禾賽集團 (HSAI) 2026 年第二季法說會：上調 SGI 財測指引

重點摘要

• 2026 年第二季營收年增約 22% 至人民幣 8.61 億元，標誌著禾賽集團 (Hesai Group) 連續第九個季度實現營收年成長。
• GAAP 淨利成長 60% 至人民幣 7,100 萬元，為該公司連續第五個季度實現 GAAP 盈利。毛利率為 40%。
• 光達出貨量成長近 80% 至超過 628,000 台。ADAS 光達出貨量超過 485,000 台，而機器人光達出貨量增加至近三倍，突破 142,000 台。
• 戰略成長倡議 (SGI) 在機器人致動模組的帶領下，首次貢獻了人民幣 4,500 萬元的營收。禾賽將 2026 年 SGI 營收指引從人民幣 1 億元上調至人民幣 2 億元至 3 億元。
• 管理層預計 2026 年第三季營收為人民幣 11 億元至 11.5 億元，光達出貨量約為 800,000 至 850,000 台。
• 禾賽重申 2026 全年光達出貨量指引為 300 萬至 350 萬台，並預計全年毛利率將保持在接近 40% 的水準。

關鍵財務結果

業務與營運表現

禾賽在該季度出貨了超過 628,000 台光達，年增近 80%。ADAS 光達出貨量成長約 60% 至超過 485,000 台。機器人光達出貨量成長約 193% 至超過 142,000 台。

在汽車光達方面，禾賽的 ETX 超遠距離光達獲得長城汽車的定點合作，預計於 2026 年底開始量產。長安汽車和其他車企選擇禾賽合作多光達項目，而理想汽車的 L8 和 L9 車型已進入量產，每輛車搭載四顆禾賽光達。該公司還提及獲得福斯汽車中國合資企業多款車型的定點，並擴大了與廣汽豐田的合作。

管理層引述法說會中提及的第三方市場數據表示，禾賽在 2026 年 6 月佔據中國遠距離 ADAS 光達市場的 44%，並已連續 17 個月排名第一。

機器人需求涵蓋人形機器人、割草機器人、自動駕駛計程車 (robotaxi)、自動駕駛廂型車 (robovan) 和倉儲應用等廣泛領域。禾賽與全球超過 50 家具身人工智慧 (embodied AI) 公司合作。管理層維持 2026 年機器人光達出貨量超過 500,000 台的預期，相比之下 2025 年約為 240,000 台。

SGI 的商業化在該季度有所加速。機器人致動模組開始產生營收，截至第二季末累計出貨量已超過 10,000 台。禾賽正朝著近期每月產能約 10,000 個模組的方向提升，並預計 2027 年出貨量將達到六位數。全身模組預計於 2026 年下半年開始量產。

Cosmo 原型機已於 2026 年 7 月出貨，並在七天內獲得初始訂單。管理層預計 Cosmo 將在第三季開始貢獻營收。已有超過 200 家跨機器人、電影、遊戲、旅遊及其他應用的潛在合作夥伴與禾賽聯繫。隨著部署規模擴大，該公司計劃透過雲端處理和 3D 空間資產授權來建立經常性營收。

管理層指引

管理層預計致動模組將貢獻 2026 年 SGI 的大部分營收。Cosmo 的貢獻預計較小，因為其商業化在今年較晚時候才開始，但公司預計隨著時間推移，其雲端服務和資產授權的比重將會增加。

風險與關注領域

• 由於禾賽在產品研發、商業化與產能方面進行投資，SGI 錄得人民幣 6,400 萬元的營運虧損。
• Sharpa 是目前致動模組需求的主要來源。擬將與 Sharpa 的年度交易上限從人民幣 1 億元提升至人民幣 3 億元的提議，仍有待股東特別大會批准。
• Cosmo 仍處於商業化早期階段，許多潛在客戶仍在進行原型機測試與試點項目。
• 汽車光達面臨價格壓力、競爭以及客戶多方採購。管理層表示，並不打算僅為了提高市場占有率而犧牲價格。
• 禾賽表示，美國 FCC 近期的行動目前對其機器人業務沒有重大直接影響，但公司將持續監控相關法規並打算保持合規。

分析師問答亮點

管理層將 Sharpa 描述為禾賽致動系統的實境試驗場。在禾賽供應更廣泛的機器人市場之前，現場部署數據可為可靠性、外形尺寸、工作週期和效能等產品需求提供參考。

關於 SGI 上調指引，管理層將此次修正主要歸因於商業化程序比預期更快以及對致動模組的需求。隨著雲端處理、訂閱和 3D 資產授權規模擴大，預計 Cosmo 將在 2027 年成為更顯著的貢獻來源。

在毛利率方面，禾賽維持全年毛利率接近 40% 的展望。管理層表示，單車光達價值量提升、成本降低、機器人光達及海外業務應能支撐集團的毛利率表現。長期來看，公司預計 SGI 的毛利率將超過 40%，其中 Cosmo 的雲端服務部分在結構上擁有高於硬體的毛利率。

管理層還強調，汽車領域的商機正從單個光達單元的價格轉變為每輛車的總光達價值量。管理層表示，結合 ETX、ATX 和 FTX 產品的高階配置可能代表每輛車約 500 至 1,000 美元的光達價值量。

完整法說會逐字稿

---

完整財報電話會議逐字稿

管理層陳述

Operator

Hello, ladies and gentlemen. Thank you for standing by. Welcome to the Hesai Group's Second Quarter 2026 Earnings Conference Call. [Operator Instructions] Please note that today's conference call is being recorded. I will now turn the call over to our first speaker today, Yuanting Shi, the company's Head of Capital Markets. Please go ahead.

