"In fact, we’re already turning a profit."
That remark from Ge Zhenwei, CEO of AI² Robotics, in an interview with Gasgoo, punches a hole in the stereotype that robotics is a money pit. He pointed to the company’s ZhiMoFang—a modular, embodied intelligence service space focused on beverage dispensing. Covering everything from soft drinks and coffee to ice cream and cocktails, these units are now operating across 11 provinces and cities in China. With some locations generating up to 200,000 yuan in monthly revenue, the business is already profitable.
Image Source: AI² Robotics
This is no isolated case. Commercial service robots have accelerated their deployment over the past two years, with successful business models emerging in food delivery, building cleaning, and industrial inspection. Yet, another industry insider offers a reality check: "Revenue is one thing; profit is another."
Niche markets are delivering value, yet R&D spending across the industry remains sky-high. So, is the robotics sector actually making money? The answer isn’t a simple yes or no. It’s a gradual process of layered implementation and localized breakthroughs.
Vertical Scenes Lead the Way, Point Profitability Is Now a Reality
For many, the perception of robotics is still stuck on flashy demos at trade shows and endless lab work, with commercial viability a distant dream. But on the ground, a group of players focused on vertical scenarios have quietly cracked the code on profitability.
The beverage-serving robots Ge mentioned are a classic example of breaking through in a niche. These units don’t require complex full-body movement or the ability to handle random tasks in open environments. They simply make and deliver drinks from a fixed location. With highly standardized processes, stable high-frequency demand, and a clear value proposition in replacing human labor, these solutions have rapidly achieved a commercial closed loop.
Similar stories are already playing out across the commercial service robot sector. As a leading global player, Pudu Robotics has long achieved large-scale commercial deployment through mature product lines focused on delivery and cleaning.
According to Frost & Sullivan’s "2025 Global Embodied Intelligence and Commercial Service Robot Independent Market Research Report," Pudu Robotics ranks first globally in both revenue and shipments for commercial service robots. Its business spans 85 countries and regions, with cumulative shipments exceeding 130,000 units. In 2025, Pudu’s performance grew by over 100%. Behind this massive volume lies rigid demand from restaurants, hotels, and office buildings for delivery and cleaning robots. These machines can reliably replace repetitive human labor and operate 24/7, offering customers a clear return on investment and a strong willingness to pay.
DeepRobotics, leveraging its quadruped robots for inspection in power grids and fire safety, achieved a net profit of 28.68 million yuan in 2025, with a gross margin of 52.83%. It has become the first to establish a profitable benchmark in the quadruped robot space.
Why have these scenarios turned profitable first? It comes down to three forms of "certainty."
First, task certainty. Whether delivering food, cleaning floors, or selling goods, the workflow is fixed. These robots don’t need the adaptability required for general environments. Technically, this allows for optimization around a single task, keeping R&D investment manageable, failure rates low, and operational costs under control.
Second, value certainty. The core value of these robots is labor replacement, and rising labor costs are a clear long-term trend.
Third, replication certainty. Standardized scenarios mean robots don’t require deep customization. They can be deployed with minimal setup after leaving the factory. Scaling from one store to a hundred drives down marginal costs and rapidly reveals economies of scale.
Image Source: ZhiShen Technology
Notably, the scenarios achieving profitability first seem to sidestep the high-difficulty track of "general-purpose humanoid robots." Instead, they rely on wheeled chassis, fixed-base robotic arms, or quadruped inspection platforms—specialized forms solving specialized problems. Essentially, this is the commercialization of "specialized intelligence." Humanoid robots, which garner the most public attention, face high technical complexity, steep unit costs, and difficult deployment in general environments, making single-unit profitability elusive for now.
This creates a fascinating dichotomy in the industry: humanoid robots, which capture the public eye, are mostly still in the investment phase, while specialized robots quietly plowing vertical niches are already making money. Profitability in pockets is no longer in doubt. The real question is when this scattered success will spread across the entire industry—particularly to the technically complex realm of humanoid robots.
Revenue ≠ Profit: The Industry Remains in an Investment Cycle
"Government support is strong, adoption is accelerating, and applications are live, so most companies have revenue. But revenue is one thing; profit is another." That comment from an insider at a dexterous hand manufacturer, in an interview with Gasgoo, captures the industry’s current reality: revenue does not guarantee profit, and profitability in one area does not mean overall success.
The most direct cause is soaring R&D spending. The robotics industry is long past the days where assembling hardware was enough to turn a profit. The core of competition has shifted to intelligence capabilities. Moving from "one brain per machine" silos to a "one brain, many forms" universal foundation requires massive R&D investment at every step.
Developing embodied intelligence is a classic "capital-intensive, long-cycle" engineering feat, spanning three dimensions: computing power, hardware, and scenario deployment.
On the model side, training a universal embodied foundation model requires not only massive amounts of real-world physical interaction data but also large-scale simulation computing clusters, all of which demand continuous iteration and optimization based on scenario feedback.
On the hardware side, every iteration of the robot body involves extensive work in structural design, joint tuning, and sensor fusion. Tooling for prototypes, reliability testing, and mass production validation all require sustained capital injection.
On the scenario side, entering a new industry requires customized development and on-site debugging tailored to that sector’s characteristics. Adapting workflows and ensuring safety compliance demand dedicated technical teams on the ground.
Image Source: DeepRobotics
For many companies, R&D is a permanent, fixed cost. DeepRobotics, for instance, reported an R&D expense ratio of 24.98% in 2025, with cumulative R&D spending over the past three years accounting for 31.52% of revenue. Market leaders like UBTECH maintain similarly high investment intensity. Startups in the humanoid robot space, still grappling with technical hurdles, see an even larger share of revenue funneled back into R&D.
