Skip to content

Podcast

AI Sovereignty Wars, Palantir-Nvidia Deal, SCOTUS Birthright Ruling, Newsom's CA Budget Lie

All-In with Chamath, Jason, Sacks & Friedberg

Source ↗ ← All highlights
  • Enterprises See Model Control As National Security
    • Enterprises view model weights and proprietary data as the “means of production” and fear giving that to frontier labs.
    • David Sacks and Jason Calacanis highlight Alex Karp’s point that control over compute, models, data, and “alpha” equals enterprise AI safety. (Time 0:00:20)
  • Model Providers Can Weaponize Platform Data Into Apps
    • Frontier model providers can vertically integrate into lucrative application categories by observing usage on top of their models.
    • David Sacks used Anthropic’s moves (Claude Design, Claude Code) and Figma’s complaint as an example of this predator strategy. (Time 0:05:14)
  • Host Open Models Locally To Protect Your Alpha
    • Do host and run open source models on your own hardware with an independent control plane to avoid leaking proprietary knowledge.
    • Chamath tested 8090’s harness: Claude + harness was 1.4x cheaper and 1.5x faster; open-source + harness was 16.4x cheaper (3x slower). (Time 0:10:51)
  • 8090 Benchmark Shows Massive Cost Advantage For Open Models
    • Chamath ran a real migration task with 8090’s software factory to compare models and harnesses.
    • Results: Claude alone vs Claude+harness and open-source+harness (16.4x cheaper) with tradeoff of slower runtime. (Time 0:13:23)
  • Plan To Fork Models And Build On Premises
    • Avoid trusting frontier labs with crown-jewel data; plan to fork and eventually build your own LLM on-prem.
    • Jason recommends vendors like Abacus for HIPAA use and says customers will migrate from hosted to self-owned models over time. (Time 0:22:41)
  • Open Models Protect The Entire AI Ecosystem
    • Chip makers, app builders, and enterprises share incentives to keep the model layer competitive to avoid a duopoly.
    • David Sacks argues NVIDIA and Palantir backing open models increases buyer diversity and prevents monopsony on chips. (Time 0:26:48)
  • Adopt A Hybrid Cloud And On Prem Model Strategy
    • Allocate model inference across cloud and local hardware to balance cost, latency, and sovereignty.
    • David Friedberg predicts a distributed spokes model: e.g., 70% cloud, 20% local, 10% other clouds for enterprise workloads. (Time 0:31:22)
  • AI Adoption Correlates With Hiring Not Mass Layoffs
    • Recent data shows high-AI-adopting firms tend to grow headcount and hire more entry-level roles, contradicting simple narratives of mass job loss.
    • David Sacks cites Ramp/Revelio study: top AI spenders grew payroll ~10% and entry-level roles ~12% over two years. (Time 0:43:12)
  • 8090 Customers Haven’t Cut Headcount After AI Rollouts
    • 8090 customers have not reported layoffs after AI deployments in Chamath’s experience.
    • Chamath and David Sacks both observed AI adopters tend to grow and hire rather than cut headcount immediately. (Time 0:46:02)
  • Anthropic’s Export Pause Ended After Negotiation And Personnel Shift
    • Anthropic’s Fable 5 export restriction was lifted after two weeks following negotiations and personnel changes.
    • David Sacks explained the shutdown stemmed from Dario’s “cyber weapon” claims, Amazon’s testing report, and Anthropic’s initial refusal to roll back Fable. (Time 0:51:57)
  • SCOTUS Keeps Birthright Citizenship Intact
    • Supreme Court reaffirmed birthright citizenship under the 14th Amendment, limiting executive power to change it.
    • Jason notes 255,000 children born yearly to non-citizen parents and flags Kavanaugh’s view that Congress could adjust laws instead. (Time 0:59:11)
  • Scrutinize Budget Tricks When States Claim Balance
    • Watch state fiscal math beyond headline “balanced” claims; look for one-time debt moves and revenue concentration.
    • David Friedberg shows California’s budget grew 65% since 2019 and top 1% pay half of personal income tax revenue, signaling fragility. (Time 1:21:48)