After the inaugural inclusion of the "AI+" initiative in last year's Government Work Report, this year's report during the National People's Congress and Chinese People's Political Consultative Conference (NPC & CPPCC) once again underscores the "AI+" action plan.

In particular, the Premier proposed in the Government Work Report to continuously propel the "AI+" initiative, fostering a tighter integration of digital technology with manufacturing and market advantages. This includes supporting the widespread adoption of large models and vigorously developing next-generation intelligent terminals such as smart, connected new-energy vehicles, AI smartphones and computers, intelligent robots, and intelligent manufacturing equipment.

Flashing back to 2015, the Government Work Report first introduced the formulation of an "Internet+" action plan, sparking the profound and cross-border integration of new-generation information and communication technologies—encompassing mobile internet, cloud computing, big data, and the Internet of Things—with various economic and social sectors and industries. A decade later, the transition from "Internet+" to "AI+" signifies AI's emergence as the core focus of a new global technological revolution and industrial transformation.

While "Internet+" prioritized resource integration and efficiency enhancement through connectivity, "AI+" leverages deep learning, large models, and other technologies to rebuild production processes and service models from the ground up. This evolution elevates AI from an auxiliary tool to a central driving force for industrial development, inevitably catalyzing paradigm shifts. At this critical policy juncture, this article delves into the potentially transformative impacts of this new technological wave from three perspectives: industrial chain reshaping, heightened competition, and cutting-edge trends.

Industrial Chain Reconstruction and Elevated Competition

The new wave of AI, epitomized by generative AI and large models, is benefiting a wide spectrum of value chains. For instance, surging demand for computing power is driving upgrades in the chip and cloud computing industries. Data has emerged as a core production resource, fostering markets for data governance and privacy computing. Furthermore, the deepening of intelligent applications is accelerating the transformation and upgrading of manufacturing, healthcare, content creation, finance, and other industries. However, these new technologies and applications will also prompt a redistribution of resources, with some enterprises poised to become winners in this new era.

A prime example is DeepSeek-R1, which employs efficient distillation technology to transfer the inference capabilities of large models to smaller, more efficient versions. This breakthrough enables the smaller versions to maintain outstanding performance while drastically reducing model size and computational resource requirements, making them ideal for edge deployment. As AI large models shift from the cloud to the edge, an "edge AI revolution" is quietly unfolding, benefiting edge computing manufacturers with edge AI solutions (such as Eurotech and Advantech) and chip companies deploying edge computing chips (including NVIDIA, AMD, ARM, Qualcomm, and even Intel).

Additionally, data management service providers stand as another potential beneficiary. AI cannot exist without data. To develop applications utilizing open models, adopters require substantial amounts of data for training and deployment, necessitating robust data management. Therefore, as the number of diverse AI models grows, data management will become increasingly critical. Companies like MongoDB, Databricks, and Snowflake, along with hyperscale enterprises offering related products, will thrive in this trend.

As the "AI+" initiative deepens, it not only reconstructs the supply chain but also catalyzes heightened competition among enterprises and even nations. Traditional competition primarily revolved around capital, talent, markets, and supply chains. In contrast, AI-era competition encompasses data sovereignty, computing power infrastructure, model ecosystems, and intelligent penetration, exhibiting more complex and far-reaching characteristics.

Competition Among Enterprises: From Product Competition to AI Ecosystem Competition

Historically, enterprise competition was mainly manifested in product functionality, cost control, marketing, and the like. However, in the AI era, competition centers on: ① Model capability: Whether an enterprise possesses the most advanced AI models determines the intelligence level of its products; ② Data closed loop: Whether an enterprise can establish a dedicated data flow to bolster the long-term competitiveness of its AI systems; ③ Intelligent ecosystem: Whether an enterprise has constructed an AI ecosystem closed loop, spanning from underlying infrastructure to end-user experience.

Leading companies have already taken action. For instance, Microsoft has deeply integrated with OpenAI, embedding AI into products like Office and Azure. Anthropic, the creator of the large language model Claude, has received strong support from Amazon, which deeply integrates the Claude AI platform with AWS, enabling enterprises and developers to leverage Claude AI through AWS.

As large models become pivotal for competitiveness, computing power shortages once posed a significant challenge for enterprises running AI models. Models like GPT-4, Gemini, and Llama 2 heavily rely on H100 GPUs, but H100 production capacity is severely constrained. Even tech giants like Microsoft and Meta need to reserve capacity years in advance to secure sufficient new chips. However, DeepSeek's emergence seems to have reshaped the game. As Li Guojie, academician of the Chinese Academy of Engineering, stated: "The rise of DeepSeek demonstrates that AI is no longer a brute-force game of simply stacking computing power but has entered a new era focused on optimizing algorithms and model architectures, while also emphasizing data quality and scale, and rationally increasing computing power."

As AI penetration accelerates, it is foreseeable that industries and scenarios across the board will be able to access top-notch AI services at lower costs. Consequently, the key to enterprise competition will become their ability to apply and integrate AI, i.e., how to leverage AI to comprehensively upgrade an enterprise's knowledge management, innovation speed, and decision-making efficiency.

