推理效率革新与Agent共振，打开万亿市场空间

Investment Rating - The industry investment rating is "Buy" with a previous rating of "Buy" as well [2]. Core Insights - The AI inference market is projected to reach a potential market size of $1 trillion by 2027, driven by the shift from training performance breakthroughs to optimizing inference efficiency [13][19]. - The demand for inference computing power is expected to grow structurally, with Deloitte predicting that the global inference workload will increase from approximately one-third of AI computing power in 2023 to about two-thirds by 2026, and potentially exceeding 80% in the long term [13][15]. - NVIDIA is transitioning from being a "shovel seller" in AI to an "AI factory," emphasizing the importance of integrated AI systems and infrastructure [19][30]. - The introduction of the NemoClaw platform aims to create a secure and controllable environment for AI agents, marking a shift towards "Agentic AI as a Service" [30][39]. - The physical AI sector is advancing with the release of the Cosmos world model and the Isaac simulation framework, which are expected to accelerate the deployment of robotics and autonomous driving technologies [41][46]. Summary by Sections Section 1: GTC Narrative Changes - AI inference chips are opening a potential market space of $1 trillion by 2027, with a focus on reducing inference costs and optimizing efficiency [13][19]. - The shift in AI narrative emphasizes the importance of inference capabilities over mere training performance [19][30]. Section 2: AI Applications - The introduction of NemoClaw enhances the OpenClaw ecosystem, providing a secure environment for AI agents [30][39]. - The physical AI domain is being bolstered by the Cosmos model, which integrates data generation and simulation for robotics [41][46]. Section 3: Investment Recommendations - The narrative in the AI industry is shifting from training-driven to inference-driven, with a focus on the cost-effectiveness and stability advantages of integrated AI solutions [47][48]. - The rise of Agent as a Service is expected to replace traditional enterprise SaaS products, indicating a collaborative evolution rather than a complete disruption of existing software [47][48].