The inflection point has emerged: 70% of the value of "AI+" comes from the Internet of Things, and AI returns to the physical world.
In the past week, there have been frequent hotspots in the AI field. On August 6th, Google released the latest version of the world model, Genie 3. This model achieved real - time interactive 3D environment generation for the first time, with amazing results. Immediately afterwards, on August 8th, OpenAI officially launched GPT - 5, once again triggering intense discussions in the industry.
Actually, long before these breakthroughs, I predicted in an article that 70% of the industrial value of "AI +" will ultimately belong to the Internet of Things (IoT). At that time, this judgment was regarded by many as bold and even radical. However, as the AI industrialization process accelerates, this view is being verified by more and more facts.
Amid the wave of AI industrialization, the IoT has not been marginalized. Instead, it has increasingly become the core driving force for promoting the real - world implementation of AI and enabling various industries. It is estimated that by 2025, the number of global IoT terminal connections will exceed 27 billion. More importantly, it is these massive IoT terminals distributed in scenarios such as production, transportation, healthcare, and cities that provide 67% - 72% of the original data sources for AI applications. It can be said that the IoT has become the most solid and extensive data foundation for AI evolution and application.
This trend is confirmed by the latest breakthroughs in AI foundation models. New - generation AI systems represented by GPT - 5 and Genie 3 are gradually shifting from simply relying on virtual data such as Internet texts and pictures to actively perceiving, understanding, and even operating in the physical world.
Behind these technological updates, the value of the IoT is becoming more prominent. It is not only a data collector but also an indispensable bridge for AI to interact with, receive feedback from, and continuously learn in the real world.
Whether it is a more powerful world model or an intelligent agent capable of autonomous action, they all rely on a large amount of real - time, multi - modal, and embodied data generated by IoT terminals. This data is not only large in quantity but also rich in physical attributes, scene characteristics, and behavioral semantics, which are the keys for AI models to break through hallucinations and move towards real intelligence.
In fact, the limitations of large models are beginning to emerge. The intelligence achieved by simply expanding parameters and stacking computing power is hitting the ceiling of the virtual world: insufficient reasoning ability, lack of physical common sense, difficulty in generalization, frequent hallucinations... For AI to break through these bottlenecks, it must return to the more real and complex physical world.
The inflection point has arrived. In the next round of the intelligent revolution, the main stage will no longer be the data stacking and algorithm show - off in the virtual world, but the sinking of intelligent agents led by the IoT, and the perception, understanding, and action in the real world. The awakening of AIoT will bring higher - level intelligence into reality.
The Limit of Virtual Intelligence vs. The Starting Point of Physical Intelligence
In the past few years, Scaling Law has become the creed driving the rapid progress of artificial intelligence. As shown in the above figure, since GPT - 3, the development of large models has almost followed a simple logic of "brute - force aesthetics": The larger the parameters, the more the data, and the stronger the computing power, the closer the intelligence is to being general.
From GPT - 4, GPT - 4o to the newly released GPT - 5, each iteration has refreshed the upper limits of scale and ability. From text generation to multi - modal understanding, these models have indeed brought about amazing leaps in capabilities. However, behind the larger and stronger models, the limitations and bottlenecks are also inevitably exposed.
As the data dividend is exhausted and the computing power cost increases exponentially, the improvement of the model in terms of accuracy and generalization ability has become increasingly slow, and even shows a trend of diminishing marginal returns.
OpenAI's highly anticipated new - generation model, GPT - 5, encountered unexpected initial reactions after its release. Some early users complained that its performance was "clumsy" and even worse than that of the previous generation products.
OpenAI CEO Sam Altman quickly responded on Friday: Plus users will be allowed to choose to continue using the previous - generation version of GPT - 4o.
What is even more worrying is that the hallucination phenomenon of large models in the virtual world is difficult to control, and many facts show that AI is still "able to talk but unable to act". They are good at filling in blanks or imitating within the existing data distribution, but it is difficult for them to break out of the sandbox of the virtual world and truly understand and handle complex and ever - changing real - world scenarios.
