From Horizon Robotics to D-Robotics: How It Aims to Become the 'Wintel' of the Robotics Industry | Unity Ventures Portfolio
Finding Breakthroughs Where No One Else Competes
Recently, D-Robotics, a subsidiary of Horizon Robotics, announced the completion of a $100 million Series A funding round with participation from Unity Ventures.
Amid the accelerating convergence of AI and robotics technology, D-Robotics has pursued a distinctive strategy of "finding breakthroughs where no one else competes," dedicated to building an open and empowering robotics infrastructure platform. As a company that doesn't manufacture robot hardware itself but serves all robot manufacturers, its innovative model has drawn considerable attention.
Cong Wang, CEO of D-Robotics, previously led Horizon Robotics' AIoT and Robotics Division, and spearheaded D-Robotics' spin-off from Horizon to focus on creating the "Wintel ecosystem" for the robotics industry.
In the following conversation, Wang discusses the three growth curves of the robotics industry, the rhythm of technology and market timing, China's unique advantages in global robotics competition, and how tech companies can find paths to growth amid uncertainty — offering a technologist-entrepreneur's perspective on the concept of the "infinite game."
Key Insights:
-
In China, great companies need to get strategy and timing exactly right. Huawei, Li Auto, and Xiaomi all possess exceptional strategic depth and organizational capability; technology is only the second thing to consider.
-
Our positioning is robotics software and hardware infrastructure. There will be thousands of robotics companies in the future, and this industry structure requires a company that builds underlying software and hardware infrastructure to support it — only then can the industry become better and larger.
-
Our strategy is to compete where no one else is competing. The robotics industry happens to meet these criteria: currently small but with massive future potential, and technically demanding.
-
Building ecosystem products fundamentally requires openness and altruism. We are friends with nearly all our partners and customers in the robotics industry. We don't compete with them for business; we help them make money together.
-
What D-Robotics wants to do is more like an infinite game: on one hand, driving robots to iterate intelligently with unlimited potential from an empowerment perspective; on the other, robots will forever give birth to new categories we can't even imagine.
-
Don't do me-too things; do things that are scarce and different, and avoid competition. Analyze what you want to do and what you can do.
-
A company's CEO, team, and mission ultimately need to be highly unified — such a team will have explosive战斗力.
Cong Wang, CEO of D-Robotics
01 Breaking Through Uncharted Territory: D-Robotics' Bold Bet from AIoT to Robotics
Please introduce yourself.
Wang: After undergrad, I went straight to a PhD program in the United States and was doing research the whole time. But I eventually realized research wasn't what I most wanted or was best at. I preferred working on products connected to daily life that could actually be deployed — this suited my personality better. So in my fourth year of the PhD, I dropped out to start a company, focusing on knowledge graphs and natural language processing.
In 2015, I founded a company with several influential Chinese scientists in the industry. We started coding and building demos in a garage in the US. That happened to be the peak year for entrepreneurship; we secured funding quickly and returned to China to build the business. The division of labor was: the CEO had a strong academic background and focused on the big picture, while I led the domestic technical, sales, BD, and product teams. At this stage, I could participate in the full cycle from product development to delivery — that end-to-end闭环 experience was incredibly satisfying.
Our business focused on "natural language processing + finance." After strategic analysis, we determined NLP could land in two major directions: search and recommendation, which the major platforms were already doing; and customer service. Among all verticals in customer service, only finance, legal, and healthcare had high dependence on NLP. But law and medicine required deep background knowledge and channel resources, so we ultimately chose finance. However, this business model tended toward customized projects and was difficult to standardize. Having researched AI for years, we kept thinking about how to build platform-level AI products rather than premium outsourcing. After three years, I chose to exit.
In 2018, I joined Horizon Robotics. When I joined, I originally wanted to work on Horizon's AI infrastructure platform. But since Kai Yu knew my background and understood I was good at pulling requirements, finding markets, and closing loops, I was soon transferred to the AIoT department to lead marketing and BD. That experience was also very rewarding. 2019 was the peak year for "AI + industry" — the four AI dragons were constantly talking about "AI + industry." I connected with almost every industry: retail, industrial, education... I even visited prisons to look at intelligent solutions. Doing BD is essentially market analysis — finding opportunities and growth points to determine market strategy.
