National Day Special Three: Empowering the New Journey with Intelligence – The '15th Five-Year Plan' Guides Artificial Intelligence into a New Form of Intelligent Economy

"14th Five-Year Plan": China's AI Industry Ranks Among the Global Top Tier. From 2019 to 2025, the number of AI-related companies registered increased from 84,000 to 757,000, and patent applications rose from 6,000 to 29,000. The performance gap between leading Chinese and US large models narrowed from 35.3% in May 2023 to 2.7% in March 2026. In terms of applications, by the end of 2025, a total of 748 generative AI services had been registered nationwide, 435 applications had been registered, and the user base of generative AI products reached 602 million, with a penetration rate of 42.8%. While the industry is expanding rapidly, three constraints are emerging: token usage has increased by more than a thousand times in two years, putting higher demands on intelligent computing supply, power supply, and cluster scheduling; model capabilities are becoming similar, and return on investment for enterprises remains to be verified; applications are being rapidly adopted, and data quality and algorithm governance need to be improved simultaneously.

"15th Five-Year Plan": AI will move from model capability competition, technical verification, and pilot applications to a new stage of comprehensive implementation of "AI +", large-scale industry deployment, and accelerated realization of commercial value. Large models are moving from content generation to task execution, and further extending to the physical world and scientific research. Simply relying on increasing parameter size, stacking computing power, or improving ranking scores is no longer sufficient to support sustainable industry development. Model companies need to address challenges such as inference cost, deployment efficiency, and business closed loops; application companies need to solve problems such as scenario adaptation, data governance, system integration, and continuous operation and maintenance. Therefore, industrial development requires coordinating computing power, algorithms, data, application scenarios, and governance systems to promote AI from single breakthroughs and local pilots to scalable applications that can be delivered, measured, and governed, thereby creating a new form of intelligent economy.

The "15th Five-Year Plan Outline" includes AI in the scientific and technological strategic layout. In the section on China's digital construction, systematic arrangements are made regarding computing power, algorithm, data supply, digital technology empowerment, and development ecosystem. It proposes accelerating breakthroughs in basic and core AI technologies, comprehensively implementing the "AI +" initiative, and improving the AI governance system. The focus of AI development has shifted from model research and demonstration applications to the simultaneous development of infrastructure capabilities, effective demand cultivation, and improvement of governance rules. From the perspective of industrial evolution, the following changes are particularly noteworthy:

Regarding industrial positioning

The "15th Five-Year Plan Outline" clearly states that "comprehensively implement the 'AI +' initiative and strengthen the integration of AI with technological innovation, industrial development, cultural construction, people's livelihood security, and social governance." In 2026, the government work report for the first time mentioned "creating a new form of intelligent economy." The previously released "Opinions on Deeply Implementing the 'AI +' Initiative" specified that by 2027, the penetration rate of next-generation intelligent terminals and agents would exceed 70%, by 2030 it would exceed 90%, and by 2035 it would fully enter a new stage of intelligent economy and intelligent society development. This indicates that the policy orientation of AI has evolved from being a technical tool that empowers single industries to becoming a new basic capability that supports the operation of the economy and society.

Regarding technological innovation

The plan proposes accelerating breakthroughs in basic and core AI technologies, "encouraging multi-modal, agent-based, embodied intelligence, swarm intelligence, and other technological innovations," and "exploring the development path of general AI." The policy focus is no longer limited to model parameters and individual evaluation scores, but promotes AI from understanding and generation to planning and execution, and gradually entering the physical world. Multi-modal, agent-based, and embodied intelligence will become important directions for technological research and industrialization during the "15th Five-Year Plan" period.

Regarding model supply

The plan proposes "promoting the synchronous development of general large models and industry-specific models." This means that while continuously enhancing general capabilities, it is also necessary to integrate industry data, expert knowledge, and business rules, so that model capabilities enter core business processes such as finance, manufacturing, healthcare, and government affairs, and gradually form a layered supply pattern of general models, industry models, open-source models, and edge models.