Facts have proved that it is difficult for AI to cross the ceiling of virtual intelligence by simply stacking data and computing power. This also makes the so - called "AI + IoT" no longer an add - on but becomes the cornerstone of the intelligent agent era. AIoT not only connects all things but also endows all things with intelligence, which is the only way for AI to break through its boundaries.
Against this background, the data in the physical world has become a new gold mine for AI evolution. When the value of text and image data is approaching its limit, the real - world data collected by IoT terminals has become the "fountain of life" for promoting the leap of AI capabilities.
As shown in the above video, the launch of Genie 3 enables the world model to achieve real - time interaction in a 3D physical environment for the first time. The research and implementation of embodied intelligent agents also emphasize the ability of AI to actively perceive, operate, and receive feedback from the physical world. The essence of these latest cases is the paradigm shift of AI capabilities from the virtual to the physical world.
Only the data of perception, interaction, and feedback from the physical world can provide AI with real generalization ability and causal reasoning ability. This type of data is not only large in quantity and high in quality but also rich in scene diversity and dynamic changes, which are the keys to supporting intelligent agents to adapt to complex environments.
Although the collection, annotation, and generalization of data in the physical world face huge technological and cost challenges, the value of "scene generalization" it brings far exceeds the data stacking in the virtual world. The evolution path of AI cannot avoid a deep embrace of the physical world.
World Model × AIoT: The Rise of New Species of Intelligent Agents
In the process of AI development, "big data" was once regarded as the universal key to intelligent evolution. Countless models have obtained unprecedented expression and understanding abilities by stacking massive amounts of text, pictures, audio, and other data. However, as AI capabilities approach the limit of the virtual world, this "quantity - based" paradigm is gradually becoming ineffective. Instead, there is an extreme desire and competition for "good data". In the future, what will really drive the implementation and evolution of AI is no longer the absolute scale of data but the quality and structure of "good data".
In the physical world, "good data" has become the core bottleneck for AI perception, understanding, and decision - making. What qualifies as "good data"? First of all, it must have physical authenticity, that is, the data comes from real environments, real operations, and real feedback, and can accurately reflect the laws and dynamics of the physical world. Secondly, it should have semantic comprehensibility, not just low - level sensor signals but data with clear labels, structures, and semantic information, which is conducive to high - level cognitive processing by models. More importantly, it should have scene generalization, that is, the data can cover diverse scenarios, complex environmental changes, and boundary conditions to ensure that the model has migration and generalization abilities.
In the era of intelligent agents, "good data" is the real fuel for AI evolution and the foundation of all technological breakthroughs. Because the awakening of intelligent agents needs to be based on embodied intelligence and world models and rely on the AIoT intelligent agent network to achieve co - evolution.
Many people mistakenly think that embodied intelligence is equivalent to humanoid robots. In fact, the essence of embodied intelligence is to endow AI with the ability to actively perceive, physically interact, and self - learn. The AIoT intelligent agent is the best carrier of this ability. Whether it is factory automation, smart cities, unmanned delivery, or smart homes, AIoT intelligent agents are quietly penetrating into every corner of the physical world in a distributed and networked form.
The evolution of the world model is enabling AI to change from "able to talk" to "able to act", and to evolve from "pixel/text" processing ability to physical causality and abstract reasoning ability. Taking the new - generation world model advocated by computer scientist Yann LeCun as an example, AI is no longer just passively reconstructing data but actively predicting environmental evolution and inferring the consequences of its own actions, achieving counterfactual reasoning and zero - sample planning.
The essence of this ability is the in - depth understanding and generalized application of the laws of the physical world. All of this cannot be achieved without the active perception, distributed decision - making, and real - time feedback supported by the AIoT intelligent agent network. Each embodied intelligent agent is like an "eye" and a "hand" in the physical world, forming a super - intelligent agent ecosystem of collaboration, sharing, and evolution through the IoT network.