For example, AIGC + automated customer service exponentially improves marketing and customer acquisition efficiency; AI code generation (GitHub Copilot) doubles R&D efficiency; AI-driven supply chain optimization enables precise demand forecasting and reduces waste, among other benefits. In consumer goods marketing, enterprises must start catering to the preferences of AI tools represented by large models. After all, an increasing number of customers use AI tools to select products and assist in purchasing decisions. Making AI better understand and recommend one's products is a new proposition for manufacturers.

When enterprise competition shifts from the "resources and management" dimension to the "level of intelligence" dimension, lagging enterprises will struggle to catch up.

Competition Among Countries: From Technological Competition to "Intelligent Sovereignty" Competition

If the escalation of corporate competition is reflected in AI ecosystems, computing power, and data closed loops, then the escalation of national competition is manifested in AI infrastructure, data sovereignty, and the integrity of the technological ecosystem.

Globally, AI has become the central battleground for national-level competition, with governments directly participating in AI investment and policy support.

Taking the United States as an example, upon taking office in January, Trump announced the revocation of the AI executive order signed by the Biden administration in 2023, signaling a possible significant shift in the US AI development strategy. Compared to the Biden administration, which placed more emphasis on technological regulation, Trump is clearly more supportive of development, aiming to achieve maximum competitive advantage for the US in the global AI field with minimal regulation. The day after taking office, Trump announced the launch of the "Stargate" plan, pooling massive funds into the construction of AI infrastructure within the US. While formulating a national AI strategy to ensure the leading position of companies like OpenAI, DeepMind, and Meta, it is also exerting efforts to restrict the development of its rivals through the "chip ban".

China has increasingly emphasized the safe development of AI in recent years, gradually improving relevant policies and regulations. The State Council issued the "New Generation Artificial Intelligence Development Plan," outlining the guiding ideology, strategic objectives, key tasks, and safeguard measures for China's new generation of AI development by 2030. It is worth emphasizing that China attaches great importance to the practical application of AI. For instance, during a briefing session to interpret the "Government Work Report," it was mentioned that this year's focus will be on three aspects: First, industrial empowerment, utilizing AI as much as possible in industries such as industrial and agricultural services; second, terminal applications, promoting a new round of rapid terminal development through digital technology in smart, connected vehicles, smartphones, computers, and robots; third, scenario cultivation, with a dedicated scenario demonstration initiative for new technologies, scenarios, and business formats, such as accelerating scenario applications in the low-altitude economy, education, training, healthcare, and other fields while ensuring safety.

As the "AI national team" accelerates its formation, "intelligent sovereignty" will replace "digital sovereignty" as the key to technological competition among major powers. Which country can independently train and operate local large models? Which country possesses the most advanced AI chips and algorithms? Which country has high-quality local data? Does the AI regulatory system have global discourse power?

Facing a new round of global technological competition, the AI development strategies and policy directions of each country may determine their future success or failure. DeepSeek's open-source strategy can be seen as a "breakthrough action" in the global AI field. The open-source model allows global developers to freely use and improve technology, thereby accelerating the popularization and application of AI technology in various industries. DeepSeek's open-source strategy also proves to history: In this AI race, whoever embraces open source will win the future. Nowadays, China has nurtured over 30 million open-source projects, spanning the entire industrial chain from chips to frameworks to applications. According to data from China's Ministry of Industry and Information Technology in December last year, the number of software developers in China exceeded 9.4 million, with the number of open-source participants ranking second globally.

Regarding AI regulation, Zhou Hongyi, member of the National Committee of the Chinese People's Political Consultative Conference and founder of the 360 Group, suggested: "The development of large models is tied to national fortune and should not restrict their creativity due to an excessive pursuit of accuracy. Flexible regulation of 'hallucinations' is needed to create a supportive and inclusive environment for innovation and mistakes, striving to enable more enterprises to 'replicate' the success of DeepSeek. It is recommended to draw on the successful experience of the 'safe harbor principle' and implement flexible regulation for large models and related products and services, moderately accommodating the 'hallucinations' of large models. For general mistakes caused by 'hallucinations,' avoid direct shutdown or removal from shelves, giving enterprises the opportunity to self-correct and promoting bold innovation and fearless competition among enterprises."

In the long run, AI competition will not be a continuation of traditional industries but a completely different set of game rules. Only enterprises and countries that embrace change will remain invincible in the future.

Cutting-edge Trends and Applications of AGI

From a longer-term perspective, current AI is still in the stage of "narrow AI" (ANI), i.e., task-specific models (such as ChatGPT, Claude, Gemini, etc.). However, with the evolution of large models, the exploration of AGI (Artificial General Intelligence) is accelerating, and several cutting-edge directions and applications are poised for takeoff:

AI Agent

An AI Agent is an intelligent entity capable of perceiving the environment, making decisions, and executing actions. They possess autonomy and adaptability, relying on the capabilities endowed by large models to complete specific tasks and continuously improving themselves in the process. Furthermore, different AI Agents can interact with each other to collaboratively complete complex tasks.

