OpenAI's $150B conversion, Meta's AR glasses, Blue-collar boom, Risk of nuclear war

At $150B, OpenAI trades at roughly 30–50x current revenue depending on which run-rate numbers you believe. The bull case is that massive new capital extends its lead in model performance (o1, Sora, voice), developer ecosystem, and infra, letting it become a trillion‑dollar platform across consumer, enterprise, and compute. The bear case warns that open‑source models (e.g., Llama 3.x), Meta and Google front‑door integrations, and commoditized models could erode OpenAI’s advantage, especially if data runs out and synthetic data races favor hyperscalers with ‘infinite’ budgets.

The Besties report using OpenAI’s o1 daily and describe it as a true ‘chain‑of‑thought’ system that decomposes tasks into sub‑steps, queries models multiple times, and shows its reasoning tree. This enables tasks that used to require analysts—e.g., cap table modeling, market analysis, operational analytics—to be done in minutes. They expect AI agents built on top of this paradigm to re‑wire call centers, back offices, and knowledge workflows, shifting companies from ‘lots of people plus some software’ to ‘few people plus powerful agents,’ dramatically altering hiring patterns and team sizes.

Chamath argues that the historic grip of systems of record (Salesforce, NetSuite, Workday, etc.) is weakened when agents can pull directly from raw databases, data lakes, or payment streams (e.g., Stripe→Snowflake) without expensive, monolithic applications on top. He predicts renewals will become brutal as AI‑first solutions undercut legacy pricing by an order of magnitude, even if overall software spend grows. Sacks, citing Marc Benioff, counters that enterprises still require authoritative records, security, compliance, and integration—suggesting more evolution than outright extinction—but both agree pricing will be deflationary and the “slice size” per vendor can shrink even as the total software pie grows.

OpenAI’s shift from nonprofit with a profit‑capped subsidiary to a benefit/C‑corp structure, plus Sam Altman’s rumored 7% stake (~$10B), raises ethical and legal questions. Chamath notes that if this nonprofit→for‑profit conversion passes muster, others could exploit the same pattern—build under tax‑advantaged nonprofit status, then flip for profit once dominant, or even flip back and forth to gain tax or competitive advantages. Sacks argues it is fundamentally unfair that early donor‑founders like Elon Musk (who reportedly contributed $50M under philanthropic assumptions) are not “made right” while insiders get huge equity packages.

Meta’s Orion AR glasses impress the group by offering a socially acceptable, lightweight form factor compared with bulky headsets like Apple Vision Pro. Friedberg frames a shift from ‘directed computing’ (keyboards, mice, phones) to ambient computing driven by five pillars: voice control, gesture control, eye tracking, audio response, and integrated visual overlays. Still, Chamath doubts glasses are the final, billion‑unit “AI iPhone,” suggesting we may end up with something far more minimal—like a tiny screen or big watch paired with AirPods—where voice is the primary modality and visual output piggybacks on existing screens in the environment.