智元彭志辉：具身智能体是AI时代最大的Token消费者

This article shares key insights from Peng Zhihui, co-founder, president and CTO of DeepSeek AI, delivered at the first Hong Kong Embodied Intelligence Industry Summit. It outlines the latest development trends and commercialization paths for the embodied intelligence industry, with key takeaways as follows:

1. Core industry insight: Embodied agents will be the largest consumers of AI tokens in the AI era. As tokens extend from the digital world to the physical world, every physical action of a robot consumes tokens, forming a closed industrial loop of data supply → token consumption → value creation. This makes embodied intelligence the most valuable core infrastructure in the token economy.

2. Industry development stage: Embodied intelligence has evolved from a technical concept to actual productivity deployment. 2026 will be a critical inflection point for large-scale industrial adoption. Once three conditions are met—breakthroughs in large model capabilities, reliable robot hardware, and a data flywheel in real-world scenarios—embodied intelligence will truly move from proof of concept to large-scale commercial application.

3. On-the-ground implementation outcomes: Peng introduced the "One Body, Three Intelligences" implementation framework. DeepSeek AI has already rolled out 10,000 robots, with over 2,000 zero-error task completions in industrial production lines. Hong Kong will serve as a key global verification hub for embodied intelligence applications.

This article sorts out industry logic, product directions and market opportunities for brands positioning themselves in the embodied intelligence track, with key takeaways as follows:

1. Industry competitive trends: The embodied intelligence industry is currently at a watershed. Competition is no longer centered on showcasing individual technical breakthroughs, but on whether products can integrate into real-world workflows and deliver consistent, measurable productivity gains. The overall industry is shifting from "selling robots" to "delivering results", and brands need to adjust their strategies accordingly.

2. Product R&D direction: Successful product commercialization must follow the "One Body, Three Intelligences" framework, where reliable hardware, motor intelligence, interactive intelligence and task intelligence are all indispensable. Products built solely on AI models or standalone hardware cannot achieve commercial deployment, so parallel capability building across all domains is required.

3. Market opportunities: Beyond industrial scenarios, mature service industries such as Hong Kong have strong high-end demand for embodied intelligence, with higher requirements for robot reliability and interactive experience. Going forward, embodied intelligence products will see broad growth potential across commercial, residential and many other scenarios, and early market entry will allow brands to seize first-mover advantage.

This article provides trend judgments, opportunity alerts and direction guidance for sellers active in the embodied intelligence sector, with key takeaways as follows:

1. Industry inflection point forecast: 2026 will be a critical inflection point for large-scale adoption of embodied intelligence. Three key conditions—large models solving perception and comprehension problems, robot hardware passing the reliability threshold, and the formation of a data flywheel in real-world scenarios—will all be met by that time, pushing the sector into a period of rapid growth. Early layout will allow sellers to capture growth dividends.

2. Market opportunity alerts: Embodied intelligence has already been deployed in actual industrial production lines, and mature service industries in Hong Kong have strong high-end demand. Leveraging Hong Kong as a hub also allows access to global markets, creating clear market opportunities across multiple scenarios.

3. Transformation directions and risk alerts: The industry has shifted from selling standalone hardware to delivering tangible value results. Sellers need to adjust their business models to provide complete solutions that integrate into customers' actual workflows. Products that only focus on individual technical demonstrations and cannot be deployed in real scenarios will be eliminated by the market, and sellers need to avoid this risk in a timely manner.

This article offers key insights for factories planning intelligent transformation or looking to join the embodied intelligence supply chain, with core takeaways as follows:

1. Product design and production demand: The core foundation for embodied intelligence deployment is stable and reliable hardware. Under the "One Body, Three Intelligences" framework, hardware reliability is the top priority. Factories capable of mass-producing high-reliability robot hardware will secure large volumes of industrial orders and clear commercial opportunities.

2. Opportunities for in-house intelligent transformation: Embodied intelligence can now work as a full member of the industrial workforce, completing tasks such as precision loading and unloading and human-robot collaboration. DeepSeek AI's robots have already achieved zero-error 8-hour shifts at Longcheer Technology's factory. Introducing embodied intelligence can effectively improve production efficiency, reduce labor costs and drive a factory's own intelligent upgrading.

3. Insights for digital transformation: The core of embodied intelligence development is building a closed-loop of real-world scenario data. Scenario data accumulated by factories during production can both feed back into in-house production optimization and support the iteration of embodied intelligence products. Long-term value can be captured through systematic data accumulation.

This article provides references on industry trends, customer pain points and business directions for service providers serving the embodied intelligence industry, with core takeaways as follows:

1. Industry development trends: Embodied intelligence has already achieved core technological breakthroughs and is about to enter the large-scale deployment phase. 2026 will see the industry inflection point, with the overall market size expanding rapidly and demand for all types of supporting service providers growing in lockstep. The sector offers broad development prospects.

2. Core customer pain points: Current core demand from industry customers has shifted from acquiring individual technologies to successfully deploying products in real workflows and delivering measurable productivity gains. Customers need end-to-end deployment support rather than single-point technical services.

3. Business expansion directions: Service providers can build their business layout around the full value chain of embodied intelligence deployment, developing solutions for links including hardware mass production optimization, model iteration, scenario data accumulation, real-world validation, and developer ecosystem building. There is clear market space for services such as scenario testing for brands and ecosystem operation.

This article provides references for ecosystem building, investment promotion and operation strategy for platform players positioning in the embodied intelligence track, with core takeaways as follows:

1. Core industry demand for platforms: The development of the embodied intelligence industry relies on open platforms to gather developers, accumulate multi-scenario data, and form a data flywheel to drive product evolution across the industry. Open platforms are core infrastructure for industrial development, with clear market positioning.

2. Platform operation and investment attraction direction: Platform ecosystem building can be structured around the "One Body, Three Intelligences" framework, attracting partners across hardware manufacturing, AI model development, scenario solutions and other segments. Investment attraction should prioritize enterprises with real-world deployment capabilities that can integrate into customer workflows, while filtering out projects that only offer technical demos without commercialization capacity.

3. Development opportunity alerts: Hong Kong is a key validation hub for embodied intelligence products and a window for exporting solutions to global markets. Platforms can connect with ecological resources in Hong Kong and the Greater Bay Area to build a global embodied intelligence industry ecosystem, capture industry growth opportunities and diversify risk from over-reliance on a single market.

This article provides the latest industry insights, development trends and business model references for researchers in the embodied intelligence field, with core takeaways as follows:

1. New research perspective for the industry: DeepSeek AI's Peng Zhihui proposed a token-based industry analysis framework, with the core argument that embodied agents are the largest token consumers in the AI era and the core interface connecting the physical world and the token economy. They can form a complete closed loop of data supply → token consumption → value creation, offering a new analytical lens for industry research.

2. Latest industry development trends: Embodied intelligence has now evolved from a laboratory concept to productivity deployment, and the core logic of industry competition has fundamentally changed. Competition has shifted from comparing individual technical capabilities to demonstrating a product's ability to integrate into real workflows and deliver sustained value. 2026 is the inflection point for large-scale application, and the four core conditions—large models, hardware, scenarios and data—are gradually maturing.

3. Reference for new business models: The new industry business model has shifted from "selling robots" to "delivering results". DeepSeek AI has proposed the "One Body, Three Intelligences" implementation framework, advancing parallel efforts in hardware mass production, model iteration, data open-sourcing and platform openness to build a self-evolving industrial system, providing a practical case reference for industrial model innovation.