Yuanting Shi

Thank you, operator. Hello, everyone. Thank you for joining Hesai Group's Second Quarter 2026 Earnings Conference Call. Our earnings release is now available on our IR website at investor.hesaitech.com, as well as via Newswire services.

Today, you will hear from our CEO, Dr. David Li, who will provide an overview of our recent updates. Next, our CFO, Mr. Andrew Fan, will address our financial results before we open the call for questions.

Before we continue, I refer you to the safe harbor statement in our earnings press release, which applies to this call as we will make forward-looking statements.

Please also note that the company will discuss non-GAAP measures today, which are more thoroughly explained and reconciled to the most comparable measures reported under GAAP in our earnings release and SEC filings.

With that, I'm pleased to turn over the call to our CEO, Dr. David Li. David, please go ahead.

Yifan Li

Thank you, Yuanting. Hello, everyone. Thanks for joining us. Let me start with the headline. The second quarter of 2026 marks a major turning point for Hesai. We are opening a truly exciting new chapter in our history. Over the past decade, we have built the technology, manufacturing engine and commercial scale required to lead the lidar industry. Today, we are setting our sights on the much larger opportunity, expanding Hesai into a full-stack infrastructure platform for robotics and physical AI, empowering them to see, understand and act. The simplest way to think about Hesai now is 3 layers. First, see our Lidar less intelligent machines see the physical world, what things are, where they are, how fast they are moving and how the theme changes around them. Second, understand cost molten real environments into reusable AI-ready 3D spatial assets, so intelligent machines learned from the world instead of guessing at it. Third, act, our robotic actuation modules turn intelligence into precise physical motion. We are already seeing commercial momentum across all these layers. Our core lidar business keeps scaling profitably and generating strong cash flow. Cosmo shipped prototypes in July 2026 secured initial orders and remains on track to contribute revenues in the third quarter of 2026.

Actuation modules started generating revenues in the second quarter of 2026 and are ramping fast. Together, these 3 layers form an integrated platform, opening up a significantly larger addressable market than we saw a year ago.

Let me start with see. ADAS was our first big commercialization market and it's still a phenomenal cash engine. Think about it. A modern intelligent car is arguably the most widely deployed robot on earth. And the industry has moved beyond the question of can the car drive itself. The question now is, whether it can do it safely every single time with backup when something fails. That's a completely different bar. Regulation is catching up fast. In China, the first mandatory national safety standards for Level 3 and Level 4 will take effect in 2027. In the U.S. proposed New Jersey legislation would require commercial autonomous vehicles to run 2 independent sensing technologies beyond cameras, typically lidar and radar. So the vehicle still sees when the cameras don't. This is what we have been saying for years, lidar is the invisible airbag.

You never think about your airbag on a normal drive. But the one time you need it, it is the difference between a bad day and a catastrophe, and nobody removes airbags to save bucks. So the market is shifting from does it have lidar to better lidar and more lidar. Better means longer range, higher resolution, rock solid reliability where it matters. This quarter, we won a design win with Great Wall Motor for mass production programs using our ultra long-range high-end ETX lidar with SOP expected in late 2026. ETX extends detection range to more than twice that of ATX while ATX has already established itself as the benchmark for cost-efficient lidar. More means multi lidar, phone coverage and higher redundancy. Momentum on our FTX blind spot lidar is strong. Changan and other leading OEMs picked Hesai for multi-lidar programs and Lee Auto L8 and L9, each equipped with 4 Hesai lidars are now in mass production. Li Auto's newly launched L6 model, price to RMB 250,000 now offers an optional 4 lidar configuration, 4 lidars at a mainstream price point. That is the moment an advanced feature goes mass market, and that is how content per vehicle compounds.

Globally, we are accelerating too. Following Mercedes-Benz, we won a major design win with Volkswagen covering multiple models from its China joint venture brands and expanded collaboration with GAC Toyota. We are also riding along with Chinese automakers that are going global with meaningful overseas volume expected in 2027.

Our leadership is also reflected in third-party market data. For [indiscernible], we took 44% of China's long-range ADAS lidar market in June 2026 and have been #1 in China for the long-range ADAS lidar market for 17 straight months.

Now here is where it gets fun. That same demand for high-performance 3D sensing is expanding rapidly into humanoid and other intelligent machines. Morgan Stanley estimates that robots could deploy roughly 6x as many lidar sensors as automotive by 2050. And you can only drive one car at a time, but in the future, you might interact with many different robots in a day.