Regarding infrastructure

The plan emphasizes "accelerating the construction of a national integrated computing power network, promoting the large-scale, intensive, green, and inclusive development of computing power resources," and makes arrangements around high-quality data sets and data element markets. As inference and agent execution become the main increase in computing power consumption, the focus of computing power construction will shift from scale expansion to system improvements in inference capacity, cluster scheduling, storage-computing coordination, computing-power-electricity coordination, and low-latency services.

Regarding governance system

The plan requires "improving laws, regulations, policies, systems, application norms, and ethical guidelines in the field of AI," and "enhancing systems such as algorithm registration, transparency management, and safety assessment." Under the policy framework that emphasizes development and regulation, high-risk scenarios such as autonomous agent execution, cross-system calls, and operations in the physical world require further clarification of authority boundaries and responsibility chains to provide institutional support for large-scale AI applications.

Companies and patents are expanding simultaneously, and the industrial development foundation is continuously strengthened,

"14th Five-Year Plan": China's AI Industry Entities and Innovation Achievements Continuously Grow. From 2019 to 2025, the number of AI-related companies registered in China increased from 84,000 to 757,000, a 40.9% increase in 2025, reaching a new high in nearly a decade; as of June 4, 2026, the number of existing companies reached 288,700. During the same period, AI-related patent applications increased from 6,000 to 29,000, a 30.7% increase in 2025, with a total of 13.1 million patents. Globally, in 2025, 23.2% of papers in the field of computer science were authored by Chinese scholars, and 74.2% of approved AI patents came from China. China's internet user base is approximately 1.296 billion, providing a broad user and data base for the spread of AI applications.

Source: Sullivan Analysis

The capital side is also showing recovery. From 2022 to 2024, first-tier market financing events in the AI field continued to decline, dropping to 696 in 2024; in 2025, with the acceleration of commercialization in areas such as generative AI, embodied intelligence, AI chips, agents, and industry applications, financing events and funds increased to 1,579 and 1,504 billion yuan respectively, with early and mid-stage projects accounting for over 80%. The recovery in financing indicates that capital allocation has shifted from contraction in the early stage to multi-track deployment, and the focus has shifted from model narratives to technological iteration, productization capabilities, and application verification efficiency, aligning with the policy orientation of the "15th Five-Year Plan" to promote AI from technological breakthroughs to large-scale applications.

Source: Sullivan Analysis

China's intelligent computing scale is expected to increase from 416.7EFLOPS in 2023 to 2,781.9EFLOPS in 2028, with an average annual compound growth rate of 46.2%, significantly higher than the 18.8% growth rate of general computing power during the same period. The structure of computing power demand is also changing: as large models enter high-frequency scenarios such as search, office work, code, content generation, and intelligent customer service, inference calls will bring more continuous, frequent, and commercially relevant computing power demand; agent execution of multi-step tasks will also increase resource consumption such as network, storage, database, and secure computing.

Source: Sullivan Analysis

Token usage has become an important indicator of the activity level of AI applications and computing power consumption. At the beginning of 2024, China's average daily token usage was approximately 1,000 billion, rising to 100 trillion by the end of 2025, and breaking through 140 trillion in March 2026, an increase of more than a thousand times in two years; in 2025, the total cumulative token usage was approximately 21,100 trillion. Additionally, in 2025, the total amount of data used for AI training and inference reached 199.5EB, of which the inference data volume was 101.3EB, exceeding the training data volume.

Source: Sullivan Analysis

"15th Five-Year Plan" proposes accelerating the construction of a national integrated computing power network, promoting the large-scale, intensive, green, and inclusive development of computing power resources. In line with this direction, computing power supply will shift from single server procurement to integrated production capacity construction that includes chips, cluster networks, storage, liquid cooling, power connection, energy contracts, and scheduling software. Unit Token cost, inference efficiency, and resource scheduling capabilities will become key indicators in the competition of intelligent computing infrastructure. Computing power providers, cloud providers, and data center operators with long-term computing power contracts, low-cost power, high-density机房 delivery, and cross-regional scheduling capabilities are expected to gain more opportunities during the "15th Five-Year Plan" period.

