HSAI (Trans): Robot revenue now over 50%; SGI seen reaching breakeven next year

Dolphin Research Trans of $Hesai(HSAI.US) Q3 2026 Earnings Call

I. Key Takeaways

1. Guidance

a. Q4: total revenue of RMB 1.1–1.15 bn. This points to mild downstream demand.

b. FY26: SGI revenue maintained at RMB 200–300 mn, with actuator modules contributing the majority. GPM target is close to 40%, while group opex is expected to rise approx. 20% YoY, with the incremental spend mainly on SGI R&D; ex-SGI, lidar opex declines YoY.

c. 2027: SGI revenue targeted at approx. $100 mn (about RMB 700 mn), contributing 10%+ of total and likely breakeven in the same year. Group GPM is expected to be broadly stable, with clearer guidance to come next year.

2. Quarterly metrics

a. Totals: revenue of RMB 1.068 bn (+34% YoY), marking the 10th straight YoY gain. GPM was 41%, and GAAP profitability has now been sustained for six consecutive quarters.

b. Mix: non-ADAS lidar businesses contributed over 50% of revenue. This reflects a broader revenue base.

c. Segments: lidar revenue was RMB 978 mn with OP of RMB 145 mn (+88% YoY). SGI revenue was RMB 91 mn (+101% QoQ), while SGI posted an operating loss of RMB 83 mn.

3. Profit and one-offs

a. GAAP NP was RMB 81 mn, while non-GAAP NP was RMB 119 mn. The YoY decline was due to a RMB 185 mn one-off last year from equity investment disposal and group IP transfer.

b. Ex the above one-off, GAAP NP rose 14% YoY. This better reflects underlying profitability momentum.

4. Opex and capex

a. Opex rose 46% YoY, driven by SGI. Ex-SGI, lidar opex fell YoY.

b. SGI investment is R&D-led. There is no need for a material increase in capex.

II. Call Details

2.1 Management remarks

1. Strategy: positioned as a full-stack infrastructure platform for robotics and physical AI. Lidar handles 'seeing', Kosmo handles 'understanding', and actuator modules handle 'moving'.

2. Automotive lidar (ADAS)

a. Share: 43% share of China's long-range ADAS lidar in Sep, ranking No.1 for 20 consecutive months. Global share in Robotaxi lidar was 62% (Yole Group).

b. Premiumization: L3 is shifting demand toward longer-range and higher-resolution products. The ET series is priced above ATX, has five awards from a top European OEM, Cadillac, Chery, Great Wall, and a leading NEV maker, and carries 1 mn+ units in backlog.

c. Multi-sensor per car: the FT series blind-spot lidar secured new awards from Leapmotor and a leading European OEM. This supports the multi-lidar-per-vehicle trend.

d. Down-market and overseas: SAIC-GM-Wuling adopted the AT series as the primary lidar. Export programs were awarded by seven Chinese OEMs including Chery.

e. Pricing: ATX pricing and margin details, cf. Q&A. The company emphasized discipline in pricing and quality.

3. Picasso chip

a. A full-color, ultra-high-sensitivity SPAD SoC capturing depth and color at the same timestamp on a single chip. ET series with Picasso is under customer validation, with potential projects progressing with top Robotaxi operators and global OEMs.

b. Partnered with JD, which plans to deploy 4 mn+ robots, Robovans and smart devices over five years. This could expand hardware and software adoption.

4. Kosmo (3D world data engine)

a. Product: globally launched on Oct 14. Kosmo One is the world's first 8K AI spatial camera with millimeter-level precision, while Kosmo World generates 3D assets 5x faster than peer 3DGS solutions and 200x faster than manual modeling.

b. Embodied AI: a single reconstruction can generate 10,000+ training trajectories, with partners like Galaxy General and Dexmo. Management estimates each humanoid robotics firm could need up to 100,000 spaces.

c. Film: building a 10,000 sqm European street set takes two months, but Kosmo completes it in a week at up to 80% lower cost. Kosmo partners with Pixar RenderMan, and is the only capture hardware cited in its official tutorials.

d. Immersive experiences: initial coverage of 1,000 pilot sites with Gaode Maps. It integrates with NVIDIA Isaac Sim, Unreal Engine, RenderMan and Houdini, and recorded first revenue in Q3.

5. Robotic actuator modules

a. Shipments: 23,000+ in Q3 (+97% QoQ), with cumulative shipments of 35,000+. Full-body modules ramped to volume production alongside dexterous hand modules.

b. TAM: a humanoid robot may require 100+ actuator modules, accounting for over half of BOM cost. The potential market could reach $1 tn.

c. Moat: in-house reducers, motors and other critical components. Anchor customer Sharpa uses Hesai modules in its full-body humanoid D-01 and dexterous hand W-02.

2.2 Q&A

Q: Given the market and competition, what is the latest outlook for 2026 and 2027? How do you see volume, pricing and margins?

