.

4. Hardware GPM fell on 'bundled sales': Product Solutions GPM was 36.2% in 1H26, down 600 bps YoY and well below the market’s 46.7% expectation. To accelerate early HSD mass-production ramps, the company bundled domain controllers with core products, diluting hardware margins.Excluding this one-off impact, GPM would have been 48.1%, up 300 bps YoY.

Additionally, while DRAM is in an upcycle, the company locked in pricing by end-2025. Memory cost swings in 2026 are not expected to further erode margins.

5. Blended GPM lifted by 'mix shift' rather than cost-down: Overall GPM reached 66% in 1H26, beating the market’s 63.2%. This was not driven by manufacturing cost efficiencies, but by a larger revenue mix from Licensing & Services, which carries very high margins (up to 90.4%).

6. 'Saturation-level' R&D spend stayed elevated: R&D expense reached RMB 2.76 bn in 1H26, up 22% YoY, with R&D ratio surging to 134%. Funds went mainly to cloud compute for rapid HSD algorithm iteration, and to tape-out and development of the next-gen strategic SoC 'Journey 7'.The new architecture is optimized for larger-parameter HSD models and native on-device deployment of the 'Kaka Shrimp' cockpit agent model.

7. Headline profit surged on paper, but cash burn continued:

Operating losses widened: 1H26 OP loss was RMB 1.67 bn, up 11.1% YoY. Adj. net loss, excluding one-offs, was RMB 1.67 bn, up 25.4% YoY and worse than the market’s -RMB 1.38 bn estimate, underscoring weak cash-generation in the core business.

Net profit turned positive purely on one-offs: Reported net profit was RMB 3.78 bn in 1H26 (vs. a -RMB 1.62 bn expectation). This was not operational improvement, but two large non-cash one-offs.

CARIAD convertible revaluation: Due to share price volatility, the company recorded a RMB 5.241 bn fair value gain on financial liabilities related to the convertible loan issued to Volkswagen Group’s CARIAD.

D-Robotics deconsolidation: Control was lost as of Mar 31, with D-Robotics becoming an associate. The company recognized a RMB 2.779 bn one-off revaluation gain, of which RMB 2.169 bn was included under discontinued operations profit.

Dolphin Research view:Overall, another somewhat 'skinny' print. Total revenue hit the guided upper bound, but mix relied heavily on less certain licensing, while the hardware anchor stalled, diminishing the quality of earnings.

The market’s core hope — high-end chips scaling (notably J6P) to drive both volume and ASP — did not show up this quarter. Mid/high-end penetration remains low, with shipments still centered on the Mono series (J2, J3, J6B), leaving ASP growth near flat and product mix upgrades behind expectations.

Profitability is also under pressure. GPM declined, and while management attributed this to early HSD ramp support via bundling domain controllers with core products, the core pain point remains the slow scale-up of high-end chips like J6P.

Coupled with surging R&D and widening OP losses, market patience for delivery on guidance is being tested. The cash buffer remains ample at ~RMB 14.8 bn, which buys valuable time for a push into the high end.

The market focus has shifted squarely to 2026 guidance, especially the ramp of J6P (560 TOPS). Whether this flagship, bearing the domestic HSD substitution narrative, can capture enough share will determine whether the stock de-rates or merely consolidates valuation.

However, management cut full-year 2026 guidance materially, signaling caution at the print:

① Revenue: full-year guide cut; growth slows further

Management lowered 2026 revenue guidance from RMB 5.9 bn to RMB 5.0 bn, taking YoY growth from 57% down to 33%. By component:

a. Shipments:Full-year chip shipments were cut from 5.40–5.50 mn to 5.00 mn units (implying +26% YoY). Dolphin Research attributes this to softer NEV passenger car growth downstream, BYD’s domestic shortfall and in-house AD chips crowding share, and slower-than-expected J6P ramp.

Implied cadence suggests 2H26 shipments of ~2.78 mn units, +37% YoY, a step-up vs. 1H26’s +12%. This indicates a back-half weighted ramp and deferred volume pressure.

More cautiously, Dolphin Research expects actual 2026 shipments around 4.90 mn units (+~23.5% YoY). The structure is estimated as follows:

Low-end ADAS (J6B/J2/J3): Demand keeps shrinking, with shipments likely falling below 2.0 mn units (estimate: 1.90 mn).

