Podcast
Socialists Sweep NYC, China Catches Up in Coding, AI Memory Crunch, Micron's Blowout Quarter
All-In with Chamath, Jason, Sacks & Friedberg
- AI As The Ultimate Economic Leveler
- Chamath argues AI is the greatest economic leveler because it turns global knowledge into actionable expertise available to everyone.
- He compares AI to Google+search but says AI acts as a personal super-founder that amplifies individual productivity without gatekeepers. (Time 0:01:05)
- DSA Platform Seeks Radical Constitutional Overhaul
- David Sacks outlines the DSA platform as radical: abolish Senate, ICE, prisons, and replace judiciary with Congress-subordinate bodies.
- He warns these elected DSA members will use Democratic ballot access to push transformative constitutional changes. (Time 0:07:10)
- Under-16 Social Media Bans May Reduce Radicalization
- Chamath links youth radicalization to social media and notes bans for under-16s in Canada/UK/Australia appear to reduce political extremism.
- He argues age-gating removes early addiction and may produce less radicalized voters as they age into politics. (Time 0:25:12)
- Chinese Open Models Are Catching Up Fast
- Gavin and others highlight GLM 5.2 as a frontier-class open-weight Chinese model matching western frontier models on coding and long context.
- They explain distillation via API scraping and reinforcement training as a low-cost cheat sheet that narrows the frontier gap. (Time 0:45:18)
- Build A Composable Model Stack
- Gavin advises enterprises will adopt composable stacks: run open-weight models locally and route hardest queries to frontier models.
- He suggests most queries (~85%) will be handled by cheaper open models, reserving frontier tokens for edge cases. (Time 0:49:23)
- Patch Vulnerabilities Quickly Instead Of Broad Bans
- David Sacks urges not to slow U.S. AI companies with excessive regulation because China will continue advancing outside U.S. jurisdiction.
- He recommends using white-hat teams to find vulnerabilities and roll out patches quickly instead of blanket clampdowns. (Time 0:55:13)
- Memory Is The Key Bottleneck For AI
- Micron’s HBM quarter shows DRAM/HBM is the critical bottleneck for AI because memory capacity and bandwidth determine model performance.
- Micron’s supply deals lock 2026 capacity and drive dramatic revenue and pricing power for memory makers. (Time 1:01:46)
- Prioritize DRAM In AI Capex Planning
- Gavin recommends prioritizing DRAM capacity in AI infrastructure planning because memory will consume a large share of hyperscaler capex.
- He estimates DRAM could be 30–40% of hyperscaler capex next year, forcing data center economic re-evaluation. (Time 1:03:51)
- Orbital Compute Becomes Viable If Launch Costs Drop
- Gavin and Chamath compute orbital vs terrestrial cost: current terrestrial gigawatt data centers cost ~$60B including power/cooling; reusable Starship could cut launch-related portions and make orbital compute competitive.
- They model $5B launch cost to place a gigawatt in space versus $35B silicon plus $25B power/cooling on ground. (Time 1:11:50)
- Anthropic Could Command Trillion Dollar Valuation
- Gavin asserts Anthropic could trade at roughly $3 trillion valuation as a public company based on inference margins and scale.
- He argues inference-dominant models can hit ~85% gross margins, making huge public market valuations plausible. (Time 1:28:21)