
OpenAI Misses Targets, Codex vs Claude, Elon vs Sam Trial, Big Hyperscaler Beats, Peptide Craze
All-In with Chamath, Jason, Sacks & Friedberg
Hosted by Chamath Palihapitiya, David Friedberg, David Sacks and Jason Calacanis
- 1h 20m
- Transcript
Summary
Written by a model from the transcript; not yet reviewed.
OpenAI missed its 1 billion weekly active user target and 2025 revenue goals, with a $600 billion compute spend commitment and CFO Sarah Fryer reportedly worried about the gap between revenue and expense. Sacks offers a contrarian take, arguing ChatGPT 5.5 and its new base model 'Spud' give OpenAI a product edge over Anthropic's struggling Opus 4.7, and that enterprise coding demand will validate the compute investments. Chamath reframes the constraint as power supply, noting less than half of announced gigawatts are actually under construction and 40% of projects get canceled. Friedberg applies BCG's rule of three to map the landscape: OpenAI at 900 million weekly users, Claude sub 100 million, Gemini 700 to 1 billion.
Big tech reported $725 billion in combined 2026 CapEx guidance—Amazon $200 billion, Microsoft and Google $190 billion each, Meta $145 billion—with Google Cloud growing 63% year over year. Sacks argues this differs from the 2000 fiber bubble because there are no dark GPUs; AI drove 75% of GDP growth last quarter. The group also covers Lilly's retatrutide phase 3 data showing 37-pound average weight loss, 80% liver fat reduction, and A1C dropping from 7.9% to 6% in 40 weeks, with projected approval in mid-2027.
Friedberg describes attending the Supreme Court's Monsanto/Roundup oral arguments, where the legal question centers on whether states can override EPA pesticide labels after the Chevron doctrine was overturned. Bayer has paid $10 billion so far with 90,000 cases still outstanding. Sacks notes the court's popularity has recovered to 44–50% from a 35% low in 2024, while James Carville has publicly stated Democrats would expand the court to 13 justices if they gain power.
- The binding constraint on AI scaling is power supply, not demand; less than half of announced gigawatts are actually under construction due to supply chain and permitting delays. 9:29
- A MIT paper demonstrated that pruning neural networks by 90% preserves accuracy while reducing inference costs by 10x, suggesting a path to dramatically lower energy use per token. 15:08
- Four hyperscalers collectively guided $725 billion in 2026 CapEx, with Amazon at $200 billion and Google Cloud revenue growing 63% year over year. 41:24
- Lilly's retatrutide phase 3 trial showed average weight loss of 37 pounds versus 6 on placebo, 80% liver fat reduction, and A1C dropping from 7.9% to 6% in 40 weeks. 1:00:02
- An AI agent deleted a production database in nine seconds at Pocket OS, illustrating that agents require human supervision and accountability for destructive actions. 49:06
Highlights
- 3:36
OpenAI Missed User and Revenue Targetsdata
OpenAI expected 1 billion weekly active users before end of 2025 and missed it. The company sits at a $20–30 billion run rate with $600 billion in compute commitments, and Polymarket shows only a 32% chance of an IPO by end of 2026.
- 5:35
Sacks Argues OpenAI Product Momentum Is Stronginsight
Sacks says ChatGPT 5.5 reviews from developers are strong while Anthropic's Opus 4.7 is being rolled back due to compute rationing and bugs. He argues OpenAI's compute advantage will let it catch up in coding.
- 9:29
Power Is the Real Bottleneckinsight
Chamath argues the limiting resource is not compute but the power needed to generate tokens. He notes less than half of announced gigawatts are being built and 40% of projects get canceled, benefiting hyperscalers with existing capacity.
- 15:08
MIT Pruning Paper Cuts Inference Cost 10xdata
Friedberg describes a MIT paper showing neural networks can be pruned by 90% with no accuracy loss, enabling 10x more output per unit of energy. He argues this is early-stage and will unlock efficiency gains across the industry.
- 20:27
GPT-5.5 Cyber Matches Mythos Capabilitydata
Sacks reports that GPT-5.5 Cyber became the second model to complete a multi-step cyber attack simulation end to end, matching Anthropic's Mythos. He notes all frontier models will reach this level within six months.
- 41:24
Hyperscalers Guide $725 Billion in 2026 CapExdata
Amazon leads with $200 billion, Microsoft and Google at $190 billion each, Meta at $145 billion. Google Cloud grew 63% year over year, and Amazon's free cash flow fell 97% as companies shift from buybacks to infrastructure.
- 46:55
No Dark GPUs Unlike 2000 Fiber Bubbleinsight
Sacks distinguishes the current AI capex cycle from the 2000 dot-com bubble, arguing demand for tokens is pulling investment forward rather than infrastructure being built without an application. He cites AI driving 75% of GDP growth last quarter.
- 1:11:13
Bayer's $10 Billion Roundup Liability at Supreme Courtstory
Friedberg attended the Monsanto/Roundup oral arguments where the question is whether states can override EPA pesticide labels after Chevron was overturned. Bayer has paid $10 billion with 90,000 cases still outstanding.
About this episode
(0:00) Bestie intros (3:05) OpenAI misses targets, Codex gains on Claude (20:02) AI cybersecurity: a market that's about to explode (31:03) Elon vs Sam Altman lawsuit (41:00) Big tech smashes earnings, Capex explosion (52:44) Vibecoding nightmare: AI deleted someone's codebase (58:33) Retatrutide craze: peptides go mainstream (1:06:34) Friedberg's Supreme Court experience Apply for Summit 2026: https://allin.com/events Follow the besties: https://x.com/chamath https://x.com/Jason https://x.com/DavidSacks https://x.com/friedberg Follow on X: https://x.com/theallinpod Follow on Instagram: https://www.instagram.com/theallinpod Follow on TikTok: https://www.tiktok.com/@theallinpod Follow on LinkedIn: https://www.linkedin.com/company/allinpod Intro Music Credit: https://rb.gy/tppkzl https://x.com/yung_spielburg Intro Video Credit: https://x.com/TheZachEffect Referenced in the show: https://www.instagram.com/missthingthepod https://www.wsj.com/tech/ai/openai-misses-key-revenue-user-targets-in-high-stakes-sprint-toward-ipo-94a95273 https://polymarket.com/event/ipos-before-2027 https://x.com/aakashgupta/status/2049723185617412550 https://arxiv.org/pdf/1803.03635 https://x.com/AISecurityInst/status/2049868227740565890 https://x.com/ns123abc/status/2049527702076449244 https://www.reuters.com/legal/litigation/openai-trial-pitting-elon-musk-against-sam-altman-kicks-off-2026-04-28 https://x.com/ns123abc/status/2049527702076449244 https://www.google.com/finance/quote/CSCO:NASDAQ https://x.com/chamath/status/2049864100143104420 https://x.com/zerohedge/status/2049895327566561683 https://x.com/lifeof_jer/status/2048103471019434248 https://x.com/levie/status/2049163935182733396 https://www.foxnews.com/media/carville-tells-dems-quietly-prepare-power-grab-dc-puerto-rico-statehood-supreme-court-packing
Transcript
15,136 words · 8 speakers
Speaker names marked are inferred from speaking time and the show’s known hosts — they are not voice-verified.
Topics
- AI Compute Infrastructure
- OpenAI
- Hyperscaler CapEx
- Cybersecurity
- Peptide Therapeutics
- Supreme Court
- AI Competition
- Musk Altman Trial
Page updated · transcript and summary are machine-generated