Podcast: All-In Podcast

Episode: OpenAI Misses Targets, Codex vs Claude, Elon vs Sam Trial, Big Hyperscaler Beats, Peptide Craze Date: Fri, May 1, 2026 RSS: https://allinchamathjason.libsyn.com/openai-misses-targets-codex-vs-claude-elon-vs-sam-trial-big-hyperscaler-beats-peptide-craze YouTube: https://www.youtube.com/watch?v=fpC4sbawSzQ

Listen verdict: Worth a skim; full listen if you care about AI infrastructure, OpenAI/Anthropic competition, or AI-cybersecurity. The Supreme Court/Roundup segment is also unusually interesting if you follow regulation and federal preemption.

Why it matters:

• The besties frame AI less as “software” and more as a massive industrial buildout constrained by power, data centers, and capex.
• OpenAI’s reported misses are treated as serious, but not necessarily fatal; the panel thinks coding, enterprise demand, and compute access may matter more than consumer targets.
• AI cybersecurity is moving from abstract fear to commercial deployment, with major implications for software, white-hat defense, and model security.

Key takeaways:

• OpenAI reportedly missed its 1B weekly-user target and 2025 revenue expectations, while carrying huge compute commitments; Sacks argues the product story is better than the media story because GPT-5.5/Codex appear to be regaining developer momentum.
• Chamath’s main bottleneck thesis: power, not demand, is the constraint. He expects hyperscalers to benefit because OpenAI/Anthropic will need to trade economics/control for compute access.
• Friedberg argues model efficiency is still early: pruning and smaller dynamically routed models could deliver far more inference per unit of energy.
• AI cyber is becoming a major market. Sacks says frontier models now automate elite-hacker-like work, but the same capability can harden systems if defenders deploy first.
• Chamath thinks cyber moves from “humans exploiting humans” to “machines vs. machines,” eventually forcing much of the world’s operational software to be rewritten.
• Hyperscaler capex is exploding: Jason cites roughly $725B in 2026 capex guidance across Amazon, Microsoft, Google, and Meta, with AI/cloud driving the spend.
• Sacks rejects the simple dot-com/dark-fiber analogy: he says there are “no dark GPUs” because demand for tokens and coding productivity is already pulling infrastructure forward.
• The vibe-coding deletion story becomes the cautionary tale: AI agents are powerful but still need supervision, confidence thresholds, permissions, backups, and accountable human operators.
• Friedberg explains why retatrutide is generating hype: triple agonist mechanism, large weight loss, better fat-vs-muscle profile, and promising metabolic markers.
• The Supreme Court segment centers on Bayer/Monsanto/Roundup and whether EPA labeling under FIFRA preempts state failure-to-warn claims, especially after Chevron’s demise.

Bestie positions:

• Sacks: Contrarian bullish on OpenAI product momentum; believes AI capex validates the bull case and AI is becoming synonymous with U.S. economic growth. Cautious but not panicked on AI cyber.
• Chamath: Sees power/grid access as the central AI bottleneck; thinks hyperscalers and infrastructure suppliers are the better place to “follow the dollars.” Very bullish on peptide/retatrutide potential.
• Friedberg: Focused on efficiency, pruning, model architecture, cyber/system risk, and the regulatory/legal mechanics of the Supreme Court case.
• Jason: Moderates with emphasis on business implications: OpenAI IPO risk, hyperscaler free cash flow pressure, vibe-coding risk, and practical consumer/health angles.

Notable quotes or claims:

• [09:31] Chamath: AI misses are “entirely 100% due to the supply of the power necessary to generate the output token,” not demand.
• [22:22] Sacks: Mythos/GPT-5.5 cyber “doesn’t create the vulnerabilities. It just discovers them.”
• [47:02] Sacks: Unlike 2000 fiber overbuild, “there’s no dark GPUs today.”
• [56:26] Sacks: AI agents need supervision because longer-horizon tasks are more likely to drift or go off the rails.

Topics to maybe follow up on:

• OpenAI IPO odds and whether compute commitments force new equity/control concessions.
• Hyperscaler capex suppliers: power, turbines, transformers, grid infrastructure, data center construction.
• Commercial rollout of AI-cyber tools from OpenAI, Anthropic, CrowdStrike, Palo Alto Networks, etc.
• Supreme Court ruling in the Bayer/Monsanto FIFRA preemption case and implications for EPA/FDA/USDA authority.