After nearly a year, Kai decided to go all-in on automotive. By late 2019, amid Horizon's strategic adjustment, I proposed to Kai that I manage the AIoT team, and he was very supportive. So from January 2020, I took over the AIoT team. 2020 should be considered the founding year of D-Robotics, because that team essentially became D-Robotics' earliest team composition. To this day, only three or four people from that team have left — that's something I'm quite proud of.
So 2020 was a critical turning point, and the prototype of what is now D-Robotics?
Wang: Yes, 2020 was also the hardest year: on one hand, we had to handle various legacy issues and couldn't let customer deliveries slip due to the adjustment; on one hand, we had to deal with employees "riding a donkey while looking for a horse"; on one hand, we had to explain the adjusted non-automotive business to customers; on one hand, we had to find a new direction ourselves. That year also happened to coincide with the pandemic — everything needed handling. We even had employees stuck in Wuhan who couldn't return. That year was constant struggle. We reoriented, developed our own methodology, broke into ECOVACS, TCL, Hisense, achieved落地 in many industries, and did many excellent things.
At the strategic level, we needed to choose an industry. In the short term, we looked at what existing capabilities could do while maximizing commercial returns; in the long term, we had to think about which industry the team could persist in and build compounding returns. Our judgment: robotics.
On one hand, robotics was our original aspiration. Many people at D-Robotics joined because of the name "Horizontal Robotics," hoping to truly build a robotics company. On the other hand, from a strategic perspective, a startup must choose an industry that is currently small but has massive future potential, and is technically demanding: if the current market is already large, major players will compete and it won't be a startup's market; if the technology isn't difficult, it will quickly become内卷. Our strategy is what Kai has always emphasized at Horizon: compete where no one else is competing.
The robotics industry happens to meet these criteria. Now the industry is "buzzing" with news every day, but it's actually still in early stages with long technology cycles, and major players haven't fully entered yet. In such a赛道, only those who believe in it, persist, and continuously accumulate can succeed. Great companies in China and the United States have succeeded this way in the past — not by chasing whatever is hot.
By late 2020, we were firmly committed to robotics. In 2021, the department was renamed "AIoT and Robotics Division," and in 2022, directly changed to "Robotics Division." We found that no one really understood the robotics industry; there were no unified standards, everyone did their own thing, but industry demand for AI was persistent. This made it possible for a company like ours to emerge — this scenario suited us well. By the second half of 2022, we brought in several robotics industry experts, and in the first half of last year began the independent spin-off process.
Why choose to operate independently in 2024, and what opportunities did you see?
Wang: It was a good window. Although there was still some distance to technology落地, by 2024 the timing of technology had reached a critical point: large models were beginning to emerge, capital attention to robotics had been rising since the second half of 2023, and talent mobility was sending the same signals.
Considering all factors, we felt spinning off would enable faster development. In fact, after becoming independent last year, we did develop very rapidly: team size grew quickly to twice last year's size, customers reached nearly 200, revenue tripled, and overall growth was very fast. After independence, we could have a new image, new incentives for the team, and new models for talent recruitment.
What is the current relationship between D-Robotics and Horizon Robotics?
Wang: We remain a Horizon-affiliated company. As Kai said at the launch event, Horizon has two major businesses — Horizon's automotive business and D-Robotics. All of Horizon's non-automotive business is operated by D-Robotics. While there are different considerations regarding shareholding structure, we refer to ourselves externally as both "Horizon" and "D-Robotics," similar to "JD-affiliated" or "Alibaba-affiliated" companies.
Spinning off from Horizon Robotics, what makes you different from other companies?
Cong Wang: Having spun out from Horizon Robotics gives us a lot of unique advantages. We fully leverage Horizon's past积累, such as the BPU and toolchain, while D-Robotics also builds its own SoCs specifically for the robotics industry. Although end customers have different needs that are hard to unify, the underlying technology is transferable.
For example, by utilizing Horizon's BPU, D-Robotics taped out its own Sunrise 5 chip in early 2024, which is now basically crushing the market and meeting robotics customer demand. We've also launched high-compute chips for Embodied Artificial Intelligence. People see D-Robotics as coming from Horizon, and they believe that only a team incubated by a company of Horizon's caliber could pull off such ultra-high-compute products — so they have more confidence in us. In reality, building SoCs this large requires continuously hiring people, raising money, and eventually dominating the market.
