OpenAI Kills Sora & Hits $100M ARR on Ads | Oura Going Public & Whoop Raises at $10BN

TL;DR

OpenAI kills Sora after single-digit millions in revenue, hits $100M ARR on ads, while Oura eyes IPO and Whoop raises $500M at $10B valuation.

Key Points

• 1.Anthropic's Claude Mythos model leaked accidentally via human error. The unreleased 10-trillion-parameter model focused on cybersecurity was exposed when staged content assets were left publicly accessible — Anthropic blamed human error, causing a 4-5% drop in cybersecurity stocks.
• 2.Agentic AI will dramatically accelerate data security breaches. Agents working 1,000x faster than humans will make proportionally more mistakes — leaking source code, PII, and sensitive data — as corners are cut in the rush to ship AI-powered products.
• 3.OpenAI killing Sora was the right call but exposed a flawed strategy. Sora made only single-digit millions in revenue while consuming approximately $1M+ per week in compute, confirming that consumer video generation was compute-intensive with negligible return.
• 4.OpenAI's $100M ARR ads business is a necessary existential bet, not a victory. With only ~5% consumer conversion rates, OpenAI needs $50-70B in annual ad revenue to justify its market cap — $100M is essentially noise compared to Google's $260B and Meta's $200B ad businesses.
• 5.OpenAI's internal drama — detailed in a WSJ story — shows CEO load-balancing as a core tax. Sam Altman spent enormous time managing feuds between Dario Brockman, Ilya, and Greg, telling each they were in charge while the company lurched through strategy reversals.
• 6.SoftBank's $40B bridge loan to buy OpenAI stock represents dangerously high leverage. SoftBank Group is already 1.5-2x levered at the equity level, meaning a 30-40% portfolio decline could wipe them out — echoing Masa Son's 85% NASDAQ loss experience in 2002.
• 7.ARR accounting practices across AI companies are misleading and often triple-counted. Anthropic uses gross revenue while OpenAI reports net; tokens get resold by multiple layers (e.g. OpenAI → Cursor → end user), with each layer recognizing the same tokens as their own ARR.
• 8.Emergent Labs' claim of reaching $100M ARR in 8 months is suspect but the product is legitimate. Suspicious billing practices — recognizing $240 annual ARR from $0 free trials — inflate numbers, yet Jason rated the vibe-coding app in the top 10% of platforms after testing.
• 9.Tranched VC rounds are being used to manufacture fake headline valuations. A lead investor might buy in at $250M pre while followers pay $1B pre simultaneously, creating a blended $600M reality — the hosts argue this gamification harms founders chasing optics over substance.
• 10.Oura is heading toward an IPO with strong fundamentals built over years from Scandinavian founders. Unlike AI-hype plays, both Oura and Whoop are described as genuinely defensible consumer hardware-software subscription businesses with real recurring revenue.
• 11.Whoop raised $500M at a $10B valuation, validating the health data wearables market. The hosts note consumer switching risk (citing Peloton's collapse) but argue fitness wearables have earned recurring revenue daily — Jason predicted 2027 as the breakout year for human health data, now looking like 2026.