谁在争夺“具身大脑”？｜产业深度

This article maps out the current state of the global embodied intelligence industry, where a wide range of players are competing to secure a position in the "embodied brain" race. Key takeaways are as follows:

1. The core industrial logic mirrors that of the early Android ecosystem: players compete to define the industry's "soul" through open-source strategies. Success lets them attract hardware manufacturers for adaptation and claim the spot as a cross-hardware "public brain"; even if they fall short, they can still build developer habits and create paid use cases in downstream segments such as post-training, computing power and storage.

2. Current players fall into two main categories. Large resource-rich internet giants lead the space: Ant Group, Tencent and Alibaba have open-sourced their embodied models and toolchains one after another. Vertical unicorns have also followed suit, with companies including Zhiyuan Robot, Unitree Robotics, ZI自变量 and Galaxy General AI rolling out open-source-related initiatives.

3. The industry is still in its early stage. Core bottlenecks include difficulties in data acquisition and the absence of a unified business model. Looking ahead, intelligence across different scenarios will develop in layers, and the "public brain" is more likely to take the form of a universal standard rather than a single model.

Embodied intelligence is an emerging high-growth industry that offers brands an opportunity to capture trends in consumer and industrial upgrading. Key insights for brands are as follows:

1. On the consumer trend side, embodied intelligence pilots are already underway in C-end and offline scenarios including home services, retail and pharmaceuticals. For example, ZI自变量 has partnered with 58.com to launch a human-robot collaborative cleaning service, while Galaxy General AI has rolled out initiatives in unmanned pharmacies and smart retail. Consumer demand for flexible intelligent services is gradually growing, and early布局 allows brands to capture user mindshare ahead of competitors.

2. For industrial implementation, the core commercial value of embodied intelligence today centers on per-task cost. It only delivers a competitive advantage when the cost savings from flexible transformation exceed total investment. Brands should not rush into upgrades blindly; instead, they need to account for two types of cost based on their own use cases: environment transformation costs and intelligent upgrading costs.

3. For布局 strategy, open-source ecosystem卡位 is the current industry norm. Brands can leverage open-source models to cut R&D costs when entering this space, and also access resources from government-enterprise cooperation initiatives. Sectors including electric power and automotive manufacturing are already opening up collaboration scenarios, allowing players to secure implementation resources on a priority basis.

Embodied intelligence is still in the early stage of industrial development, bringing new incremental opportunities for sellers alongside clear risks that require mitigation. Key insights are as follows:

1. On the opportunity side, profit distribution in the embodied intelligence track has not yet been finalized. Beyond model development itself, new paid demand exists across multiple links including post-training, data collection, computing storage, simulation and deployment. Pilots are already open in scenarios ranging from home services, retail, industrial manufacturing to logistics sorting, and early卡位 can secure first-mover advantage.

2. Two proven pilot paths from leading unicorns offer actionable references: one is a horizontal open approach, where open-source models attract developers, who are later converted into paid service customers; the other is a vertical closed-loop approach, where players tie their offerings to specific scenarios to quickly acquire data and generate revenue.

3. Risk warnings: The per-task cost of embodied intelligence is still higher than that of manual labor or dedicated automation. The industry lacks unified measurement standards, and business models remain project-based rather than standardized. Sellers should not rush into large-scale capacity expansion, and should prioritize small-scale pilot validation first.

The growth of embodied intelligence opens up new directions and business opportunities for factories advancing digital and intelligent transformation. Key insights for factories are as follows:

1. In terms of production demand, non-standard production tasks in variable environments—such as complex assembly, intelligent quality inspection and irregular part sorting—are the priority implementation scenarios for embodied intelligence. It can help factories improve flexible production capacity and reduce investment in transforming fixed automated production lines.

2. For business opportunities and partnership paths: Large tech players have already fully open-sourced their embodied foundation models and training toolchains. Factories do not need to build intelligent systems from scratch, and can complete testing and adaptation at low cost by leveraging the open-source ecosystem. Meanwhile, a large number of embodied intelligence companies are actively seeking industrial scenario partnerships, allowing factories to access technical resources through collaboration and lower the barrier to upgrading.

