Ola Dimensions Raises Hundreds of Millions in Angel+ Round, Completing Fourth Funding Round in Six Months | Unity Ventures
Huawei's Former Embodied Intelligence "Employee No. 1" Shunbo Zhou Wants to Redefine Home Embodied Robots
Household embodied intelligence robotics company Ola Dimensions recently announced a nine-figure RMB angel-plus round, its fourth funding round completed within six months.
Founded in March 2026, Ola Dimensions secured seed, seed-plus, and angel rounds in just three months, with cumulative funding reaching several hundred million RMB. Unity Ventures was the seed-plus investor.
The "Huawei Genius Youth" label is the most eye-catching part of founder Shunbo Zhou's résumé, but the real story begins elsewhere. Before choosing entrepreneurship, Zhou had already spent twelve years studying robotics, starting in the robotics lab at The Chinese University of Hong Kong. After completing his PhD in 2020, Zhou and his advisor Yunhui Liu had already begun preparing to start a company.
During his time at Huawei, Zhou focused on how technology could lower the barriers to robot development. After starting his own company, he hoped to truly bring robots into people's lives, exploring "trainable" robots for household scenarios.
In a conversation with ChinaVenture, Zhou shared why he chose a track that "requires everyone to redefine and recreate from scratch."
In 2026, embodied intelligence has produced many notable names. Among them, Shunbo Zhou is likely one of the more prominent.
Before choosing entrepreneurship, Zhou had already spent twelve years studying robotics, starting in the robotics lab at The Chinese University of Hong Kong. There, he developed a therapeutic panda robot tasked with accompanying children with autism and elderly patients with Alzheimer's in hospitals, and also led development of a heavy-duty autonomous towing system for Hong Kong Air Cargo Terminals Limited (HACTL). As his research deepened, he realized he genuinely "loved robotics" and deeply enjoyed "bringing that passion to everyone around him." After earning his PhD in 2020, he joined Huawei through the "Genius Youth" program, formally making "robotics" his career, and achieved the following:
As the first head of Huawei's Embodied Intelligence Innovation Lab, he led the ground-up construction of Huawei's embodied intelligence system; served for years as Huawei's Chief Technology Expert for Embodied Intelligence, responsible for strategic planning of Huawei Cloud's embodied intelligence efforts, and led development of the Huawei Cloud embodied platform CloudRobo, completing "zero-to-one" technical validation for industrial-scenario robots, with benchmark POC deployments at the national embodied intelligence training facility, in advanced manufacturing, and in government services, achieving internationally leading technical metrics.
During his Huawei period, Zhou focused on how technology could lower the barriers to robot development. After starting his own company, he hoped to truly bring robots into people's lives. Ola Dimensions distilled its mission into one sentence: "We build robots so people have more time to live, explore, and create." Choosing household scenarios and exploring trainable robots were concrete steps toward this goal.
Looking at the timeline, things appear to have unfolded accordingly: Ola Dimensions was founded in March 2026, and in just three months secured seed, seed-plus, and angel rounds with cumulative funding reaching several hundred million RMB, with investors including Hillhouse, 5Y Capital, China Merchants Capital, JunShan Capital, SAIF Partners, Unity Ventures, Baidu Venture, and other leading VCs and corporate investors most active in the embodied intelligence track in recent years.
And now, they have completed their fourth round in six months. Ola Dimensions recently finished a nine-figure RMB angel-plus round. This round was jointly invested by new and existing shareholders including Chengyuanxiang Fund, XinNeng Venture Capital, Xizhou International, Topband, and HONY Capital. In addition to existing shareholders' oversubscribed follow-on investments, corporate investors such as Topband also joined as new shareholders, with both parties exploring further business cooperation; ZhiGuan Capital continued to serve as exclusive financial advisor.
However, telling Zhou's entrepreneurial story through the lens of "Genius Youth rides the startup wave" is perhaps the most boring version possible.
In fact, after completing his PhD in 2020, Zhou and his advisor Yunhui Liu had already begun preparing to start a company, with a very clear scenario in mind: they planned to develop a hauling robot for mine shaft operations. But 2020 was far from today's environment—intelligence wasn't ready, supply chains weren't ready, the market wasn't ready, and it was even less something a fresh graduate could perfectly handle. Constrained by the technological maturity and industrial conditions of the time, the project did not move forward. Zhou came to Huawei carrying this "regret," wanting to observe "changes in the industry's technical paradigm" up close, to see how to truly "push robots toward the mass market."