We are already leading this trend. We work with more than 50 embody AI companies worldwide with recent orders coming from Unitree, Robbie And, Gauba, Galacia, Dexmal and many others. Our JT128 has become the go-to lidar for humanoid and quadruped robot. And the numbers are, frankly, remarkable. Robotics lidar shipments in the second quarter almost tripled compared with the same quarter last year following year-over-year growth of 47% in 2023, 67% in 2024 and 426% in 2025. And we expect another 2 to 3x shipment increase in full year 2026.

The next leap is going from geometry to mini. Classic lidar gives you the skeleton of the world distance, shape and position. A robot needs to know what a thing is, not just where it is. That is Picasso, our full colored ultrasensitive fixed DespatSOC. People love to frame the debate as camera versus lidar. That is like asking a human to choose between color and death. You need both. Picasso fuses them on one chip. We are pleased to share that the Picasso SOC is now SOP ready. Since its launch in April 2026, the full color ETX powered by Picasso has secured initial design wins, including cargo bot. It is now undergoing customer validation and advancing into RFI and RFQ discussions with leading robotaxi operators and global automakers.

Because depth and color come off 1 chip under 1 time stamp, you get intrinsically time aligned multimode information, exactly the fuel world models need. We believe Picasso will be the eyes of the next generation of robots.

Then layer 2, understand, helping robots understand and learn from their environments. This is where Cosmo comes in. Cosmo is a spatial intelligence platform that integrates an AI spatial camera, AI algorithms, 3D spatial assets and cloud services into 1 unified system for capturing, reconstructing and understanding the physical world. It addresses the SIM to real gap bottleneck. Robots need enormous digital representations of the physical world that are geometrically accurate and physically grounded. Training on bad inputs is like teaching a kid from a blurry textbook, plenty of pages, but they may learn the wrong things. Cosmo makes the textbook sharp. It turns real environments into high fidelity, editable interactive 3D assets, reality itself as the training ground. That is exactly the foundation robotic scaling loss need without the model collapse risk from low-quality inputs.

Validation came fast. Pro types went out this July. And within 7 days, we had orders from a lineup of leading humanoid robotics companies, including Galbot. Customer feedback has been extremely encouraging. Cosmo reconstructs a 200 square meter restaurant at roughly 5x the efficiency of a leading alternative with such high fidelity that even 4 millimeters text on a menu remains clearly visible. Scale that across thousands of scenes and you get the generalization robots need to walk into an actual home.

Cosmo also goes well beyond robotics, cultural tourism film and TV, games, advertising and beyond. Imagine standing on stage at a concert or on the field for historic sporting moment. Cosmo turns real places into reusable digital assets. Since the April preview this year, more than 200 prospective partners have come to us and initial revenues are expected to land in the third quarter within SGI.

But the part which I am most excited about is the business model of Cosmo. The AI spatial camera is the front door. Behind it is proprietary algorithms, cloud services and a compounding library of 3D spatial assets. Every deployment adds assets. A richer library enables more applications, more applications attract more users, more users create more assets. That is a flywheel, recurring revenues, network effects and real operating leverage. Cosmo doesn't digitize a space once, it compounds the value of that space over time.

Layer 3 is act, the muscles. JP Morgan sees 2026 as a milestone year for humanoid. Actuation modules can be more than half a robot [indiscernible] and 1 robot may need more than 100 modules. Run that math and long-term demand could approach 1 trillion units, 1 trillion. It is also brutally hard engineering. You need strength and precision in something compact, efficient, durable and consistent like asking a pro athlete to run a marathon every day for years without losing a step. This is exactly our wheelhouse.

For more than a decade, we have built precision electromechanical systems, in-house chips, material science, thermal engineering and automated manufacturing. People think lidar is a sensor, but it is not. It is a precision machine where optics, electronics, software, motors and encoders stay in perfect sync through heat, cold and vibration for years. The physics of a lidar scanner and a robot joint are cousins, both demand precise control of position, speed and force. Going from helping robots see to helping robots move is the most natural extension we have ever made. We rebuilt the entire actuation stack from first principles, materials, structure and system integration. The result is a breakthrough in actuation modules, roughly 3x the torque density and power density of leading products available today in a package 37% smaller with transmission efficiency above 95% and performance validated through 2 million operating cycles.

We deliberately started with the hardest system in humanoid robotics, the dexterous hand. Walking get the robot to the workstation, hands or what makes it useful once it arrives. Starting with this demanding hand application allowed us to validate our architecture at the highest level before extending it across the rest of the body. That strategy is already converting into revenues.

We are supplying actuation modules to Sharpa, a global front-runner in AI robotics. In August 2026 in Shanghai at a Dairy Queen store, a Sharpa humanoid is expected to complete what we believe is the world's first 0 retrofit commercial deployment of its kind. The robot is set to autonomously work a full TQ shift preparing signature blizzard ice cream treats, a complex long horizon challenge, while demonstrating the potential to generalize across diverse real-world tasks. That is the whole point. Robotics is not about impressive lab demos. It is about useful work done repeatedly and reliably in a real operating environment. And as Sharpa earns market recognition and its backlog grows, our volume grows with it.