02 Three S-Curves + "Wintel for Robotics," Positioning for Next-Generation Infrastructure
Could you explain D-Robotics' positioning?
Cong Wang: Our positioning is robotics software and hardware infrastructure. We've always had this thesis: the categories of future robots will be extremely diverse — we already serve over 200 companies, and there will inevitably be more. Even if general-purpose humanoid robots emerge someday, there will still be many different brands. Every scenario will have different categories, and behind every category there will be a cluster of companies — thousands of robotics companies in the future. An industry structure like this needs a company doing underlying software and hardware infrastructure to support it, so the industry can get better and bigger. Just like the internet industry — without Amazon and Alibaba building infrastructure, a twenty-person team would have to set up its own servers. D-Robotics' positioning is that we don't build the robot bodies themselves, but we serve all the companies that do.
Specifically, our positioning has two dimensions: "horizontal industries" and "vertical products." On the horizontal industry dimension, we see three growth curves: first, the intelligent upgrade of traditional robots (like vacuum cleaners and lawn mowers); second, the explosion of new-category robots that use large models and even some embodied technology while maintaining the mobile base attributes of traditional robots — golf robots, tennis robots, companion robots, and so on. Third, Embodied Artificial Intelligence robots. The timing differs for each of these three curves.
Vertically, it's the product dimension: we layout the entire industry through a product matrix of "chips" and "algorithms + software platforms." Additionally, we have an "ecosystem line" to cultivate more developers, letting them benefit from our technology and products early on.
The hardest part is judging the timing of each curve, and timing in turn determines resource allocation and positioning for each line. At every strategy meeting, we put up these three curves and the team collectively judges the timing and what approach to take for each curve.
What methodologies and approaches do you currently use to judge timing?
Cong Wang: Looking at the overall market, we compare the three curves to different periods in Chinese history. On the first curve, the traditional robotics market has basically reached a "Three Kingdoms" structure — the era of warlords carving up territory is over, and now only the "Wei, Shu, and Wu" major players remain. The customers and products in this market are clear, you just go execute.
The second curve is more like the "Warring States period" — everywhere there are many startups competing like "minor feudal lords." Most customers at this stage are underwater, and reaching them requires going through ecosystems, channels, even investor introductions.
The third curve is more like the "Spring and Autumn period" — there's not even a methodology yet, people haven't even unified on ideology: should you create a new technology from 0 to 1, or follow the market's major trends with higher leverage? For example, everyone's doing VLA now, and I discussed with my team whether we should self-develop VLA for embodied intelligence. First, we look at whether the technology development is certain; second, we look at the difficulty of replicating it. Algorithm replication isn't hard — once the algorithm paper is public, we can replicate it in a month. By contrast, autonomous driving has reached the stage of competing on overall systems engineering integration capability; it's no longer about algorithms. The embodied intelligence industry hasn't reached that stage yet, so we should pursue broader industry partnerships and find higher-leverage methods. These three stages require different approaches, and this is a key test of a company's operational and strategic capabilities.
A truly strong company never relies on single-point technology. Historically, no company became great on single-point technology alone — at best, they got acquired for it. In China, great companies need extremely accurate judgment on strategy and timing, like Huawei, Li Auto, and Xiaomi. They all have very strong deep strategic and organizational capabilities; technology is the second thing to consider. What I greatly admire about Li Auto is that they were among the last to do autonomous driving, yet produced one of the strongest products — precisely because their strategic timing and organizational strength were excellent.
D-Robotics' vision is to become the "Wintel" of robotics. How do you understand this positioning?
Cong Wang: "Wintel" is essentially software and hardware infrastructure — just as the PC era was Windows + Intel, in robotics it's a large toolchain. Over the past 5-10 years, traditional robotics manufacturers invested more in hardware R&D, but in the next 5-10 years, robotics companies will invest more in software and algorithms. Just like in autonomous driving, what burns the most money is still algorithms; hardware can mostly be solved through foundries. The robotics industry will follow the same path.
Since software investment is becoming heavier, software will depend on several elements in the future: first, increasingly usable chip platforms that can run more applications; second, large toolchains that reduce customer human resource needs for robot algorithm iteration from 100-200 people to 50-100, or enable faster new product launches with the same headcount. Whoever captures this will be the new "Windows," and that's what we want to do.
