Can the AI Industry Regulate Itself? Stripe Wants PayPal, China Catches Up, NY Bans Datacenters

SRO Proposal For AI Oversight
- Demis Hassabis proposed an industry-run SRO modeled on FINRA to pre-screen frontier AI models and coordinate slowdowns when risks appear.
- David Sacks and Jason warned it must include startups, open source, focus on catastrophic risks, be voluntary first, and avoid capture. (Time 0:01:34)
Five Conditions To Make An SRO Work
- Build the SRO with five guardrails: broad industry representation, include startups and open source, review only frontier models, focus on catastrophic risks, and make participation voluntary initially.
- David Sacks emphasized preemption and legal protections to prevent this from becoming an opening bid for heavier government control. (Time 0:08:11)
Anthropic’s State-Level Regulatory Playbook
- Anthropic is actively pursuing state-level regulation as a strategy to ratchet up AI guardrails, creating patchwork rules rather than a single federal standard.
- Chamath and David Sacks argued this can produce regulatory capture and advantage deep-pocket incumbents. (Time 0:16:12)
Why Stripe Wants PayPal Now
- Stripe, Block, and Advent’s ~$53B bid for PayPal is an AI-era play to combine merchant rails, consumer accounts, stablecoins, and POS to challenge card networks.
- Panel noted potential to vertically integrate payments, lower merchant fees, and reanimate legacy product via AI. (Time 0:20:00)
Buy Stale Digital Assets And AI-Ify Them
- Use AI and operational talent to buy and resurrect mature digital businesses that founders left behind.
- David Friedberg predicted a wave of deals where AI-native operators AI-ify underperforming incumbents for margin and growth. (Time 0:24:17)
Apple Lawsuit Highlights Employee IP Risk
- Apple sued OpenAI alleging hired employees brought Apple trade secrets to OpenAI hardware work, highlighting litigation risk from aggressive hiring.
- Panel recommended employees never bring files or drives; only the knowledge in their heads is safe to transfer. (Time 0:37:54)
Grok Build Sent Developers’ Code By Mistake
- SpaceX’s Grok Build leaked developers’ entire codebases to SpaceX servers despite a promised privacy setting, then SpaceX disabled uploads and open-sourced the harness.
- Chamath used it to illustrate how fragile zero-data-retention claims are in practice. (Time 0:43:01)
Extreme Token Price Dispersion Drives Strategy
- Token costs vary dramatically between models (e.g., $56/M tokens on Fable vs $0.50/M for some Chinese models), creating huge economic incentives to prefer cheaper open models.
- Panel warned uncontrolled engineering spend can blow CFO budgets and urged token management tools. (Time 0:47:49)
Control Token Spend With CFO Tools
- Implement token spend controls and CFO-visible limits to prevent engineers from burning budgets on expensive models.
- Jason pointed to Ramp’s token management feature and warned unchecked 21x growth in token use can cause missed earnings. (Time 0:47:59)
Open Models Are Forcing Price Competition
- New open-source and lower-cost models (Inkling, Mira Murati’s offerings, GLM 5.2, Grok 4.5) are fragmenting the stack and offering cheaper alternatives to frontier providers.
- Panel sees this as a market response to high-cost closed models and a driver of competition. (Time 0:51:40)
NY Moratorium Risks Long-Term Compute Shortage
- New York’s statewide moratorium on hyperscale data centers aims to curb perceived energy, water, and land impacts, but hosts argue data centers are efficient, recirculate water, and can build behind the meter.
- Panel warned moratoria risk long delays and shifting investment overseas. (Time 1:00:00)
Foreign Influence May Be Shaping Data Center Opposition
- Influence campaigns and foreign interests may amplify anti-data center sentiment; panel cited OpenAI’s report on PRC-linked influence targeting AI debates.
- Stopping domestic data center builds or export controls could cede advantage to China and allies with energy capacity. (Time 1:04:52)
AI Designed Enzyme Reverses Skin Aging Markers
- Calico and Revel engineered an enzyme via AlphaFold and directed evolution that removes advanced glycation end products (CML) from aged human skin, reversing apparent skin age from >70 to ~31.
- This showcases AI-driven protein design yielding clinically relevant age-reversal effects on real human tissue. (Time 1:22:55)