At the same time, real-world field experience feeds directly back into engineering, creating another powerful flywheel for our product development. Our dedicated production line is fully operational. Cumulative shipments exceeded 10,000 modules by the end of the second quarter. We are ramping towards roughly 10,000 modules per month in the near term and expect 6-digit volumes in 2027.

Meanwhile, full body modules SOP is expected in the second half of 2026, expanding into shoulder and wrist joints.

Beyond robotics, these modules fit anything needing dense, precise, reliable motion. This is only the beginning. As intelligent machines proliferate across industries, we believe the addressable market for high-performance actuation will expand dramatically.

Let me close by returning to the bigger transformation we are witnessing today. If digital AI gave machines of voice and a mind, physical AI is giving them eyes to see, a body to move and the ability to act in the real world. Hesai is becoming a full stack infrastructure platform for robotics and physical AI, empowering them to see, understand and act so as to power the AI-driven fourth industrial revolution.

With that, I will turn it over to Andrew for our financials and outlook. Andrew, please go ahead.

Peng Fan

Thank you, David. Hello, everyone. The second quarter of 2026, once again demonstrated our ability to deliver solid growth and scale while investing in Hesai's next chapter. As we introduced last quarter, we now manage and report our business through 2 segments: our core lidar business and our strategic growth initiatives, or SGI, giving investors greater visibility into their distinct financial profiles and the growth trajectories.

With that framework in mind, let me begin with our consolidated financial performance. Total net revenues for the quarter reached RMB 861 million or USD 127 million, representing an increase of approximately 22% year-over-year. The second quarter of 2026 marked our ninth consecutive quarter of year-over-year revenue growth.

Gross margin remained healthy at 40%. On operating expenses, we remain disciplined while continuing to invest in long-term growth.

Sales and marketing expenses were RMB 50 million, General and administrative expenses were RMB 67 million, and research and development expenses were RMB 231 million. The increase in R&D primarily reflected our targeted investments in the SGI opportunities that David discussed earlier.

Even with these continued investments, we maintained solid profitability. GAAP net income reached RMB 71 million or USD 10 million, representing an increase of 60% year-over-year, marking our fifth consecutive quarter of GAAP profitability demonstrating the strength of our operating model and the financial foundation built by our lidar business.

Non-GAAP net income was RMB 101 million or USD 15 million.

Now let me turn to the performance of each segment, beginning with our core lidar business. Lidar revenues reached RMB 816 million or USD 120 million. The segment generated operating profit of RMB 66 million or USD 10 million, demonstrating strong profitability and the cash-generating capacity. Total lidar shipments reached over 628,000 units, up close to 80% year-over-year. ADAS lidar shipments increased approximately 60% to over 485,000 units while robotics lidar shipments grew approximately 193% to over 142,000 units. The accelerating growth in robotics alongside the continued expansion of ADAS further demonstrates the breadth and resilience of our lidar business.

Turning to SGI. The second quarter marked the segment's first revenue contribution, an important step from technology development to commercialization. SGI revenues reached RMB 45 million or USD 7 million, driven by strong early demand for our robotic actuation modules. The segment recorded an operating loss of RMB 64 million as we continue to invest in product development, commercialization and production capacity to support the future growth of both our robotic actuation modules and Cosmo.

I am especially pleased to share that SGI commercial momentum has significantly exceeded our expectations. Our robotic actuation modules are already generating revenues while Cosmo remains on track to begin contributing revenues in the third quarter of 2026.

Given this strong momentum, we are raising our full year 2026 SGI revenue guidance from RMB 100 million to a range of RMB 200 to RMB 300 million. We are particularly encouraged by SGI's growth trajectory and expect the business to reach approximately USD 100 million in 2027 in revenues and achieve breakeven in the same year. This give us increasing confidence in Hesai's dual-engine growth model.

Our lidar business continues to deliver scale, profitability and cash generation, while SGI is rapidly emerging as a powerful new growth engine. Together, they position Hesai to capture the enormous opportunities ahead in robotics and physical AI.

With that, this concludes our prepared remarks today. Operator, we are now ready to take questions.

Operator

[Operator Instructions] Our first question comes from Tina Hou with Goldman Sachs.

分析師問答

Tina Hou

So my question is really regarding the actuation module. So obviously, you have raised your revenue guidance for the strategic growth initiatives. And also, I think, starting from June second quarter, we've seen more news regarding [indiscernible] collaboration with not only NVIDIA, but also Google Gemini robot. So just wondering if you could share more details requiring these and other future potential collaborations. At the same time, what would be the long-term revenue potential for this business? And also, how do we achieve the strategic synergies for the partnership with Sharpa between Sharpa and Hesai?