What makes us especially happy is seeing customers mass-produce robot products connected to daily life. The joy of consumer electronics is making many interesting products that you can actually use at home.
Standardization in the PC industry was built on converging demand and converging technology, while robotics is in a phase of exploding demand and diverging technology. Does a "Wintel" opportunity really exist in robotics? What is D-Robotics' path and leverage to achieve it?
Cong Wang: In the PC era, demand at the operating system level was relatively convergent, while at the application level it was completely different — some people gamed, some used Office, some made browsers. Everyone using Windows had different skill sets and different needs.
Although robotics technology hasn't fully converged, when you break down robotics software there are still many commonalities: communication architectures are similar, everyone needs to avoid excessive memory usage and ensure sufficient real-time performance; while algorithm directions differ, data collection and generation is the same — all domains need data generation to extend cases or do scene editing; while VLAs may have different final forms, the joint optimization of operators inside chips and support for bandwidth are still the same.
We can now help a customer run through the full loop from data collection, data generation, training to validation and testing, and scale it further. The current state of Embodied Artificial Intelligence is that every company is launching algorithms and promoting them through WeChat public accounts. We need to replicate them, but find no public test sets, no unified environments, everyone testing on different real machines. So we partnered with Shanghai Jiao Tong University to do a CVPR challenge, trying to create the Kaggle for embodied intelligence (ZP note: Kaggle is an internationally renowned data science competition platform), then iterate on the test validation set and evaluation access. So even in scenarios with diverging demand, you can find anchor points, accumulate clues bit by bit, and eventually collapse into standardized solutions.
03
Ecosystem First, Word-of-Mouth Virality: Making D-Robotics the Most Usable Robotics Platform
What does the emergence of AI mean for the robotics industry right now, and how is D-Robotics seizing this opportunity?
Cong Wang: AI is now being applied in every domain of robotics, from Locomotion, Navigation to Manipulation. In the past, Locomotion was mainly traditional control theory; now it uses simulated Imitation Learning. Manipulation was previously Rule-based, and is now also adopting Learning approaches. In fact, every module in robotics is now using Learning techniques. Our tool generation part is large models — every part from 2D to 3D to trajectory uses Learning algorithms. You could say that if a company lacks Learning technology talent or know-how, it simply cannot do robotics.
D-Robotics is investing in all these domains, just with different rhythms in each. For example, in Embodied Artificial Intelligence Locomotion, the industry is doing reinforcement learning under simulators — while the movements differ between dancing and martial arts, the underlying pipeline and algorithms are not fundamentally different. So we're trying to land this on our own chips for joint optimization and efficiency gains.
In the Manipulation direction, one popular approach is Learning + Rule — for example, grasping now uses a combination of 6D Pose + AnyGraph, though some use pure end-to-end approaches. We definitely pay attention to the joint optimization of these algorithms on our own chips — these are problems customers cannot solve; the underlying joint optimization can only be done by us.
On the other side, we're also doing research on the robot brain itself. For now, we're mainly tracking the academic frontier — the team needs to know who's actually doing good work in the industry and who isn't, and we have to try things ourselves. At the same time, we're collaborating with university professors. Our current approach is based on our judgment that we haven't yet reached the timing that Li Auto or Huawei had when they went all-in on intelligent driving. When we do reach that timing, we should assemble a team of several hundred people and go hard.
Do you think D-Robotics' strategy will resemble NVIDIA's?
Cong Wang: Yes, and we've learned a lot from NVIDIA too. NVIDIA is truly the greatest company in the global chip and even AI industry — there's so much worth learning from them. Conversely, the robot industry is such a broadly divergent space that this approach really makes sense here. Looking at history, great companies follow only a few models. There's the Apple-Tesla model, where they ultimately deliver an incredible blockbuster product to consumers.
NVIDIA has launched a "robot full-stack solution" and opened parts of its toolchain for free use. What is D-Robotics' competitive advantage?
Cong Wang: From an endgame DNA perspective, we will definitely have differentiated advantages from NVIDIA. NVIDIA's core is data centers — the compute network supporting the digital virtual world. They've been laying out their cloud strategy all along, with the ultimate goal of selling more cloud products and services.