The performance gap between leading Chinese and US large models continues to narrow, the relative gap dropping from 35.3% in May 2023 to 2.7% in March 2026. Domestic models such as DeepSeek, Qwen, and Doubao-Seed have entered the global top tier. As model capabilities become similar, companies are more focused on inference cost, response speed, tool invocation capabilities, deployment methods, and industry adaptation capabilities when selecting models. Surveys show that 71.5% of companies prefer deploying fully or partially open-source models, with 48.8% choosing partial open source to balance model controllability, data security, and deployment efficiency.

The plan proposes promoting the synchronous development of general large models and industry-specific models, which aligns with the layered pattern being formed in industrial supply: closed-source flagship models continue to define the upper limit of capabilities, open-source models drive the popularization of capabilities, industry-specific models accelerate into core businesses, and efficient architecture models reshape the cost curve. Application supply is also evolving from model API encapsulation to enterprise-level Agent engineering systems. Agent platforms, MCP and other tool protocols, and enterprise knowledge systems jointly affect whether models can integrate into real business processes. Sullivan believes that during the "15th Five-Year Plan" period, system-level SLA achievement rates, adaptation to complex production environments, and business scenario ROI will gradually replace general benchmark scores as the core criteria for evaluating vendor competitiveness, and the focus of industrial supply will shift from model availability to system operability.

"15th Five-Year Plan" proposes strengthening the integration of AI with industrial development, people's livelihood security, and social governance. The focus of application implementation has shifted from verifying technical availability to integrating into business processes and forming measurable business results. On the enterprise side, the number of registered generative AI services increased from 302 in December 2024 to 868 in April 2026; in 2025, the average daily usage of enterprise-level large models increased from 10.2 trillion Tokens in the first half of the year to 37.0 trillion Tokens in the second half, an increase of about 263%.

Source: Sullivan Analysis

From the industry structure perspective, large models are mainly applied in five scenarios: internet, finance, industrial manufacturing, consumer retail, and education, with application cases accounting for 31.8%, 14.7%, 12.2%, 8.1%, and 6.5% respectively, totaling 73.3%. Industries such as internet and finance have sufficient data accumulation, short feedback chains, and clear business indicators, allowing value verification to occur relatively quickly, and are expected to expand deployment scale first. Industries such as industrial manufacturing, energy, healthcare, and government affairs have higher value density but stricter requirements for professional knowledge, data security, and result reliability. In the early stage of the "15th Five-Year Plan," private deployment, industry model adaptation, and key process verification will still be the main approaches.

Source: Sullivan Analysis

On the consumer side, the user base of Chinese generative AI products increased from 230 million in June 2024 to 602 million in December 2025, with a penetration rate rising from 16.4% to 42.8%; in the first quarter of 2026, the monthly active users of AI-native apps reached 440 million. Usage entry has expanded from independent apps to mobile applications, phone manufacturer AI assistants, and PC applications, and usage scenarios have extended from question answering to image and video generation, text processing, lifestyle services, and entertainment interactions. The competition on the consumer side focuses on shifting from model capabilities to user entry, product experience, and scenario retention. AI phones, AI computers, and operating system-level AI capabilities will be key for terminal manufacturers to compete for entry points.

The plan lists agents as a key area of encouraged technological innovation. The "Opinions on Deeply Implementing the 'AI +' Initiative" also includes the penetration rate of agents as a quantitative target. The industrial value of agents is not only reflected in dialogue and generation but also in goal planning, task decomposition, tool invocation, process execution, and result verification. Enterprise-level Agents are suitable for prioritizing high-frequency, cross-system, well-defined, and verifiable tasks. By integrating knowledge, experience, rules, and system capabilities, personal efficiency can be transformed into reusable, auditable, and scalable organizational efficiency.

