Riskgaming How compute and AI will create next-gen superapps
82 snips
Sep 10, 2025 
Dylan Patel, the founder and CEO of Semianalysis, and Shahin Farshchi, partner at Lux Capital, dive into the fascinating future of AI and compute. They discuss the implications of GPT-5 on AGI and user applications, questioning if efficiency is overshadowing quality. The team explores the rise of super apps and innovative monetization models, reflecting on how they could reshape consumer tech. Additionally, the complexities of the semiconductor landscape are highlighted, especially in light of geopolitical tensions affecting supply chains.
AI Snips
Chapters
Transcript
Episode notes
Compute Spend Is Revenue-Driven
- OpenAI and Anthropic's fundraising signals massive compute spend and real inference ROI.
- This revenue-driven compute demand explains rapid infrastructure and chip market growth.
Router-Based Models Change Cost Tradeoffs
- GPT-5 uses a router across multiple models to optimize cost and performance.
- System-level and utilization efficiencies matter more than raw chip efficiency alone.
Manage Spiky AI Demand Strategically
- Optimize infrastructure for utilization and workload spikiness, not just peak performance.
- Use batching, tiered models, and time-shifting to reduce peak-hour compute costs.