Everything unfolded as Zhou had hoped. Through the robotics platform they developed, a team member who barely coded and had almost no robotics development experience built a material picking application in four days; in the past, similar projects required professional developers at least a month. This stood in sharp contrast to the strong sense of powerlessness Zhou felt about "building robots" during his PhD: back then, developers needed to study physics, mathematics, control systems, programming, and hardware just to barely develop a robot that "could only move in constrained scenarios." The rapid decline in development barriers convinced Zhou that his long-held dream of "bringing that passion to everyone around him" had become possible.
So what next? Continue developing general-purpose platforms to further lower robot development barriers, or dive deep into a specific scenario, running closed loops on data, product, and business simultaneously, to build a robot that could truly create high value?
This is where Ola Dimensions' real story begins. Zhou believes "general-purpose robots" is an ideal goal, but to truly achieve this, the most realistic path is to get the data flywheel spinning in a real scenario, make the economics work, and gradually expand outward from that base. In his view, households have real, sustained, and diverse task demands, and also allow the team to continuously test and improve systems through execution and feedback.
For this reason, Ola Dimensions appears quite "contrarian" in the current embodied intelligence landscape. When discussing household scenarios, people instinctively default to "emotional companionship" assumptions, and consequently believe humanoid form is necessary. But Ola Dimensions chose a mobile base with dual arms for its product form, and insists on "prioritizing exploration of the base-plus-dual-arms form at this stage." When discussing consumer tech products, people instinctively apply "consumer electronics" framing, believing universal applicability should take priority. But Ola Dimensions' first product targets Makers and developers, inviting people more willing to tinker, train, and accept early product forms to participate in product development.
In the current embodied intelligence industry, it's even hard to find a company as a direct benchmark. Ola Dimensions' product is neither a pure vacuum robot nor service robot, nor a sci-fi-flavored companion robot. Zhou says they are in a track that "requires everyone to redefine and recreate from scratch."
These elements may be more compelling than the "Genius Youth" label. After all, "Genius Youth" is more an accolade for past accumulation. And the reason we were originally so excited about the emergence of artificial intelligence and embodied intelligence is that they made us realize we have the opportunity to personally shape the future we dream of.
Currently, Ola Dimensions' latest releases are model-stage results: the OLA-Sem and OLA-Geo models, plus household scenario demos, used to optimize the connection between user task intent, what the robot currently sees in the world, and final action. Immediately upon completing this funding round, Zhou spoke with ChinaVenture. Below are excerpts from the conversation:
Four Rounds in Six Months: What Did Investors See?
ChinaVenture: Typically, when a six-month-old company completes four consecutive rounds with different "round names," it usually means the company keeps evolving—whether in team, financial model, technical breakthroughs, or product milestones. So I'm curious what evolutions Ola Dimensions achieved in these six months that kept investors participating?
Zhou: Investment is a comprehensive decision. The first two rounds of investors valued the household track more, the team's past accumulation in robotics, and how clearly the founding team had thought through the problems. They were judging our starting point, potential, and subsequent execution speed. By the third round, including the current ongoing round, institutions began demanding to see actual progress: whether previously planned technical milestones were delivered, whether the team was continuously improving, whether the company was reducing uncertainty.
Over the past few months, we built data and model infrastructure. In August, we trained our first version of the foundation model, achieving strong results on several mainstream global benchmarks. It proved the team could catch up to industry progress in a relatively short time along a differentiated path. We also completed end-to-end validation in real scenarios, first verifying our overall concept in relatively controlled environments; related results will be released soon.
From model R&D to household scenario validation, Ola Dimensions hopes to gradually establish a Real-world RSI (Recursive Self-Improvement) system: letting robots execute real tasks, receive feedback, then through data processing, training, and evaluation, return to scenarios to verify improvements. What we're showing now are stage results along this route.
