OpenAI just killed its erotica chatbot, video app, and shopping feature in the same week, and the pattern tells you everything about where AI money actually lives.

The Summary

  • OpenAI is shutting down Sora video, scaling back shopping, and scrapping adult content features after a year of experimental consumer plays
  • The pivot is clear: enterprise revenue from text generation and agent deployment beats consumer novelty features
  • Apple just opened Siri to rival chatbots Claude and Gemini, shrinking OpenAI's consumer moat while it refocuses on business tools
  • Despite the retreats, ChatGPT still commands 900M weekly users and 50M paying consumers, a base OpenAI now needs to monetize harder

The Signal

OpenAI spent 2025 trying to be YouTube, Amazon, and OnlyFans all at once. Now it's becoming Oracle. That tells you where the actual revenue models are in AI: not in consumer entertainment, but in enterprise productivity tools and agent infrastructure.

The numbers explain the retreat. OpenAI has 900 million weekly active users but only 50 million paying consumer subscribers, a 5.6% conversion rate that won't support a company burning cash on model training at this scale. Meanwhile, business customers pay for text generation and agent deployment with clear ROI tied to headcount reduction. When your enterprise client can justify the spend by cutting three analyst positions, you have a business model. When your consumer feature is an erotica chatbot that can't reliably filter out bestiality references, you have a liability.

The timing matters. Anthropic's Claude is gaining ground, and Apple just announced it will let rival chatbots integrate with Siri, directly threatening OpenAI's consumer distribution advantage. When your platform exclusivity evaporates and competition intensifies, you focus on the customers who write the biggest checks. That's not teenagers making AI videos or shoppers buying through chat interfaces. It's companies deploying agent workforces to replace $80K/year employees with $800/month API calls.

The shutdowns also reveal what actually works in AI infrastructure versus what generates headlines. Video generation (Sora) is compute-intensive with unclear monetization. Shopping features require payment processing, logistics partnerships, and regulatory compliance. Adult content carries moderation costs and reputational risk. Text generation and agent frameworks? Those are pure software leverage with enterprise budgets behind them.

The Implication

Watch for OpenAI's IPO filing in the next 18 months. This housecleaning is textbook pre-IPO focus: kill the distractions, clean up the risk profile, show revenue concentration in defensible segments. For founders in the agent economy, the lesson is clear: consumer AI features drive buzz but enterprise agent infrastructure drives revenue. Build where the business model is obvious, not where the demo is impressive. And if you're betting on a consumer AI platform, ask whether it can convert 5% of users to paid or whether it needs enterprise revenue to survive. OpenAI just gave you the answer.

Source: Axios