Yifan Li

Thanks for the question. This is David. I will take this question. Look, Hesai is building a full stack infrastructure platform for robotics and physical AI, empowering them to see, understand and act. And these are capabilities that take years of R&D, engineering and real-world validation to build, right? Since day 1, Hesai's vision has always remained the same, empower robotics, elevate lives. Lidar happened to be the first commercial product through which we demonstrated that we could build a leading technology platform and scale it globally. But that is not the end of our journey you see, it is just the beginning.

And I, myself, am also a Co-founder Sharpa, a humanoid robotics company that has gained meaningful recognition globally, including recent adoptions by Gemini Robotics and NVIDIA. And very soon, Sharpa is expected to begin its first real-world commercial deployment at a Dairy Queen store in Shanghai. We believe this will be the first deployment of its kind, not a lab program demonstration, not a pop-up showcase, but a regular operating store with the robot performing tasks autonomously in a real commercial environment. And that means this is an important milestone, right, because it demonstrates what humanoids can potentially achieve when they move beyond demonstrations and into a real operating environment.

For Hesai, you know Sharpa can serve as a valuable real-world proving ground. We already supply lidar to a diverse pool of robotics companies and expect to ship more than 500,000 units this year. Through Sharpa, we can also gain first-hand insight into what actuation systems actually need to deliver in real-world environments. And that creates a powerful feedback loop. Hesai puts its technologies onto robots. The robots in turn, tell us what the market really needs from performance metrics and form factors to reliability requirements, cases, duty cycles and total [indiscernible]. We can not [indiscernible] to insight in a right rollback loading in the real world, day after day and even more importantly, from a company founded by the same founding team firsthand understanding of both the [indiscernible].

Let me give you some examples of this flywheel. For a humanoid, lidar is typically more about resolution, field of view, some packs and will and simply maximizing the testing launch [indiscernible]. And for actuation, we retain very well and with respect to reliance rental performance, consistent, durability and performance [indiscernible] all matter. These are the pension reform to be a much clear when you are building and operating [indiscernible]. And that is a difference, right, between being a component supplier and being a supplier with first-hand knowledge of how the end product actually works in the real world.

Our supply of products and provision of manufacturing services to Sharpa act as a high signal test bed and learning platform before bringing our products to the broader market. As disclosed in the announcement and circular for the continuing connected transactions with Sharpa, we are seeking to increase the annual cap under the supply of products framework agreement with Sharpa up from RMB 100 million to RMB 300 million, subject to shareholders' approval at the EGM, with the majority of this year's transactions expected to be in actuation modules.

Longer term, the opportunity is much broader than Sharpa. Our products are all designed to serve third-party customers at scale. If fiscal AI is entering a period of rapid adoption, Hesai's role is to provide the foundational infrastructure effectively selling the shovels in the physical AI gold rush. Sharpa gives us a way to sharpen those tools faster, validate them in the real world and understand what the market will need next. That is the strategic value of the flywheel. Sharpa is the proving ground, while the broader physical AI market is the opportunity. That's my answer to your question. Thank you.

Operator

Our next question comes from Tim Hsiao with Morgan Stanley.

Tim Hsiao

This is Tim from Morgan Stanley. Congratulations on this exciting new chapter as management just mentioned. So just a quick question about the actuation modules. Just could you please give us more colors on Hesai actuation modules and how they fit into a broader robotic strategy? And what kind of financial profile do you expect these products to have over time? And are you currently shipping those products to customers beyond Sharpa? Those are my questions.

Yifan Li

Thank you. This is David. Okay. Three questions really, do we want to do it? Is the market ready to do it? And can we do it well? [indiscernible] easy, yes. Actuation can be more than half of robot's bam. One humanoid may need over 100 modules. Run that math and long-term demand approaches 1 trillion units, 1 trillion, and the margin profile is currently around 40%. That's the market we want, right?

The second question, ready? Yes. Humanoids are hitting volume and real-world deployment starting this year. Every robotics company is hunting for great joints, but they're hard to find. A lot of suppliers don't have firsthand knowledge of what leading robot makers need or why those metrics matter. Joints are both critical and hard. Power, precision, size and durability all at once, like asking a pro athlete to run a marathon every day.

The third question, can we....

Operator

Next question comes from Jeff Chen with Citi.

Ming Chung

Congratulate on excellent result. My question is about the SGI. What drove the increase in sGI revenue guidance previously set? And how should we think about the revenue mix between robotech actuation modules and the Cosmo's SGI?

Peng Fan

Okay. I know a lot of our investors care about this. We actually raised the full year 2026 SGI revenue guidance from RMB 100 million to a range of RMB 200 to RMB 300 million. For 2027, we are now looking at roughly USD 100 million, about RMB 700 million. And most excitingly, SGI is expected to reach breakeven in 2027. This is because commercial demand and the speed of tech validation came in well ahead of what we originally modeled. That's also a strong signal. Hesai is evolving into an infrastructure platform for robotics and physical AI, not just a lidar company.