3. Key takeaways for implementation: Factories should prioritize cost accounting when advancing intelligent upgrades, calculating both the cost of adapting environments to machines and the cost of adapting robots to existing environments. Only when the cost reductions delivered by embodied intelligence exceed the combined cost of the model and deployment should factories move forward with full implementation. Prioritize pilots for high-value non-standard tasks, and avoid replacing mature, dedicated automation equipment blindly.

The embodied intelligence industry is in the early stage of卡位 competition, bringing new business growth opportunities for all relevant service providers. Key insights are as follows:

1. Industry development trends: The "embodied brain" has become the core competitive卡位 for the entire embodied intelligence industry. Both large internet giants and vertical unicorns are ramping up resource investment, and capital markets have already priced in the future premium of general intelligence in current valuations. Demand will keep growing in supporting segments including data collection, post-training services, simulation deployment, and standard setting.

2. Core customer pain points: Two core industry pain points remain unsolved. First, embodied data is scattered across physical scenarios, and cannot flow freely due to restrictions around real-time performance, data security and network conditions. Second, the industry lacks a unified unit of measurement comparable to tokens for large language models, so business models cannot escape the project-based model and achieve standardized reuse and cost dilution.

3. Reference solution paths: Service providers can choose between two entry strategies. They can either build a horizontal open model and tool service, attracting ecosystem customers through open source before converting them into paid users; or build integrated solutions for vertical scenarios to achieve fast implementation and generate revenue. They can also build infrastructure related to universal standards to secure future industry卡位.

The rise of the embodied intelligence industry opens up new growth space for all types of industrial and cloud platforms. Platforms can adjust their布局 to align with industry demand. Key insights are as follows:

1. New industry demand from platforms: As the industry develops, customer demand has expanded beyond the early homogeneous computing power supply to cover multiple links including data processing, simulation training, post-training tools, and interface adaptation. Customers now expect platforms to provide integrated supporting services, rather than simply selling computing power.

2. Latest卡位 strategies from leading platforms: Today's top cloud platforms use open-source embodied foundation models to attract hardware manufacturers and developers to adapt to their ecosystems, locking users into their cloud infrastructure ecosystem and expanding into more paid service scenarios. In the future, the core of platform competition will be the ability to aggregate data and experience from scattered projects into reusable assets, and convert these into sustained platform revenue.

3. Operation and risk warnings: Platforms should not simply copy the development path of large language models. Embodied intelligence requires deep engagement with physical scenarios such as factories and warehouses, so platforms need to build out service networks gradually. Since both technology and business models remain immature, platforms should prioritize pilot recruitment using scenarios open through government-enterprise cooperation, build capabilities incrementally, and avoid the risk of blind expansion.

This article maps the latest industry developments in the "embodied brain" track, raises a number of new open research questions, and summarizes the business models of current players. Key insights are as follows:

1. Latest industry developments: Across the embodied intelligence sector, all types of players now view the "embodied brain" as their core卡位 direction, sparking an industry-wide open-source wave. Industry players are shifting their core role from simply selling training resources to building ecosystems. Scenario implementation is clearly driven by government and enterprise initiatives, and capital markets have already priced future general intelligence premiums into current valuations.

2. Open research questions: Several core industry problems remain unsolved. First, embodied data is scattered across the physical world and difficult to collect and flow due to multiple restrictions. Second, the lack of a unified value measurement unit traps business models in a project-based structure and prevents standardization. Third, current per-task costs still do not deliver an advantage for large-scale replacement, so the industrial value has not yet translated into actual benefits for customers.

3. Summary of existing business models: The industry has currently formed four player paths: the ecosystem model of large players, which layout across all segments to sell computing power and tools; the in-house brain model of hardware manufacturers, which builds proprietary brains to break through shipment bottlenecks; the platform model of model vendors, which build open horizontal ecosystems to convert to paid services; and the vertical integrated closed-loop implementation model of scenario-focused players. All four paths have pros and cons, and remain to be validated by the market in the future.