Not Investment Advice 260: Nvidia GTC, AI Inference Explained, Jensen new Steve Jobs & Super Micro’s $2.5B Smuggling Scheme
8 snips
Mar 25, 2026 A fast rundown of Nvidia GTC highlights and the company’s shifting product stack. A lively take on Jensen Huang as the new tech visionary. Clear explanations of AI inference, pre-fill versus decode, and why agents are driving token demand. A wild look at Supermicro’s alleged $2.5B smuggling and the geopolitics of chip supply chains.
AI Snips
Chapters
Transcript
Episode notes
Jensen Sells An Industry Not A Product
- Jensen Huang functions like an industry-level salesperson, not a consumer product pitchman, because NVIDIA must convince a small set of major customers across the entire semiconductor supply chain.
- Trung Phan explains the "five layer cake": energy, fabs (TSMC), chips, models, and applications — Jensen must coordinate CEOs across all layers to secure capacity and capex.
OpenClaw Will Explode Token Demand
- OpenClaw (the open-source agent) will massively increase token usage because agents run continuously and perform many chained tasks versus single Q&A calls.
- Trung Phan reports his token usage likely rose ~1,000x while sending agents to fetch clips, research, and automation tasks.
Prioritize Performance Per Watt When Buying AI Hardware
- Focus on performance per watt (tokens per unit energy) when evaluating AI infrastructure purchases.
- Trung Phan notes NVIDIA pitches system-level efficiency, not just chip price, as the decisive metric for data centers.