A: Q4 revenue guidance is RMB 1.1–1.15 bn, with full-year GPM near 40%. Auto demand is softer than expected, while robotic lidar broadly met expectations; higher ASP and margins in Robotaxi and humanoid lidar offset weaker low-ASP lawn-mower lidar.

Quarterly GPM will fluctuate with mix, but full-year GPM is still expected near 40%. With customer validation and commercial demand running ahead of plan, SGI investment is stepping up, taking opex up about 20% YoY; in 2027, ATX pricing and margins should stabilize further, SGI revenue should reach about $100 mn (10%+ of total) and could breakeven that year, with clearer guidance next year.

Q: ASP and GPM improved QoQ in Q3. Will ADAS lidar ASP and margins hold in Q4, and can you split ASP and GPM between auto and robotics for Q3?

A: No ASP or margin breakout by sub-biz; ATX pricing and GPM are guided to be relatively stable in 2027. This is underpinned by signed contracts for next year and a more rational ADAS pricing environment; with core components and ASICs largely in-house, further cost-down headroom narrows, and the company will not trade quality for price.

Blended ASP is mix-driven, and ADAS should be viewed via per-vehicle value: L3 is pushing from a single main lidar to a high-end main lidar plus multiple blind-spot lidars. The group GPM target remains near 40% in 2026 and is expected to be broadly stable in 2027, supported by stable ADAS margins, higher margins in robotic lidar and overseas projects, and Kosmo margins potentially 40%+ at scale.

Q: What is the latest on actuator module scale-up?

A: Monthly capacity has reached 10,000 units, and based on the current pipeline, annual shipments could reach six figures by 2027. Actuator modules account for the vast majority of SGI revenue this year.

There are three drivers: Sharpa continues to ramp and remains the largest customer, with commercial deployments including Dairy Queen ice cream-making. Per-unit value is rising as modules move from dexterous hands to full-body, requiring 100+ modules per robot.

The third is customer expansion: torque and power density are about 3x leading products, and the company is in talks with several AI firms that have announced humanoid strategies, with names not disclosed; it also sees opportunities beyond humanoids.

Q: What are Kosmo's TAM, biz model and near-term order progress?

A: Kosmo revenue is hardware-led this year, with software and asset revenue set to rise significantly next year. No split is provided for now. Revenue will come from Kosmo One hardware, Kosmo World subscriptions and usage-based fees, and 3D asset licensing, with more devices driving more data and recurring software and asset revenue.

Use cases span embodied intelligence (Galaxy General, Dexmo), film (RenderMan), and immersive experiences (Gaode 1,000 pilot sites), with customers in China and overseas. A white paper estimates the humanoid market at about $7.5 tn by 2050 and global media and entertainment at about $2.8 tn.

Q: Can you reconcile one-offs in Q3 this year vs. last year to normalize YoY profit?

A: There was a RMB 185 mn one-off last year; ex that, GAAP NP rose 14% YoY. The one-off came from equity investment disposal and group IP transfer.

Management also explained the variance vs. plan: the shortfall was mainly in ADAS, as several OEMs cut 2026 volume targets, weighing on shipments, which came in slightly below prior expectations. On the robotics side, one customer issue reduced JT-series shipments, but Robotaxi and humanoid lidar exceeded plan, leaving the overall outcome in line; robotics-related revenue (robotic lidar, Kosmo, actuators) already accounts for 50%+ this quarter.

Q: What is the shipment outlook and pipeline for robotic lidar?

A: Robotic lidar accounts for about 50% of lidar revenue, and JT128 shipments could reach six figures next year. Cumulative robotic lidar shipments more than doubled YoY in the first three quarters, and high-end products like Pandar, OT, QT and XT grew revenue by 60%+.

Three factors drive this: L4 Robotaxi fleet expansion is running ahead of expectations, supported by overseas capacity, VDA 6.3 certification in Germany, and an MSCI ESG AA rating. Humanoids use multiple lidars per unit, with JT128 adopted by a top domestic embodied AI player and rising interest from global majors.

Industrial automation, retail and pharma are also key physical AI use cases. Management believes the robotic lidar market is at least 10x the ADAS lidar market.

Q: When will export programs with Chery and six other OEMs reach mass production, and at what scale?

A: No MP timing or volume guidance for export projects; management called them 0-to-1 incremental opportunities. Ten awards have been secured via a JV with global OEMs, which are expected to ramp at larger scale next year.

Q: When will Western OEMs follow Chinese peers in adopting lidar? Any view on Mercedes' L2 plan without lidar?

A: No direct comment on Mercedes' L2; Mercedes L3 projects are progressing per customer timeline. The company is a strategic and awarded supplier for multiple Mercedes L3 platforms in Europe and China and serves as the reference lidar for NVIDIA's Hyperion platform, lowering integration barriers for overseas OEMs.

Overseas ADAS offers a larger vehicle base with higher ASP and margins, and some global OEMs are moving straight to L3 with a high-performance main lidar plus multiple blind-spot lidars. New awards already reflect this architecture.

Risk disclosure and statements:Dolphin Research Disclaimer and General Disclosure