Mid-end ADAS (J6E/M): Management had been upbeat, guiding >3.0 mn units (with BYD and Geely contributing ~1.0 mn each and targeting BYD’s 'Shenyan B' AD system in 2026). Given the lowered overall guide and OEM in-house risk, Dolphin Research now expects ~2.80 mn units.

High-end HSD (J6P, etc.): Management guided 0.30–0.40 mn units, led by Chery Fengyun, iCAR and Jiyue V27. In a heated HSD race, Dolphin Research expects ~0.20 mn units, with significant ramp challenges.

HSD mix: management previously guided that over half of HSD volume will come from 'HSD + J6P' (mainly nine Chery models), with the remainder from J6M.For HSD (estimated ~0.40 mn units in 2026), official nominations are progressing well (20 mass-production nominations), covering China’s top-selling OEMs. Horizon says talks with leading OEMs on HSD are advancing well.

There are three HSD delivery modes. J6P solution: total ASP USD 700 (USD 500 chip + USD 200 software), targeting ~RMB 150k vehicles.

Dual J6M solution: total ASP USD 400 (USD 200 chip + USD 200 software), targeting RMB 120–130k vehicles.

Single J6M solution: pricing under negotiation, targeting ~RMB 100k vehicles.

b. ASP: implied cut; high-end mix-up slower than plannedBack-solving management’s latest guide of RMB 5.0 bn revenue, 5.0 mn units, and flat Licensing & Services YoY implies Product Solutions must grow 85% YoY to RMB 3.0 bn, with core chip ASP around RMB 600. This is clearly lower than the prior implied near-RMB 700.

The ASP reset exposes weaker-than-expected mid/high-end mix. In particular, J6P (560 TOPS), central to the 'volume + ASP up' thesis, is ramping slowly, even though its chip ASP is USD 500 — nearly 10x the 2025 Avg. ASP of USD 56.On a more conservative view, Dolphin Research expects a 2026 overall chip ASP of ~RMB 524, still below the RMB 600/unit management target.

② Margins: multiple levers to defend a 60%+ baseDespite fierce competition and memory cost pressure, management aims to keep blended GPM above 60%, anchored by three levers. First, revert to a high-margin model: system-level hardware delivery (domain controllers) is a transitional tactic for early HSD ramps, and 2H26 may see a shift back to a 'SoC + software licensing' model with near-100% margins.

Second, hedge supply chain swings: DRAM prices were locked by end-2025, limiting cost noise this year. Third, replace price wars with architecture innovation: a cockpit-driving integrated scheme based on the 'Starry' chip merges cockpit and AD memory systems, saving thousands of RMB per car (memory, harness, cooling, PCB, etc.).

The company seeks to help OEMs cut costs via technology rather than SoC price cuts. This solution is slated for mass production in Q4 this year, with a very short cycle from nomination to SOP.

Valuation & call: wait for a safer 'hitting zone'Horizon remains a scarce domestic 'software + hardware' AD chip play. But J6P, the high-compute product underpinning the 'NVDA alternative' narrative, has not ramped as expected, limiting near-term realization of the substitution story in earnings.

Dolphin Research models 2026 revenue at RMB 4.7 bn, with growth slowing from 58% in 2025 to 25% in 2026. On a current market cap of ~RMB 63.0 bn, that implies a 2026 P/S of 13.4x.

This is above NVDA’s forward P/S of ~12–13x, despite NVDA’s far stronger growth and profitability base. With J6P yet to scale and limited proof of durable moats, the valuation looks stretched.

Competition also intensifies in the value high-end band, with Momenta and others bringing in-house 'chip + algorithm' stacks in 2H26. The backdrop is turning tougher.

Bottom line, until J6P shows up in the P&L at scale, upside optionality appears limited from here. For a better risk/reward and margin of safety, Dolphin Research suggests a safer entry at RMB 47–50 bn market cap (about 10x P/S).That implies 21%–25% downside from current levels, corresponding to RMB 3.5–3.7 per share, offering a more favorable setup.

Earnings exhibits:

Dolphin Research deep dives on Horizon:

《Is the 'NVDA alternative' story for Horizon overheating?》

《Horizon: can a domestic AD 'substitute chip' unseat NVDA?》