On the ecosystem front, we have already accumulated over 1,000 high-value seed users; established strategic partnership with HomeKing, the world's largest premium housekeeping company, with plans to jointly produce high-quality household data in 2027; and are also strategic partners with D-Robotics. The team currently numbers about 70, with core positions basically filled. Core members mostly come from long-accumulated collaboration networks, with mutual trust and shared working foundations.
Where Does the Most Complex Household Data Come From?
ChinaVenture: You mentioned HomeKing has excellent SOPs. Is the data generated by housekeeping service personnel in real households? Can we understand this as you distilling aunties' experience into robots?
Zhou: HomeKing's value lies not only in household access, but in its professional housekeeping service SOPs. It breaks down housekeeping services very finely, including cleaning, organizing, childcare, cooking, and other specialties, with professional procedures for each task within each specialty. Take organizing as an example—it's work requiring professional training and even certification exams. If robots can bring professional organizing capabilities to ordinary households, this is more valuable than having non-professionals simply repeat motions.
ChinaVenture: You said organizing requires professional training, but I've visited many data collection centers where companies arrange young people to repeatedly teleoperate robots folding clothes, ironing clothes, and putting them in wardrobes. What do you think of this data collection approach? HomeKing is a good entry point, but this scarce data is ultimately limited.
Zhou: Different training stages require different data strategies. The pre-training stage needs large amounts of in-the-wild data; we collect only a small portion ourselves, mainly obtaining it through data consortium exchanges, purchasing from industry companies, and partnerships with companies like HomeKing. The middle-train stage relates to final product experience and requires end-to-end closed loops, so private data will form. The post-train stage will be lighter, building data closed loops through compliant partnerships, R&D collection, and voluntary user feedback.
ChinaVenture: Companies like Roborock and Dreame that have already entered household service scenarios are also continuously accumulating their own data, forming data flywheels. If household robots gradually become consensus, why can Ola Dimensions better utilize this data? Compared to Roborock and Dreame, where is your latecomer advantage?
Zhou: First, I don't believe household robots have already become industry consensus. Many companies ostensibly choose households but actually enter from hotels, commercial services, industrial, or specialized scenarios. Hotels are standardized scenarios requiring ROI calculations; companies can easily add standardized process orchestration, visual markers, and various engineering tricks, ultimately precipitating into customized solutions for hotels. This differs greatly from real household data, model, and infrastructure capabilities. The household scenario remains early-stage, with different companies choosing different entry tasks and investment depths.
Roborock and Dreame are certainly large companies, but large companies doing new integrated hardware-software products doesn't necessarily mean absolute advantage. DJI similarly faced competitors that had been established for years with capital and talent advantages in its early days, ultimately succeeding through the geek-first, mass-market-later route, plus organizational and technical decision-making flexibility.
I believe in this wave of embodied intelligence, data has primacy. Large companies can buy pre-training data, but once products enter the real physical world, large amounts of private data will be needed to guarantee experience. This kind of accumulation takes time. Ola Dimensions began accumulating data collection and private data closed-loop know-how from day one, and enters real scenarios through industrial partners like HomeKing and China Merchants Capital.
Adding one robotic arm to a vacuum robot, then later adding a second hand, waist, or greater operational space—each change in body configuration may force re-collection of previously teleoperated data. This is extension-style innovation. Ola Dimensions accumulates human data from the start, researching how to let human data drive robots. Hardware can evolve gradually; human operation data and methods can continue accumulating. Household embodied intelligence is a new category that has never existed in history; simply extending existing appliance forms makes it difficult to grow sustainable, evolvable household robots.
ChinaVenture: There's a saying that the reason people are fixated on developing "humanoid robots" is because tables, cabinets, and doors in human society are designed according to human body structure. Past household robots replaced not family members, nannies, or housekeepers, but were closer to replacing a broom, while Ola Dimensions develops with people at the center. Is this understanding correct?
Zhou: It can be understood this way, but I want to add: people-centered first means using human data to drive robots; it doesn't mean the body must be bipedal humanoid. The human world is indeed designed according to human body structure, but human configuration is also a comprehensive trade-off over hundreds of millions of years of evolution, with no single aspect being absolutely optimal. Human mobility efficiency, speed, and strength are not the highest.
Robots currently serve humans, and in the future should also accomplish things humans cannot. Therefore, bodies will pursue capabilities beyond human ones. I don't believe robots will ultimately converge on bipedal humanoid form. Human data can teach robots, but robot bodies can still take other forms.
