张璐 / Zhang Lu

张璐 appears in E230｜1万亿收入预期背后：英伟达的巅峰与软肋 as an investment-side guest interpreting Nvidia’s GTC claims. Her central contribution is to frame the AI infrastructure demand story through Inference as Cash Flow: training is closer to a large one-time build cost, while inference, agents, and long context create continuing token consumption.

She also treats Nvidia’s software moves, including NeMo Cloud and Agent as a Service, as efforts to shape deployment standards and expand token usage rather than merely sell application-layer services. Her caution is that inferencing chips and startup opportunities still exist, but the space narrows when Nvidia’s full-stack ecosystem, model change, and system integration absorb more of the value.

Connections

• Nvidia, Jensen Huang, and AI Infrastructure Full-Stack Moat - platform and moat frame she helps explain.
• AI Inference Cost Structure, Inference as Cash Flow, and Token per Watt - token-demand and efficiency concepts.
• Agent as a Service, NeMo Cloud, and AI Native SaaS Threat - software and agent deployment branch.
• Low-Latency Inference Chip, AI Chip Specialization, and Physical AI - areas where she discusses remaining challengers and future vertical attacks.