In a recent speech at the 2026 Chongli Forum, Zhou Hongyi, founder of 360 Group, delivered a keynote address titled "From Large Models to the Era of Billions of Agents - A New Path for AI Evolution." He forecasted that 2026 will usher in a global "era of billions of intelligent agents," and China is well-positioned to seize this strategic opportunity.

Zhou pointed out that 2024 was a year of industry-wide pursuit of large models, while 2025 represents a transition and upgrade period. Although large models possess general capabilities, they often manifest as "chatbots" and struggle to directly solve complex business problems. Only by upgrading to intelligent agents can AI be truly implemented effectively.

During his speech, Zhou constructed a "Five Forces Model" - "Electricity -> Computing Power -> Intelligence + Human Power -> Productivity." He analyzed that while electricity is converted into general computing power via chips, it can create an illusion of "computing power surplus" if it fails to solve specific problems. Intelligent agents are the crucial hub that transforms general computing power into "specialized intelligence" capable of "getting work done." This transformation process relies on industry experts injecting unique "tacit knowledge" and "unwritten rules" into the models, enabling agents to develop the ability to solve practical pain points under human guidance and supervision, thereby truly closing the value loop from energy foundation to new productive forces.

Regarding the computing power market, Zhou believes the industry frequently confuses "training computing power" with "inference computing power." Training computing power serves as the entry ticket for large companies, while the real future demand explosion lies in "inference computing power."

He predicted that as intelligent agents are applied to complex tasks like short drama production and education, computing power consumption will grow exponentially. Over the next decade, computing power consumption might increase by a billionfold, driving a watershed moment in the chip market: the training field will still require top-tier general GPUs, while the inference field will develop toward specialized chips with extreme cost-effectiveness.

Zhou also revealed that 360 will soon officially launch a "short drama intelligent agent." This product aims to significantly lower the barrier for video content creation, granting ordinary people unprecedented production capabilities. Users simply need to input online novels or scripts, and the agent can generate animated blockbuster series. Zhou stated he has personally tested the product to create multiple episodes, verifying the feasibility of individual creators producing video content through "solo operations" empowered by intelligent agents, providing new tools for public innovation and employment in the new media sector.

Zhou emphasized the concept of the "intelligent agent economy" in his speech. He indicated that as agent capabilities improve, the way humans use software and the internet will undergo fundamental changes.

First, the internet will split into "two internets." One will serve humans for information consumption and entertainment; the other will serve intelligent agents for high-frequency data interaction and task execution. Future websites and apps may no longer be primarily designed for humans but will function as underlying toolkits for agents to call upon. Humans will only need to state their intent, and agents will automatically complete complex interactions and transactions across platforms.

Second, intelligent agents will reshape e-commerce and service industries. The traditional "people find goods" model will transform into an agent proxy model. For example, in e-commerce scenarios, users simply need to define their goals, and agents can automatically complete the entire process including price comparison, ordering, and payment. This will lead to numerous transactions occurring between agents (Agent-to-Agent) rather than between people and screens.

Third, new trust and settlement systems will emerge. Zhou stressed that in the intelligent agent economy, key challenges will include verifying agent identities (such as confirming whether an agent represents "JD.com" or "Taobao"), ensuring transaction security, and enabling automatic settlements. This provides genuine application scenarios for technologies like blockchain and smart contracts. Building identity authentication, contract signing, and payment settlement systems based on intelligent agents will be the core direction for developing the intelligent agent economy's infrastructure.

Zhou concluded that China, with its robust power infrastructure, comprehensive industrial system, and excellent open-source model ecosystem including DeepSeek and Tongyi Qianwen, is fully capable of seizing the strategic opportunities presented by the era of billions of agents. He called on enterprises to build an "AI-native" culture, allowing those who embrace AI to evolve into "super individuals," while simultaneously emphasizing the need to maintain security red lines, guard against risks potentially brought by collective intelligence, and promote high-quality development of China's economy.