Why Did Ola Dimensions Choose "Geek Co-Creation"?
ChinaVenture: One interesting aspect is that you've chosen the "geek first, mass market later" route in product development. You screen experience officers and have accumulated a group of seed users with development enthusiasm to participate in co-creation. But from consumer electronics logic, the end target is ordinary households. For consumer electronics, consumption matters; product extensibility and explorability should yield to universality and ease of use. Handing initial development to geeks first—won't this skew the development direction? Might geek-created demands be pseudo-demands or transitional demands? Finally, the company still has to subtract from excess demands—won't this increase product development burden?
Zhou: Serving ordinary households with out-of-the-box usability is certainly the ultimate goal. But embodied intelligence is not as mature as outsiders imagine. Doing products first requires clearly defining boundaries. Ola Dimensions' early customers are not end household users, but Maker households. At this stage, both hardware and software need a group of people with higher fault tolerance, stronger curiosity, and willingness to participate in co-creation.
And Makers are not a fixed niche group. Large models and Agent tools are continuously lowering development barriers. We've already lowered the barrier from simultaneously understanding robotics and AI, to not needing robotics experience, not needing to have trained large models—just having conventional development sense to participate. As tools advance, more people will qualify as Makers every day. The relationship between early users and mass users is more like a continuously sliding process.
"Training" is not a temporary solution when product capabilities are insufficient either. As long as society continues developing, people will constantly emerge with new ideas, preferences, and demands. Robots need a long-term interaction method to transmit human innovation and preferences to them. We believe training will be a long-term core capability of household robots.
Has Embodied Intelligence's GPT Moment Not Yet Arrived?
ChinaVenture: You mentioned earlier that embodied intelligence seems to have had several GPT moments, but ultimately didn't rapidly emerge and proliferate like large models.
Zhou: Perhaps I didn't express myself clearly. What I meant is that in the industry, people have repeatedly claimed embodied intelligence has reached its GPT moment, but actually everyone has overestimated what robots can achieve in the near-to-medium term. Industry expectations for staged demonstrations sometimes outpace actual usability; the reality is that generalizable robots remain genuinely difficult to deploy.
ChinaVenture: If a true embodied intelligence GPT moment does arrive in the future, what should it look like?
Zhou: My understanding of an embodied GPT moment is that an embodied foundation model exists in the industry, based on which users need only pay extremely low costs—for example, simply recording one demonstration of their own—to let robots learn to execute tasks. Essentially, the marginal cost of using robots needs to be driven very low.
For example, recent ICL, or in-context learning, has shown the embryonic form of this capability. Companies like Generalist AI and Skild AI have demonstrated some related demos, but there's still a long distance to true usability. Currently ICL mainly manifests as fast learning speed, and mostly concentrates on simple tasks with sufficient training; complex tasks remain difficult.
ChinaVenture: Recent public information mentions Ola Dimensions released the OLA-Sem and OLA-Geo models. What improvements did each make?
Zhou: I believe OLA-Sem and OLA-Geo are not two isolated model optimizations, but are consistent with the Action-Native Foundation Model technical route we are advancing, and are the first batch of stage deliverables along this route.
Our core thinking is to let Action participate in the model's understanding of the physical world from the pre-training stage, rather than simply connecting an action module after visual and language models are trained. Along this direction, OLA-Sem focuses on strengthening the connection between spatial semantic understanding and action generation, while OLA-Geo focuses on reinforcing spatial geometric representation, letting the model understand both "what to do" and "where objects are, how to interact with them."
Both models achieved first-place global rankings in their corresponding evaluation settings on the RoboTwin Benchmark, preliminarily validating the effectiveness of this technical route. For us, this is not merely single-point capability improvement, but the first step toward unified Action-Native Physical Representation. Next, we will further unify semantics, geometry, action, and physical interaction into the foundation model, letting robots truly possess the ability to understand the physical world and act within it.
Main Directions for the Next Twelve Months?
ChinaVenture: Returning to a very practical question: the company completed four rounds in six months, raising money very quickly. Where is this money mainly being used now?
