GTC大会开幕，首提“Token经济学”勘误版

Group 1: Core Insights - The report highlights the emergence of "Token Economics" as introduced by NVIDIA's CEO Jensen Huang at the GTC 2026 conference, predicting that by 2027, AI computing demand will reach $1 trillion [2][4] - The AI industry is experiencing a trend of collaborative advancement in computing power, models, and applications, with a shift towards self-developed chips and large-scale clusters [2][3] - The report emphasizes the transition of AI infrastructure towards systematization and asset-heavy models, driven by the explosive demand for inference [3][5] Group 2: Computing Power and Infrastructure - NVIDIA's GTC 2026 introduced a comprehensive hardware ecosystem aimed at reducing token production costs, including the Vera Rubin platform and long-term computing power agreements with Meta and Nebius worth $27 billion [3][4] - Tesla is advancing its Terafab wafer fabrication project to achieve vertical integration in AI chip manufacturing, reflecting a trend among leading companies to enhance their autonomous computing capabilities [3][5] - The report notes that Nebius and Meta's agreement locks in long-term computing supply, indicating a strategic shift towards securing computing resources [3] Group 3: Model Technology - The competition in the AI industry is shifting towards agent workflows and model layering, with OpenAI launching GPT-5.4 mini and nano models optimized for high-frequency, low-latency scenarios [5] - The introduction of GLM-5-Turbo by Zhizhu enhances capabilities in tool invocation and long-chain execution, indicating a move from price competition to capability pricing among model vendors [5] - MiniMax's M2.7 model demonstrates self-optimization capabilities, highlighting the importance of collaborative scheduling among models of varying costs and abilities [5] Group 4: Application Layer - Major internet companies are accelerating their deployment of AI agents, with Alibaba launching an enterprise-level agent platform and Baidu integrating agent capabilities into smart home applications [6] - The report indicates a trend where AI applications are expanding into both enterprise process execution and consumer-level high-frequency entry points, enhancing the depth and breadth of application deployment [6] - The shift from auxiliary tools to embedded, multi-scenario operational systems is emphasized as a core trend in AI application development [6] Group 5: Policy Factors - The Chinese government's "Artificial Intelligence+" action plan provides policy guidance for the AI industry, setting development goals for 2027, 2030, and 2035, which supports long-term confidence in AI concept stocks [6]