Huayuan Securities Ning Keyu: Dual-wheel Drive of Computing Power and AI Applications Opens New Growth Cycle

The 2026 National People’s Congress and the Chinese People’s Political Consultative Conference just concluded. The Fourth Session of the 14th National People’s Congress approved the resolution on the 15th Five-Year Plan for National Economic and Social Development, outlining a clear blueprint for China’s modernization efforts over the next five years. Meanwhile, the rapid rise of open-source AI agents like OpenClaw has prompted industry players to reevaluate the direction of computing power consumption, application scenarios, and business model evolution.

At this critical intersection of policy and industry resonance, Ning Keyu, Chief Analyst of the Computer Department at Huayuan Securities, recently told China Securities Journal that the leap from digitalization to intelligence is reshaping the investment logic of tech stocks. Infrastructure for computing power and intelligent application scenarios will jointly drive and usher in a new growth cycle.

From Digitalization to Intelligent Transformation

The recently approved “14th Five-Year Plan” explicitly emphasizes “deepening the construction of Digital China and enhancing the level of digital intelligence development,” with particular focus on cultivating emerging and future industries.

Regarding the government’s first proposal this year to “create a new form of intelligent economy,” Ning Keyu believes this differs fundamentally from the previous concept of “digital economy.”

“Digital economy mainly addresses issues like online information, process software, and network connectivity—essentially making data a production factor, enabling business operations on the cloud, networks, and platforms. In contrast, the intelligent economy goes further, focusing on systems that can perceive, understand, decide, execute, and continuously optimize,” Ning said. He noted that the government work report this year also proposed “accelerating the promotion of new-generation intelligent terminals and agents” and “cultivating new business forms and models rooted in intelligence,” indicating a policy shift from “digitalization” to “intelligentization + native development.”

In his view, the biggest breakthrough in industry implementation is no longer simply moving existing business online but reconstructing business processes, organizational efficiency, and product forms. “In the past, the main focus of the digital economy was ERP, cloud computing, and industrial internet; now, the main focus of the intelligent economy will shift to intelligent agents, smart terminals, industry reasoning infrastructure, high-quality data sets, and AI services aimed at results delivery. Digital economy improves information processing efficiency, while intelligent economy enhances decision-making and execution efficiency,” he explained.

Regarding the layout of “strategic emerging industries” in the 15th Five-Year Plan, Ning Keyu divides the investment pace into two categories for the computer industry.

“The first category is ‘infrastructure first, application expansion later,’ which aligns with the development logic of emerging pillar industries. Examples include computing power scheduling, cloud infrastructure, intelligent transformation of industry software, and network security. These directions have mature business models, clear customer budgets, and will enter order and revenue realization stages early,” Ning said. “The second category is ‘preliminary validation first, product maturity later,’ which aligns with future industry development, such as embodied intelligence and brain-computer interfaces. Currently, these are more about technical route validation, making short-term financial realization difficult, but long-term growth prospects are clearer.”

AI Narrative May Shift Toward “Task Delivery”

Recently, the open-source AI agent OpenClaw has rapidly gained popularity domestically, even leading to scenes where people queue offline to install it. The emergence of this landmark product offers a new reference for observing the evolution of AI technology routes.

“OpenClaw’s core disruptive point isn’t just larger models but shifting AI’s value assessment from ‘answering like a human’ to ‘whether the task is completed,’” Ning said. This marks a shift in AI industry storytelling from a simple “model parameter race” to “agent ecosystem competition.”

“Models are still important, but they are more like engines. The real difference-makers are toolchains, workflows, data feedback loops, user interfaces, and ecosystem network effects. Future winners may not necessarily be companies with the strongest single models but those that best integrate models, tools, data, terminals, payments, and distribution into a platform,” he explained. Industry competition will increasingly focus on task completion rates, call chain stability, tool integration depth, permission management, user retention, and ecosystem expansion.

Regarding the reasons behind OpenClaw’s explosive success in China, Ning said: “First, China has a rich and high-frequency digital life scene—payments, transportation, content, e-commerce—providing natural interfaces for AI agents. Second, Chinese users highly value efficiency tools and care a lot about whether they can get things done. Third, Chinese companies iterate quickly on applications and have strong channel and scene integration capabilities.”

More importantly, OpenClaw has propelled AI industry development from “dialogue” to “execution.” “The biggest change in the agent model is expanding AI’s computational needs from ‘low-frequency, batch, offline training’ to ‘high-frequency, fragmented, continuously online inference calls,’” Ning explained. When a task involves planning, retrieval, and tool invocation over a dozen or even dozens of steps, token consumption and inference requests will increase exponentially.

Balancing Vision and Reality

Under the dual drive of new infrastructure during the 15th Five-Year period and the high-frequency inference demand brought by OpenClaw, Ning believes three segments of the computing power industry chain deserve particular attention.

He said: “First, inference infrastructure and cloud scheduling platforms. The true ceiling for commercialization isn’t just theoretical peak computing power but the availability, billability, and elastic scalability of inference capabilities. Second, high-speed interconnection components, including switches, optical interconnects, and cluster networks. As applications evolve toward intelligent agents, system operation will rely more on concurrency, real-time response, and cross-node collaboration. Third, supporting capabilities of data centers, especially liquid cooling, power supply, and energy efficiency optimization. Future industry competition will focus on full-stack efficiency improvements rather than just equipment scale.”

On the AI application side, Ning suggests prioritizing products that have already established business models—those closest to high-frequency user needs and capable of forming a paid closed loop. He believes the hardware business model is undergoing a fundamental shift from simple equipment sales to an integrated delivery model of “hardware + models + subscription services + data feedback.” Companies that can convert one-time hardware revenue into ongoing service income are more likely to establish a competitive advantage.

Currently, the market’s focus on AI and tech innovation tracks is high, with some valuations at historically elevated levels. For the 2026 technology stock allocation strategy, Ning offers three recommendations:

First, long-term vision determines valuation ceiling. Current policies like “Smart Economy” and “AI+” provide clear macro support for the mainline of tech growth.

Second, recent financial reports determine valuation floors. As industry development deepens, the market will pay more attention to fundamentals such as order fulfillment, revenue recognition, and cash flow. Especially in high-attention sectors, stocks with weak performance realization will face significant differentiation.

Third, allocation should adhere to the principle of “mainline intact, performance prioritized, layered deployment.” He suggests allocating some positions to infrastructure and platform companies that are already in performance release phases for more certainty; others can focus on future industry directions with technological barriers and ecological advantages to retain upward potential.

(From China Securities Journal)