Actually, across any industry, there are very few edge-side products that can achieve large-scale mass production. So we'll end up playing two different games, though there are many areas where we can learn from or even directly use NVIDIA's supporting products. In simulation, for example, NVIDIA has invested enormously — there's no need for us to replicate that from scratch. Omniverse's output can run better on our chips.
While NVIDIA has indeed developed very comprehensively in these areas, we'll stay focused on our differentiated path, especially in edge-side deployment.
What is D-Robotics' business model?
Cong Wang: Our business model has three main pillars. By revenue share, the core is selling chips, followed by overall ecosystem layout and product technical services based on chips, and third, cloud services which just started generating its first revenue. All three will grow, just at different stages.
Currently in phase one, we're mainly relying on chips, and the primary scenarios and customers for chips are consumer electronics. From a shipment volume perspective, consumer electronics will remain the mainstay for the next two to three years, with more products emerging in this space.
From an embodied AI perspective, our core strategy lies in overall layout and positioning. Compared to consumer electronics, embodied AI products have much higher unit prices: take cars, for example, where the average unit price reaches 100,000 to 150,000 RMB, with intelligent driving main chips accounting for nearly 10% — similar to the proportion in phones. I think embodied AI product revenue could match consumer electronics in about three years, though shipment volume might be only one-tenth.
How do you plan your overseas strategy?
Cong Wang: We're taking a three-step approach to going global.
Step one is "following customers overseas" — selling to domestic customers who then sell to overseas companies. This isn't really true international expansion.
Step two is "ecosystem developer出海" [going overseas]. Starting from the second half of last year, we've gradually completed channel setup and overseas certifications, mainly targeting Japan, South Korea, Southeast Asia, and Europe. Within three to four months, we covered developers in over 20 countries, including many robot KOLs, with decent results. Ecosystem expansion abroad also helps us build momentum. When we initially tried to crack major overseas customers directly, it was tough going — they held rather traditional views, skeptical about how a Chinese chip company could make good products, with some resistance in their minds. Later, through "teaching by example," gradually forming our ecosystem layout, some "signals from the nerve endings" got through, and we built reputation step by step. Neither purely top-down nor purely bottom-up customer development works — you need both ends pushing together to win customers.
Step three is direct overseas expansion — selling chips directly to customers doing R&D overseas. This is also the hardest step. We started looking at overseas placements this year and have already made considerable progress. We're already conducting overseas partnerships: one very large European robotics company has confirmed strategic cooperation, a German robotics customer is in the process of confirming cooperation, and a Swiss company is also collaborating. Leveraging the influence of these anchor customers, we're also exploring broader overseas layout.
What do you think are China's long-term advantages?
Cong Wang: China has two core long-term advantages. First, supply chain advantage. Second, engineering talent advantage. I once told a client from a certain European country that China graduates about 10 million engineering students annually, while that country's total population is only 1 million — you can see how strong China's engineering talent density is.
There's another point: overseas clients are now more willing to work with China. Chinese technology and products have improved significantly compared to before, and China's trade policies and international policies are consistent. China-foreign cooperation is mutually beneficial and win-win, and the trust in cooperation has increased.
What are D-Robotics' plans for the next 2-3 years?
Cong Wang: We're currently pushing forward in all these areas and laying out more customers. In embodied AI, we plan to launch a higher-compute embodied chip early next year, while also working on a series of new algorithms at the software level.
We're not trying to be the inventor of some era-defining algorithm. Rather, we want our products to be sufficiently compatible with all algorithms on the market, so everyone can use them "out of the box." I often tell the team: in three years, when people mention D-Robotics, I want them to immediately think of the most usable robotics platform — I want the two to be synonymous. Just as it took us many years at Horizon Robotics to establish that "persona": Horizon = autonomous driving chips. Robotics will require more accumulation, but I think we can draw that "equals sign" in about three years.
04
The Philosophy of Infinite Games: Breaking Boundaries, Driving the Future of Robotics
What is D-Robotics' current team configuration?
Cong Wang: Excluding operations roles, D-Robotics currently has roughly equal numbers of chip and software employees. The software team is divided into three parts: one doing algorithm research itself, one doing software supply chain at scale, and one embedded software engineering group. Much of the technology builds on Horizon's past积累 [accumulated experience] — for example, our ROS system was developed internally before, and now we're just doing some底层架构 [underlying architecture] optimization on that foundation. Looking at future development, we still need to recruit talent in algorithms, VLA, and new architectures. Other team building is already complete. The entire project started in 2021, and the sales team has worked together for 7 years. The team has already gone through its磨合期 [breaking-in period].