Another factor driving up R&D costs is the industry’s fragmentation. Hundreds of robot body manufacturers populate the domestic track. In the humanoid space alone, there are dozens of products with different configurations, utilizing electric, cable, or hydraulic drives, and featuring bipedal, quadruped, or wheeled motion. Communication protocols, control interfaces, and hardware parameters vary by manufacturer, with no unified industry standard yet established.
It is difficult for general intelligence solutions to achieve "develop once, adapt everywhere." Integrating with a new robot body requires redoing driver layer adaptation, kinematic calibration, and full-process joint testing—essentially semi-custom development. The more models adapted, the higher the investment in R&D manpower and time.
Beyond R&D, hardware costs remain a significant barrier to profitability. Specialized robots can profit largely because their supply chains are mature, production volumes are high, and costs are controllable. However, core components for humanoid robots—such as high-performance joint modules, dexterous hands, and high-resolution LiDAR—currently have limited production scale.
Data acquisition and operational costs are often overlooked. Unlike large language models, which can train rapidly on public internet data, robot intelligence requires feeding on real-world physical interaction data. Every valid data point must be collected by a physical robot operating on-site, making acquisition far more difficult and costly than harvesting web data.
Post-deployment operational costs are also significant. On-site setup, debugging, fault response, and system upgrades add up. In the early stages of product maturity, failure rates are relatively high, causing operational costs to further erode profit margins. Many projects may look impressive on contract value, but after deducting R&D, hardware, and operational costs, the actual profit space is very limited.
This leads to a typical industry divergence: mature product lines generate cash, while new businesses and technologies burn it. A single scenario may be profitable, yet the company as a whole cannot report a net profit due to high R&D spending. This isn’t a sign of an unhealthy industry; it is the inevitable law of technological development.
From Pockets to the Whole: Three Barriers to Scalable Profitability
Localized profitability has validated the commercial value of robots, but for the entire industry to enter a period of scalable profitability, it must cross three hurdles: technology, supply chain, and scenario. This is not a challenge for a single company, but a collective test for the entire industrial ecosystem.
The first hurdle is technology reuse. The core is amortizing R&D costs through a generalized architecture.
The industry’s early "silo" development model led to significant waste of R&D resources. The core direction of evolution is to build a layered, decoupled universal intelligence foundation: a bottom layer of unified perception, planning, and control algorithms; a middle layer of standardized system software platforms; and a top layer that connects to different robot forms and application scenarios via adaptation layers.
In industrial logic, this is the pursuit of economies of scope. The more models and scenarios a single technology foundation can cover, the lower the R&D cost allocated per unit, and the sooner the profitability tipping point arrives. Once core intelligence capabilities achieve cross-form reuse, the adaptation cycle for new products can shrink from months to weeks, dramatically boosting the return on R&D investment.
Mainstream players are all evolving in this direction. Some are extending from mobile navigation capabilities, others expanding from industrial operations, and still others rooting general large models into the physical world. Technical paths vary, but the core goal is identical: maximize technology reuse efficiency and turn R&D spending into reusable, long-term assets.
Image Source: MagicLab
The second hurdle is supply chain maturity, relying on mass production to drive down hardware costs.
High hardware costs are essentially a scale problem. With shipments in the thousands, core components can only be customized in small batches, keeping prices high. When shipments reach the scale of hundreds of thousands or millions, the supply chain’s economies of scale will fully kick in, driving down the cost of joints, sensors, and chips.
Domestic substitution for core robot components is proceeding steadily. The localization rate for traditional industrial parts like harmonic reducers and servo motors is already high. However, parts required for humanoid robots—such as high-performance joints, torque sensors, and dexterous hands—are still in the production ramp-up phase. As shipments rise, manufacturing processes mature, and yield rates improve, costs for these components are poised to fall rapidly.
On the policy front, the "Special Action for Real-Scenario Training of Humanoid Robots and Embodied Intelligence," jointly launched by the Ministry of Industry and Information Technology and the State-owned Assets Supervision and Administration Commission, is driving demand for physical robots and expanding shipment scale, thereby pressuring the supply chain to mature.
The third hurdle is scenario deepening: moving from "replacing labor" to "creating value."
Currently profitable robot scenarios mostly停留在 replacing simple, repetitive labor, which has a relatively low value ceiling. The truly high-profit scenarios of the future will come from robots performing tasks humans cannot do or do poorly, such as operations in hazardous environments, high-precision manipulation, and 24/7 fully autonomous service.
As intelligence capabilities improve, robots will evolve from simple physical executors into productive factors with autonomous decision-making capabilities, penetrating more high-value scenarios. Complex precision assembly in industrial settings, emergency rescue and inspection in special environments, and end-to-end autonomous service in commercial settings all offer unit prices and added value far exceeding current basic services, bringing richer profit margins.
At the same time, business models will shift from selling hardware alone to offering integrated solutions and long-term subscription services, significantly enhancing both the sustainability of revenue and the stability of profits.
Ultimately, profitability in the robotics industry is not a sudden arrival but a continuous process of layered emergence. Specialized robots profit first, humanoids later; simple scenarios first, complex ones later; single-point solutions first, platform ecosystems later.
Are robots really making money? The answer is: some are, and more are on the way. The players who will ultimately define the industry’s profitability rules will be those that simultaneously refine their technology foundations, streamline their supply chains, and cultivate real-world scenarios, consistently closing the commercial loop in the physical world.
