In this episode we go deep on the AI infrastructure stack. Perplexity has become the default AI search engine, growing usage fast and reportedly raising at a higher valuation. Groq is shipping incredibly low-latency inference on its custom LPU hardware and signed several enterprise deals. Modal makes serverless GPU compute dead simple for AI teams and has strong developer adoption. Ollama continues to dominate local model running with a huge open-source community. We also touched on Anthropic and OpenAI on the frontier-lab side, but the real story this week is the infrastructure layer.