Zheshang: Agentization drives the explosive demand for tokens, and investing in AI industries from the perspectives of computing power, models, and applications.

Zheshang released a research report stating that currently, AI is transitioning from being an auxiliary tool to becoming autonomous labor force, driving a surge in Token demand through Agent transformation, and pushing the computing power ecosystem from general to efficient ASICs and premium cloud services. The investment paradigm in AI follows the principle of reducing costs and generating revenue, achieving productivity scale through extreme cost reduction, thereby creating incremental value in areas such as content creation and cutting-edge research. In 2026, the three dimensions of AI industry investment are: 1) model perspective, with longer context and memory capability breakthroughs and stronger abilities in autonomous thinking, reflection, and creativity; 2) computing power perspective, from "computing power accessibility" to "computing power inflation"; 3) application perspective, finding scenarios for rapid income growth. Zheshang's main viewpoints are as follows: - The post-training paradigm drives qualitative changes in intelligence, with Agents transitioning from tool assistance to autonomous labor force. - By 2025, the large model industry has shifted its focus from expanding parameter scale to expanding reasoning capabilities. Represented by OpenAI's o-series and DeepSeek R1, reinforcement learning has verified that model capabilities can achieve nonlinear growth in the post-training phase. Looking ahead to 2026, model evolution will focus on ultra-long context memory, multimodal perception, and world models. AI is transitioning from an auxiliary copilot to an Agent delivering results, with the potential to handle complex tasks for more than 8 hours. The enhancement of the ability to deliver end-to-end projects from fragmented tasks marks the formal transformation of AI from an efficiency tool to scalable digital labor force, substantially triggering the fourth industrial revolution. - The surge in demand for reasoning drives the restructuring of computing power cost structures, with ASIC chips and cloud resources reaching a turning point for premium realization. - With the popularization of Agent workflows, global Token usage is experiencing exponential growth, with Google and bean Token usage continuously breaking through, validating the bottom-up certainty of computing power demand. ASIC chips, such as Google's TPU v7, are gradually invading the market from general GPUs due to their advantages in single chip energy efficiency and low inference costs. With upstream core components seeing price hikes and downstream demand surging, the pricing model for cloud resources has shifted from "price for quantity" to "premium realization." Edge cloud, CDN, and AI software infrastructure supporting RAG architecture (such as vector databases, inference acceleration engines) will be the first to realize performance improvements. - The AI investment paradigm focuses on cost reduction and revenue generation, with extreme cost reduction driving the acceleration of new scenarios. - The core rule of AI commercialization lies in the deep coupling of cost reduction and revenue generation, with extreme cost reduction often serving as the starting point for revenue generation. On the cost reduction side, digital employees and AICoding are significantly reducing enterprise research and development costs by restructuring per capita productivity. On the revenue generation side, when the marginal cost of specific scenarios reaches a critical point, incremental value points will be derived (such as AI short films/comics, Agent outsourcing, etc.); meanwhile, AI is deeply penetrating high complexity systems such as AI4S and power trading to achieve precise value extraction. Risk Warning: - Delay in the commercialization of the AI industry; - Market competition risks; - Policy risks; - Geopolitical risks.