The main SGI driver this year is robotic actuation built on more than a decade of engineering experiences. These products matured fast and started generating revenue in Q2. Demand from Sharpa has been a major source of our revenue for robotic actuation. Sharpa produced one of the first humanoid, maybe the first actually deploying into real restaurants with a path to scale. That demand for our actuation products run ahead of expectations, which is why we raised the [indiscernible] for the continuing connected transactions under the supply of products framework agreement with Sharpa of from RMB 100 million to RMB 300 million.

A quick clarification. That's RMB 300 million is the maximum amount for the transactions with Sharpa under the supply of product framework agreements, subject to shareholders' approval at the forthcoming EGM.

Cosmo, our special -- special intelligence platform is the other SGI pillar, and it's moving fast too. Prototype shipped in July. Initial orders came within 7 days, is well on track to start contributing revenue in Q3. We are not breaking out any exact revenue split between these 2 at this early stage. Directionally, actuation is expected to be the majority of SGI revenue in 2026. Cosmo's initial contribution this year will be smaller, low 8 digits revenue because commercialization starts later this year and the higher-value cloud services compound with deployment scale over time. That revenue mix between the 2 pillars should look differently in 2027.

As Cosmo matures from the AI spatial camera into cloud processing, subscription and licensing of high-quality 3D spatial assets as contribution should raise meaningfully. More importantly, a growing share of Cosmo's revenue is expected to be recurring cloud services under the business model we described earlier at structurally higher margins as expected. So the raised SGI guidance is really about the actuation, commercialization faster than expected this year and the next, with Cosmo setup to become another meaningful contributor as the platform scales.

Operator

The next question comes from Nora Min with UBS.

Nora Min

This is Nora from UBS. I had a quick question on Cosmo. Can you brief us the business model of Cosmo, its existing and potential customers, and lastly, the delivery schedule of Cosmo?

Yifan Li

Thank you. This is David. I apologize, there's like some errors about muting different lines on the operator side. Can you hear me okay?

Nora Min

Yes, I can, David.

Yifan Li

Okay. Good. Good. Thank you. Yes. Well, I guess this is a part that we still work on the AI or the training data, I am sure it's not a part of the reinforcement learning experiment Okay. So Cosmo is actually a super exciting product. We previewed the Cosmo in April and more than 200 prospective partners have reached out, robotics film, gaming, tourism, luxuries and more and keeps extending. And the breadth is the signal that this isn't a niche budget for 1 vertical. I think about it what 2D camera became over decades the infrastructure of how the world gets captured and consumed. Cosmo is that, except this time, it's 3D and AI driven.

Anywhere a regular camera works today, there is a much bigger opportunity digitizing the physical world in 3D. I prefer not to put out a specific cap order value for Cosmo right now at this early stage because we simply don't know how big this could be. And a lot of customers are still in prototype testing on pilots. We'll manage to need is who is buying into this, who is testing hard and who is already ordering and how fast that moved? We started shipping prototypes in July. Within 7 days of the first deliveries, we've got strong feedback and locked in initial orders. That is the real pace.

Early demand is closer in 2 key areas as we see. And the first one is the part we're super familiar, the robotics. It's a sharper near-term pull because humanoids are hitting early mass production in this year. And there are hungry -- they are hungry for high-quality training inputs to [indiscernible]. The ceiling to date is still [indiscernible]. A robot can look flawless in simulation and fall apart the second, it walks into a real factory your home. Train on weak spatial data that you're getting from a blurry textbook, we need 100x that [indiscernible]. Cosmo turns real environment into high fidelity physically grounded 3D assets, richer training stacks, better real performance, less risk of model collapse from a [indiscernible] data. The second one is more interesting is actually the media digital content. We're talking about the gaming, [indiscernible], luxury tourism, AI short-form advertising, et cetera. Content has been stuck in 2D for decades. It needs a dimensional upgrade. These applications share the same goal, to make the physical world immersive and reusable. A store set a concert a historical site stops being a one-off shoot and becomes a 3D spatial asset that's reusable.

You're not photography in the world, you're turning it into a new class of content. Film production is a clear example. Traditional pipelines leaning on green screens and clear practical shoots, expensive, inconsistent and slow. Cosmo changes that equation, higher quality, higher consistency and lower cost. I can't name who we are in close talk with yet, but they are household names globally.

Cosmo is a spatial intelligence platform, AI spatial capital algorithms, 3D assets and cloud services in 1 system. The device is just the front door. The real value compounds behind it. More devices mean more inputs that deepens the asset liability, unlocks more applications, pull in more customers and drive recurring revenue through cloud usage and asset licensing. That's where the operating leverage shows up.

So early commercial traction is real. Orders and prototypes are just the first turn of the wheel. Recent feedback has already validated both our technology and the business blueprint. We are very excited about the next batch orders coming through. Thank you.

Operator

Our next question comes from Dan Linen with CICC.

Unknown Analyst

[indiscernible] from CICC. And congratulations on the strong results and impressive I confirm. And my question is about your guidance. Could you please update us on your guidance for both the third quarter and full year 2026, adapting updates to our 3.5 million units lidar shipment guidance for this year?