For example, an autonomous driving Agent can perceive information such as surrounding traffic conditions and road conditions (perceiving the environment), formulate the next action plan based on the perceived information, such as deciding whether to accelerate, decelerate, turn, etc. (making decisions), and then control the car's accelerator, brakes, steering wheel, etc., based on these decisions (taking actions).

The "White Paper on Agent Communication Network (ACN) (2024)" released by China Mobile's think tank mentions: AI Agents will become the mainstream form of large model commercialization. AI Agents not only make it possible for everyone to own virtual agents (such as digital humans, intelligent assistants, etc.) but will also propel a large number of physical embodied intelligent agents, including humanoid robots, robotic dogs, smart vehicles, intelligent drones, etc., into commercial use. Agents will be the "new citizens" of future network connections, igniting an Agent economy boom and ushering human social life into a new stage and normal state of collaboration between humans and Agents, as well as among Agents themselves.

Embodied AI

"Embodied AI" and "intelligent robots" have debuted in this year's Government Work Report, marking a significant advancement in artificial intelligence (AI) discourse.

According to the "Embodied Intelligence | CCF Expert Terminology Discussion" published on the China Computer Federation (CCF) official website in July 2023, embodied intelligence denotes an intelligent system that senses and acts through a physical body. This system gleans information, comprehends problems, makes decisions, and executes actions via the interplay between the agent and its environment, fostering intelligent behavior and adaptability.

In contrast to traditional AI, the crux of embodied intelligence lies in its "embodiment." Historically, much of AI resided solely in the digital realm, exemplified by large language models that engage in online conversations. Embodied intelligence, however, imbues AI with a tangible form, enabling it to not only converse with humans but also perform real-world tasks like serving tea, pouring water, or carrying items. Hence, humanoid robots are widely recognized as the optimal embodiment of embodied intelligence.

Traditional industrial robots, characterized by their relatively low intelligence levels, have been pervasive in manufacturing. Essentially, they are a fusion of "fixed programs" and "robotic arms." Conversely, embodied intelligent robots represent a further evolution, enhancing their "multimodal perception" and "brain decision-making" capabilities. This year, the explosion and innovation of edge models, driven by DeepSeek, have addressed some of the challenges in constructing a closed-loop cognitive system for robots in complex decision-making scenarios. This has facilitated the integration and deployment of edge models with humanoid robots, accelerating the realization of embodied intelligence at the edge. As humanoid robots edge towards mass production, 2025 is poised to be the inaugural year of embodied intelligence, with applications broadening from to industrial healthcare automation, home services, and beyond.

World Models

Currently, there is no universal definition for world models. NVIDIA's official website defines them as generative AI models that comprehend the dynamics of the real world, encompassing its physical and spatial attributes. These models utilize inputs such as text, images, videos, and motion to generate videos. Through learning, they grasp the physical properties of the real-world environment, thereby representing and predicting dynamics like motion, stress, and spatial relationships within sensory data.

AI industry insiders foresee world models as the next big wave in AI. Unlike language models that still linger at the stage of understanding human semantics, world models transcend from one-dimensional digital intelligence to three-dimensional spatial intelligence. They begin to comprehend the tangible physical world, create internal representations of how the world functions, grasp the laws of the real world, and deduce the consequences of actions accordingly.

Earlier this year, Google's DeepMind formed a new team, appointing Tim Brooks, a core member of Sora, to spearhead the development of AI models capable of simulating the physical world. Coincidentally, Jen-Hsun Huang also announced at the 2025 Consumer Electronics Show (CES) in Las Vegas the launch of Cosmos World Foundation Models, specifically designed to comprehend the physical world and predict/generate "physically aware" videos. Additionally, Fei-Fei Li's World Labs, the startup Decart, and Odyssey are also active in this domain.

Conclusion

The "AI+" initiative signifies more than just an expansion of technological applications; it represents a profound transformation of China's economic structure and social morphology. Through policy guidance, scenario innovation, and ecological synergy, AI is evolving from a mere tool to an integral infrastructure, propelling the development of new productive forces and occupying a strategic high ground in the global technology race. Looking ahead, with the proliferation of technology and the enhancement of ethical governance, "AI+" is anticipated to become a pivotal engine driving high-quality development and achieving common prosperity.

References:

• Government Work Report Deploys "AI+" Initiative, Adding Momentum to the Transformation of Old and New Growth Drivers, China Financial Network
• From "Internet+" to "AI+": Technological Innovation Leading Industrial Transformation, Baijiahao
• What Policy Guidance Does China Have for "AI+"? Interpretation by a Member of the Drafting Group of the Government Work Report, Beijing Evening News Online
• A 58% Surge in 7 Days! Why Did DeepSeek Shock the World? Li Guojie Discusses the Third Wave of AI Again, China Fund News
• The "Embodied Intelligence" First Mentioned in the Government Work Report, China is Leading the World, China Report
• "World Models" - The Next "Battleground" for AI, NVIDIA and Google Both Enter the Fray, Wall Street CN
• "Agent Communication Network (ACN) White Paper (2024)", China Mobile Think Tank