Ultimately, the generalization ability and adaptability of intelligent agents must rely on the physical - world closed - loop of AIoT. The world model is the foundation of cognition, and AIoT is the framework for action. Only through their collaboration can intelligent agents awaken in the physical world.
From the Battle of Hundred Models to the Intelligent Agent Economy
With the rapid evolution of AI technology, the industrial landscape is facing an unprecedented inflection point.
In the past two years, AI has rapidly expanded in the "battle of hundred models". Countless large models, applications, and platforms have emerged, trying to lead in terms of algorithms and scale. However, the windows of technological and traffic dividends are closing. The real focus of competition is shifting from the comparison of model capabilities to the control of platform - based, software - hardware integration, and data closed - loops. Large models have become infrastructure, and those who can achieve "intelligent agents as an ecosystem" in a broader industrial scenario are more likely to lead the next round of the intelligent revolution.
This shift in the focus of AI marks the in - depth evolution of the AI business model from "models as a service" to "intelligent agents as an ecosystem". In complex real - world scenarios such as factories, logistics, cities, and healthcare, a single AI model API can no longer meet the full - process requirements from perception, decision - making to execution. Enterprise and city customers are more eager for integrated software - hardware platforms to achieve end - to - end data closed - loops and continuous evolution.
Taking an automated factory as an example, only by connecting the entire chain of equipment, sensing, AI decision - making, and robot execution can a self - learning, self - optimizing, and self - managing intelligent production system be formed; the demand for the active collaboration and dynamic scheduling of intelligent agents in the logistics industry also determines the irreplaceability of platform - level AI capabilities.
What is worth noting in this process is that the mission of AIoT is being redefined. It is no longer just a networking tool or a transit station for data collection, but an enabler for each physical device to evolve into an active intelligent agent capable of perception, decision - making, and action, and to continuously produce high - value data.
The value of AIoT is rising from the foundation of digital transformation to the new infrastructure of the intelligent agent era. In cutting - edge fields such as smart factories, smart cities, and digital healthcare, AIoT has become a super - connector for the in - depth integration of AI and the real economy. The future real - world intelligent economy is essentially a globally coordinated, data - driven, and intelligence - emerging economy promoted by AIoT.
This trend is also driving changes in the industrial ecosystem. The AIoT platform, embodied intelligent models, and the Agent ecosystem are forming a trinity of resonant development. The AIoT platform provides a unified base for perception, communication, and execution. Embodied models endow each intelligent agent with autonomous learning and reasoning abilities. Various intelligent Agents continuously evolve and collaborate in specific scenarios to form a self - organizing and self - adaptive intelligent agent network.
Conclusion
Looking back at the evolution path of the AI industry, we are standing at an unprecedented historical inflection point.
The craze for large models will eventually return to rationality, and the real value of AI is accelerating its migration to the physical world. 70% of the value of "AI +" comes from the IoT. This judgment is not only verified by more and more real - world cases but has also become the most trustworthy strategic consensus for the next decade. With the awakening and maturity of AIoT infrastructure, the future of intelligent agents is being defined and led by the IoT.
For all industrial decision - makers, developers, and academic researchers, now is the best time to embrace the integration of "AI + the physical world". Whether it is to promote the intelligent upgrading of the real economy or to build new - type infrastructure for the future, AIoT has become an indispensable key cornerstone.
Looking forward, only by deeply embracing the physical world can intelligent agents truly awaken. When AI is no longer limited to the virtual space but is deeply integrated with the perception, interconnection, and intelligence of all things, the entire society and industry may enter the next golden decade led by intelligent agents. The next industrial miracle will be ignited by the spark of AIoT.
References:
1. Genie 3: A new frontier for world models, Author: Jack Parker - Holder and Shlomi Fruchter, Source: deepmind.google
2. Altman's failure in expectation management and GPT - 5's inability to achieve AGI, Source: Tencent Technology