Peng Fan

Thanks for the question. For Q3, we expect that the total revenue of renminbi between RMB 1.1 billion to RMB 1.15 billion, with lidar shipments around 800,000 to 850,000 units. SGI should be high 8 digits of that revenue in Q3. Actuation is ramping faster and is approaching 10,000 modules of monthly production capacity soon. Most excitingly, Q3 is expected to be the quarter where revenue from outside ADAS lidar gets close to or even more than half of total revenue. That's another clear step into our role as infrastructure for robotics and physical AI.

That acceleration also reinforces our confidence for the full year. We have raised SGI expectations, but lidar remains the core. We are reiterating 2026 shipment guidance of 3 million to 3.5 million lidar units. We shipped about 1.1 million units in the first half, almost double last year. And here is what people sometimes miss. Our ADAS business follows auto seasonality. Second half is usually much stronger. Last year, roughly 2/3 of our full year volume came in second half. So this second half ramp is normal for our business. And on top of that, there are 3 things driving lidar growth. First, penetration. Lidar is not just a story for the new EV players anymore, traditional OEMs like Geely and Changan, they are putting more lidar into their EV architectures, too. Lidar penetration on EVs was close to 20% in 2025. We think it could get to roughly 30% to 40% this year, and we've been #1 in China's long-range ADAS lidar market for 17 months straight according to Gasco.

So bigger market, strong position. We like that setup. Second, don't think of penetration as capped at 100%. It can go well beyond 100% because it's no longer 1 lidar per car. China's mandatory L3 and L4 safety standards came out in August and should take effect in July 2027. Automakers can't wait until then and start drilling holes in the car to add lidar. These centers have to be designed in and integrated at the factory. That's why we are already seeing multiple lidar setups for broader coverage and also more safety redundancy.

Remember how we've always described lidar, like airbags. You don't just want 1 airbag in car, and we've already secured a multi-lidar design wins with lia, Xiaomi and Changan with 3 to 6 lidars per car. And I don't think this stops at L3. Over time, we are also seeing advanced L2 picking up blind spotting lidards. So multi-lidar is becoming a major trend in 2026. And once people experience a safer, more capable driving system, it's very hard to go backwards.

Third, robotics. Humanoid, low modes, robovans, robotaxi, demand is picking up across the board. [indiscernible] AI is especially interesting. We are already working with more than 50 leading companies in this space, including Unitree. Morgan Stanley estimate that the robot lidar hand would be 6x the size of cars. Personally, I think it can be even larger because any robot moving through the real world needs to know where it is, what's around it and what's changing around it, especially when it's interacting with people all the time. Lidar is becoming a basic sense organ for robots, the way eyes are for humans.

So put it together, the core lidar business can keep compounding and frankly, remaining our cash cow, while SGI is the new growth driver, that's already taking off, okay? Hope this answers your question.

Operator

Our next question comes from Jia Lou with BOCI.

Jia Lou

This is Lia Lou from BOCI. My question is regarding robotics lidar. Any upside or downside risk to our robotics lidar shipment guidance for 2026. In terms of downstream market, beyond the lawn mowers, lower van and 2-wheelers, where is the penetration accelerating? And that the FCC's recent action on following produced robotics devices affect our robotics business?

Peng Fan

Thank you. Okay. We are still expecting the full year outlook for our robotics lidar over 500,000 units. Last year, it was around 240,000. We are on track for the delivery. Q2 robotics lidar shipments are almost tripled year-over-year. Demand is broad, not concentrated in 1 niche.

On the FCC piece, we don't see a material direct impact at this point. Most of our robotics lidar volume this year is robotic lawnmowers. Those customers are largely Chinese manufacturers focused on Europe as their key overseas markets. U.S. lawnmower penetration is still relatively low. We are watching the rules closely and we'll stay fully compliant. But as of today, it doesn't change our shipment outlook. More importantly, lawnmowers are just 1 wedge of a much bigger opportunity. The core view is simple, any robots moving through the physical world has to do what humans do, see, understand and act in real time.

Lidar is becoming a fundamental sense of organ for robots. A robot without reliable 3D perception is like working through an unfamiliar room with your eyes closed. We already hold leading positions across humanoids and robotaxi, robovans and robotic lawnmowers per GGI [indiscernible]. Humanoid demand is especially strong. We are working with more than 50 [indiscernible] AI companies worldwide, including Uniti. JT128 is quickly becoming a go-to light up for human noise and paves navigation, [indiscernible], avoidance, safe interactions with people and objects.

Warehouses and logistics are moving fast too. In some high net performance setups, a single machine can carry as many as 15 JT128 units for full coverage. JT128 also carries a higher ASP than the JT16 used in lawnmowers, harder performance requirements, more lidar content per machine. So robotic lidar isn't just an add-on to auto, over time, it can get substantially larger, more environments, potentially far more machines.

We like to say lidar is the shovel in the physical AI gold rush. We don't need to call which category scales first. Humanoids, warehouse robots, robotaxi, lawnmowers or something else, whichever wins reliable spatial perception is essential. Hesai is set up to ride that growth across the whole ecosystem.