Zhou: Funds are mainly directed toward three categories of expenditure. First is compute and data, accounting for about one-third of total spending. Second is personnel costs, also about one-third. The remainder is mainly early-stage non-recurring investments including operations and equipment procurement.
ChinaVenture: From a speed and rhythm perspective, these several rounds all went very smoothly. Funding negotiations and decision processes are usually lengthy; four rounds in six months looks almost seamless. What questions did investors ask you?
Zhou: It's not like that. These rounds were not evenly spaced. The first two rounds mainly looked at people and potential, so closed relatively quickly. Starting from the third round, investors would ask: you previously planned a path, what have you proven now? Relying on over a decade of industry experience, the team rapidly delivered progress in data, R&D, and training toolchain. Third and fourth round investors would personally experience stage prototypes or household scenario demonstrations, so continuous delivery and fulfilling commitments were still required.
ChinaVenture: What do investors ask most frequently? And what was the hardest question to answer?
Zhou: Most common questions include: how large is this early niche group exactly; where will the product get stuck when moving from niche to mass market; how is data privacy handled; whether there are overseas plans; and if large companies enter, what gives Ola Dimensions the right to succeed. Our answer is that the Maker group will continue expanding as tool barriers lower, and the company can also let investors periodically feel that robot development barriers are lowering.
ChinaVenture: What metrics should the outside world use to measure Ola Dimensions in the future? And what is your own development plan?
Zhou: In Q1 2027, we will advance small-scale validation and iteration of the first-generation product, determining sales and scaling rhythm based on reliability and user experience. 2027 will involve rapid iteration; single-generation production volumes won't be particularly large. By the second half of the year, based on pitfalls encountered earlier, we will form clearer mass production plans.
Additionally, two metrics matter more than sales volume. The first is success rate: the success rate of robots executing out-of-the-box high-availability tasks in households. It reflects Ola's foundation model capabilities and is the basis for users to easily train new tasks. The second is usage frequency per household and single-session duration. It can determine whether users are truly using the product, helping the robot grow, and continuing to unlock new scenarios.
ChinaVenture: I have a somewhat personal question. How do you hope people will evaluate Ola Dimensions in the future?
Zhou: I hope that one day, when people mention Ola, they will think of household embodied robots, just as Bambu Lab is to 3D printing, or DJI is to drones.
ChinaVenture: So you hope Ola Dimensions becomes synonymous with this track in the future, while also defining how this track is played?
Zhou: I believe that real tasks, user feedback, and continuous iteration are problems that all household robot routes need to solve right now.
ChinaVenture: To some extent, future industry benchmarks will also be partly set by you?
Zhou: I wouldn't put it so absolutely. Industry rules certainly need to be established together with peers. Right now, no one yet knows what robots can actually do or which users will accept them; cooperation still outweighs competition. We first need to make household robots feasible together.
Embodied intelligence ultimately won't have just one super-large company either. Autonomous driving can be seen as an embodied intelligence sub-track; in the future there will also be parallel tracks like industrial, warehouse logistics, specialized, and household, with each track forming its own industrial division of labor. Ola Dimensions hopes to become a full-stack self-developed company in the household track, similar to Wei Xiaoli.
ChinaVenture: You are a scholar-type entrepreneur, and now you're the company boss, needing to simultaneously handle funding, orders, supply chain, and organizational management. How do you view this issue? Do you have a clear remedial plan?
Zhou: Our current customers are Makers; most team members are themselves this group—we are the people who best understand user needs. The situation truly requiring remedial work is when the team doesn't even know precisely who the users are and cannot do detailed design. We are currently serving the group we know best, and through external user interviews, task testing, and how usage feedback affects product trade-offs, needs are easily aligned within the team. Early user accumulation has also exceeded expectations; they are willing to participate in co-creation—this is a good starting point.
Furthermore, we are a team spanning robotics research, consumer hardware, systems engineering, and product commercialization. Every founding team member is top talent with leading representative works, capable of standing on their own. There is full trust and clear division of labor between team members; it is everyone together supporting the company's continuous advance. For us, technical views and long-term beliefs have always been consistent: driving the democratization of robot technology and robot services. As academic and industrial progress continues, available tools will change, so methods for achieving goals will adjust.
Source: ChinaVenture
Author: Pu Fan