Summary of Key Points
In 2026, the domestic embodied intelligence sector experienced a funding gap characterized by a surge in investment in "robot brains" (controllers and software), with over 50% of the funds flowing towards these components. However, companies like Xianggong Intelligence, known as the "first stock in the robot brain market," still rely on low-margin robot hardware sales to generate revenue. This mismatch—where high valuations are driven by the potential of the "brain" and profits come from the "hardware"—reflects the industry's transition from being hardware-driven to software-defined. While the value of the "robot brain" is recognized by investors, commercialization still depends on hardware. Additionally, the industry faces challenges such as a lack of standardized definitions for what a "robot brain" actually is, insufficient data, high deployment costs, and differing technical approaches. This is a necessary stage in the industry's evolution. Ultimately, those who can balance the intelligence of the "brain" with the profitability of the "hardware" will emerge victorious.
I. All the Money Goes to the Brain, but Brain Companies Rely on Hardware for Revenue?
In the first half of 2026, out of the 43.8 billion yuan in funding for the embodied intelligence sector, 50.8% went to companies specializing in robot brains, while only 12.8% went to companies that manufacture complete robots. Although investors emphasize that the "second half of robotics depends on the brain," the reality is more challenging:
- The Paradox of Xianggong Intelligence: As the "first stock in the robot brain market," Xianggong Intelligence has a high gross margin of 79.8% for its controllers (the core of the brain) and 89.3% for its software, but together these two components account for only 24.6% of its revenue. The main source of revenue comes from its hardware business, which has a lower gross margin of 38.4%, yet it accounts for 67.9% of total sales. Ironically, despite a 210% increase in controller sales over three years, the unit price has dropped from 25,900 yuan to 10,700 yuan, turning the controller from a profit-making component into a low-margin commodity.
- Financial Pressure: Xianggong Intelligence's revenue grew by 33% annually from 2023 to 2025 but still incurred a cumulative loss of 137 million yuan, and it did not turn a profit in 2026. The company has experienced negative cash flows for two consecutive years, with the payment collection period extending from 61 days to 111 days—meaning the more products sold, the harder it is to collect payments.
Why is this the case? The founders aim to develop controllers rather than complete robots, but customers (such as factory integrators) want ready-to-use robots that can be plugged in without additional customization. Therefore, the high margins from the brain component must support the lower-margin hardware business.
II. Why Does the Robot Brain Need to Rely on Hardware?
It's not that the robot brain doesn't want to be independent; it's just that the current stage of the industry doesn't allow it:
- Customer Demand: Over 80% of customers are integrators who seek complete solutions, not just individual components. For example, factories buy robots for loading tasks and don't want to purchase controllers and assemble them themselves.
- Hardware Constraints: Humanoid robots are still in the early stages of mass production, so manufacturers focus on reducing hardware costs. Third-party controllers must offer low prices to attract customers, effectively devaluing their software capabilities.
- Unmature Business Models: Revenue-generating models like algorithm subscriptions and SDK licenses have not yet become widespread, and customers are only willing to pay for tangible hardware.
In short, current robot brains need to rely on hardware as a platform to be sold.
III. What Does a Robot Brain Actually Look Like? The Industry Is Still Debating
Although investors are keen on robot brains, there is no consensus on their standard definition. There are three main technical approaches:
- VLA Model (Most Popular): This approach integrates vision, speech recognition, and action execution into one model. For example, when commanded to "pick up a cup," the robot can identify the cup and reach for it directly. This model accounts for 42% of funding in the industry and is easy to implement but has limited generalization capabilities.
- World Model: This approach aims to predict the outcomes of actions, similar to how humans think before executing them. Companies like NVIDIA and Ant Lingbo are working on this technology, which accounts for 27% of funding.
- Brain-inspired Architecture: This mimics the human brain structure, with layers for decision-making, coordination, and execution. Zhifang Square recently released the world's first brain-inspired VLA model, representing a newer direction.
Each approach has its advantages, but no one can prove which one is the best solution, as there is no consensus on the initial problems that need to be addressed by the robot brain.
IV. For the Brain to Be Independent, Three Major Barriers Must Be Overcome
Even if a technical path is chosen, several challenges must be overcome for the brain to become profitable:
- Data Gap: To make robot brains universally useful, millions of hours of real-world interaction data are needed, but currently, only 500,000 hours of such data are available globally. Moreover, investing tens of millions in data collection may only improve model performance by 5%—real-world data is extremely difficult to obtain.
- High Deployment Costs: The savings from lower hardware prices are offset by the high costs of customization. For example, VLA models cannot be retrained for different robots or scenarios, requiring additional data collection and debugging, which is time-consuming and costly.
- Uncertain Technical Paths: With multiple competing approaches, investors' valuations are based on potential future success. If the chosen path proves incorrect, the high valuations could collapse.
Without overcoming these barriers, even the most promising robot brain technologies will not generate actual profits.
V. The Gap Is Not the End Point, but a Necessary Stage in Industry Evolution
The current gap reflects the fact that investors are betting on software-defined futures, while reality is still dominated by hardware. The future success of this industry depends on two key factors:
- Balancing Brain and Hardware: Ensuring that the brain is intelligent enough to accumulate data across different scenarios and that the hardware generates sufficient revenue to support the company.
- Overcoming Barriers: Solving issues related to data, deployment, and technical approaches to turn concepts into scalable products.
Xianggong Intelligence's mascot, a slime-like character from video games, may symbolize its ambition: to become a versatile platform that transcends its initial role as a controller, evolving into a foundational component for embodied intelligence. The ultimate outcome of this industry is still uncertain, but one thing is clear—once hardware becomes homogenized, intelligent systems will be the true source of profit. The current gap is merely a temporary phase in the transition from a hardware-driven to a software-defined future.
(Note: This analysis does not constitute investment advice; market risks should be assessed independently.)