Operator

Your next question comes from Jesse Lowe with Bank of America Securities.

Yu Jie Lo

This is Jessie from Bank of America. I just have a quick 1 around our key customers, Xiaomi. Xiaomo has included [indiscernible] its lidar supplier on the latest model on. So how do you see the size developing in Xiaomi future models? And more broadly, as more suppliers enter into our customer supply chain, how do you expect the competition pricing pressure and also the supplier shares to evolve?

Peng Fan

Okay. First, Xiaomi, moving further into [indiscernible] lidar is actually encouraging for the whole industry, even on L2, they are adding a real lineup where there used to be 0. That tells you something, automakers and consumers are recognizing what lidar is worth. Our long-term view hasn't changed. Every intelligence vehicle eventually carries at least one lidar. Many will carry several for coverage and redundancy. So it isn't a flight over 1 liar per car. As lidars per vehicle rise, the opportunity can become several times larger. The pie itself is turning out to be much, much bigger than people thought.

On competition, market sourcing is normal in auto industry, especially as programs scale. We don't comment on allocation for individual future models. That may shift sometimes with performance, cost, capacity and platform needs. Better to let the data speak. According to GasCo, Hesai has been #1 in China's long-range ADAS lidar market for 17 straight months with roughly 40% to 50% share. And we've done that while keeping relatively healthy pricing and margins. As the leader, our goal is at 100% share at any cost. A disruptive price war is the worst outcome for everyone. It starts, technology, quality and safety. We'd rather stay clearly ahead on tech, deliver more value to customers and earn a fair return on that value. That confidence comes from structural advantages. Proprietary ethics, deep system engineering, large-scale automated manufacturing, a broad product portfolio, automated grade reliability and years of mass production, those let us stay competitive even when we are not the cheapest option.

Customers aren't buying a cute little box on the blue, they are buying safety and reliability for the moment that camera fails. New cars, lidar is the airbag, as we've always said. We are also investing in the next cycle, Picaso, our full-color ultrasensitive 60D integrated debt and RGB color at the chip level. To the best of our knowledge, Hesai is currently the only company in China taking [indiscernible] chip level full-color lidar into mass production on track for the second half of 2026. That's the kind of innovation that lidars compete on more than price. Over time, the lidar versus camera debate phase, they become 1 integrated system. The only question that remains is when the car drives itself, is my family safe enough?

People also obsess over shipments or revenue share but misprofit share. Just look at Apple in smartphones. Its share of industry profits has been fairly high -- far higher than its share of unit shipments. That wasn't a price war, it was technology, product, brand, and a differentiated value. We think about Hesai the same way. We'll keep working to hold our leading position like 40, 50-ish, but leadership isn't just shipping the most units, it's capturing a leading share of the industry economic value, protecting healthy gross margins, reinvesting in innovation and creating long-term values for users and for society.

Operator

Our next question comes from Aaron Wang with Jefferies.

Weijie Wang

This is Aaron from Jefferies. My question is on the LIDAR ASP and the margin side. Given the ongoing industry-wide pricing pressure and also the competition from peers, could management elaborate more on the ASP trend for lidar products over the next few quarters? And also, how should we think about our margin profile going forward?

Peng Fan

Okay. We are positive on the resilience of gross margin. Full year 2026 outlook is unchanged, close to 40%. When -- and we expect to keep a healthy margin profile going forward. First, look past the half line blended ASP. The L3 shift change the mass from one lidar price to total lidar content per vehicle. An L2 car typically carried one large long-range lidar. Now we are already seeing L2 as a real live sport unit, 2 lidars. Entry-level L3 might be 1 ATX plus 2 FTX sporting lards. More advanced setups with ETX and actual can push total content to roughly USD 500 to USD 1,000 per vehicle. So the lidar value we deliver per car is rising, and that's showing up in vehicles launching this year.

Gross margin is just about price, it's price minus cost. Our price, we're already #1 in the market share. So we have no interest in giving up pricing just to chase more share. We compete on brands, technology and quality. On cost, years of in-house ethics, system level integration, automation and scale have led us keep bringing costs down. Net of that, we target a relatively stable ADAS margin profile. The mix helps the group too. Robotics lidar and overseas business generally run higher margins than domestic ADAS. Robotics lidar almost tripled year-over-year in Q2, and the global business should keep expanding. As those scale, they remain important support for healthy group margins.

SGI also started contributing revenue in Q2. Too early to comment on near-term financials, while products are still ramping. But long term, we expect SGI margins above 40%.

Cosmo especially, with cloud services in the mix, should carry a structural higher margin than pure hardware and become increasingly accredited to the growth.

Operator

There are no further phone questions at this time. I'll now hand the call back over to Yuanting Shi for closing remarks.

Yuanting Shi

Thank you once again for joining us today. If you have further questions, please feel free to complaint our IR team. This concludes today's call, and we look forward to speaking to you again next quarter. Thank you, and goodbye.

Operator

That does conclude our conference for today. Thank you for participating. You may now disconnect.