Which of Horizon's business philosophies and strategic choices have had significant influence on building D-Robotics today?
Cong Wang: Spending seven years with Kai [Yu Kai, Horizon founder] has influenced my strategic thinking quite a bit.
First, don't do me-too things; do things that are scarce and different. Don't compete; analyze what you want to do and what you can do. More fundamentally, a company's mission, vision, culture, and what it can withstand must match the CEO's personality.
Like Steve Jobs and Elon Musk — they both pursued极致 [extreme/ultimate] experience, almost to the point of偏执 [obsession], and ultimately actually built such consumer products. This kind of product needs to be "ground out," needs "not listening to others." Jobs even said customer research was "bullshit," and he had the ability to "force out" products.
And doing this kind of ecosystem product essentially requires openness and altruism. I can say quite frankly: with almost all partners and customers in the robotics industry, we're friends. We don't compete with them for business; we help them make money together. Our internal company culture is also open and transparent — people can say whatever they want to me.
A company's CEO, team, and what it does ultimately all need to be highly unified. I once attended an event where I heard the founder of Purcotton speak. He said he wanted to bring consumers the most natural, healthiest贴身 [close-to-body] products, and all his executives' messaging was unified. A team like that must have爆表 [off-the-charts]战斗力 [combat effectiveness].
From being a business unit head to being the number one at a company, what's the most direct feeling from this identity shift?
Cong Wang: A lot of things are actually quite similar. Even when I was running a business unit, I had to handle fundraising and government relations — now there's just more of it. The biggest difference is that at Horizon Robotics, if something fell through the cracks, Brother Kai had my back. Now it's all on me. And a lot of it isn't about competence anymore; it's about grit and endurance.
I never used to pray or visit temples. Now I find myself dropping by Yonghe Temple whenever I get the chance (laughs). There was a stretch last year when new problems popped up every single day. The first day, solving one problem would put me in a good mood for half the day. The next day, a fresh problem would have me worried all over again. But looking back, when you finally land a new client and their product works beautifully — that joy hits harder than all the pain.
How did you settle on the name "D-Robotics"?
Cong Wang: "D-Robotics" means "the fruit that grew from Horizon Robotics." I was at dinner with some investors recently and they asked about the name too — they even got into a debate about whether "digua" refers to sweet potatoes or white potatoes.
Which entrepreneur, book, or person has influenced you the most?
Cong Wang: I didn't do well on the gaokao, so for those four years of college I basically lived in the library. I read over a hundred books a year, diving into philosophy and psychology. Philosophically, existentialists like Sartre and the logical positivists left a mark on me. That's why I don't get hung up on things.
Existentialism doesn't waste time on materialism versus idealism. The point is: I exist first. Existence itself is the meaning. You exist, and then you go find yourself — rather than obsessing over some abstract "present moment." Camus's Sisyphus keeps pushing that boulder up and watching it roll back down, feeling life is suffocating, until he starts noticing the flowers and grass along the road and realizes life is still beautiful. Existence itself has no meaning. My favorite poet is Fernando Pessoa from Portugal, another existentialist — he influenced me deeply.
After entering the working world, many entrepreneurs and their stories inspired me: Elon Musk's first-principles thinking, Li Auto's story, books by Richard Liu and Xing Wang. I basically read and absorb everything. But at my core, I'm still pretty existentialist.
What's the most exciting industry change for you recently?
Cong Wang: I'd still say ChatGPT. Its emergence basically made my PhD studies obsolete (laughs). Things have been relatively calm since then — most of the wave after ChatGPT was more or less predictable.
What kind of company do you want D-Robotics to become?
Cong Wang: Our mission and vision have never been about becoming number one in anything. What D-Robotics wants to be is more like an infinite game: on one hand, pushing robots to iterate intelligently — from an enablement perspective, there's infinite potential; on the other hand, robots will always give birth to product categories we can't even imagine, and we want to push entrepreneurs to innovate there. We never box ourselves in. We don't let boundaries limit us.
Source: Z Potentials