Source: Sullivan Analysis

On the ToB side, Agent commercialization will gradually deepen through pilots, integration, processization, platformization, and organizationization. Its progress depends on factors such as entry and integration costs, task frequency and single-value, SLA auditing and reusability, and budget allocation and organizational adoption. Usage volume alone does not directly equate to commercial value. Only when Agents enter stable enterprise entry points, core processes, and governance mechanisms can they be converted into annual fees, seat fees, exclusive concurrent fees, and renewal contracts. On the ToC side, general assistants expand penetration through free entry and ecosystem bundling, vertical tools generate cash flow through subscriptions, high-computing-power tasks achieve revenue through pay-per-use, and traffic-based scenarios enhance ARPU through advertising and transactions.

The plan encourages technological innovations such as embodied intelligence and swarm intelligence. Physical AI is an important carrier for extending the intelligent closed loop from the digital world to the real physical environment. Compared with generative AI and agent-based AI, physical AI needs to form a complete closed loop of perception, decision-making, verification, execution, and feedback, driving robots, autonomous vehicles, and industrial equipment to independently complete tasks. Since physical actions may have irreversible consequences, safety requirements must run through the entire system architecture and operation process.

Source: Sullivan Analysis

The scale of China's physical AI market is expected to increase from $38.5 billion in 2025 to $33.75 billion in 2035, with a compound growth rate of 24.3% from 2025 to 2035. The growth rate is expected to increase to 27.8% in 2030-2035, and the global market share is expected to rise from approximately 31% to approximately 37.5%. The main drivers are China's large manufacturing and supply chain system, ongoing policy guidance, active capital market, and diverse application needs in industrial automation, business services, and household scenarios. In the early stage of the "15th Five-Year Plan," physical AI will still focus on engineering verification and large-scale pilots in scenarios such as industrial manufacturing, warehousing and logistics, and autonomous driving. As strategic models, world models, simulation, and digital twin capabilities gradually mature, local leading companies are expected to evolve from application integrators to core technology and platform rule setters.

AI-driven scientific research (AI4S) is becoming an important lever for technological innovation during the "15th Five-Year Plan" period. The "15th Five-Year Development Plan" of the Chinese Academy of Sciences specifically addresses the transformation of the scientific research paradigm through AI, proposing to strengthen the construction of new scientific and technological infrastructure such as large-scale scientific basic models, intelligent scientist systems, scientific corpora, and scientific data centers. AI4S embeds intelligent models into key stages such as hypothesis generation, experimental design, simulation computing, and result verification, expected to shorten the knowledge discovery cycle and reduce the trial-and-error costs of complex experiments.

Source: Sullivan Analysis

From 2020 to 2025, the market size of China's AI4S platform tools increased from $14.9 billion to $29.5 billion, with a compound annual growth rate of 14.9%. It is expected that by 2030, the market will further expand to $62.5 billion, with a compound annual growth rate of 16.2% from 2025 to 2030. Simulation computing and experimental intelligence remain the core application areas, while knowledge extraction continues to grow steadily as a fundamental capability. Since entering the commercialization stage in 2023, scientific intelligent agents have accelerated their development, and it is predicted that the market size will reach $19.3 billion by 2030. During the "15th Five-Year Plan" period, the growth momentum of AI4S will shift from individual tool penetration to platform-based research processes, large-scale procurement of industrial R&D scenarios, and closed-loop scheduling of scientific intelligent agents. Additionally, potential for result transformation will be realized in fields such as life sciences, materials energy, and engineering manufacturing.

Based on the "15th Five-Year Plan Outline" and related policy arrangements, during the 15th Five-Year period, the development of artificial intelligence will advance simultaneously in basic theory and core technology research, general large models and industry-specific models, the construction of a national integrated computing power network, as well as the "Artificial Intelligence+" initiative and the improvement of the governance system. Under this context, industrial competition will shift from model capabilities to delivery and organizational capabilities. The focus of computing power demand will move from training to inference and agent execution. Application forms will evolve from content generation to task execution, further expanding into the physical world and scientific research. In the next five years, the key to enhancing China's artificial intelligence capabilities will be whether it can deeply integrate into real business processes, create measurable operational value, continuously improve industrial efficiency, promote scientific discoveries, enhance social operations, and lay the foundation for a new form of intelligent economy.