Ep 15 – China’s Open-Source AI Models Are Winning — And Google Still Can’t Fix Gemini

Seventy-seven companies just signed a letter asking Washington not to ban AI models — so who are they actually afraid of? Anthony Batt and Harry DeMott dig into the open-source (really, open-weights) boom coming out of China with models like Kimi K3, DeepSeek, and GLM, and why the frontier labs asking for “regulation” might be less worried about safety than about protecting their own moat. They break down the real math a CIO has to run: rent intelligence from Anthropic or OpenAI, or build your own stack on an open-weights model and control your own data and costs.

From there they turn to Google, and Harry lays out his thesis on why the stock is down even after a blowout $120B revenue quarter: 43% of searches now end with zero clicks thanks to AI Overviews, consumer Gemini keeps failing basic tasks that Claude handles instantly, and the company is fighting a six-front war — search, cloud, chips, models, hardware, YouTube — without a single focused win in AI. They also cover why $200-a-month and $20-a-month AI tiers will both have hundreds of millions of subscribers, what actually breaks for American startups if open-weight models get regulated away, and why competition — not regulation — is what makes capitalism (and AI) work.