OpenAI abruptly killed its video-generation app Sora last month after burning through $1 billion, marking a stark shift toward profitability over flashy demos. Both OpenAI and Anthropic are facing what The Verge's Nilay Patel calls a "monetization cliff" — the point where hundreds of billions in investment must translate to actual revenue or the bubble pops. The catalyst isn't model capabilities, it's AI agents that customers actually want to use but consume compute at rates these companies never anticipated.

This isn't theoretical anymore. Agent workloads through products like Claude Code, Cowork, and OpenAI's Codex are burning tokens so fast that it's forcing hard decisions about which products to support and which to kill. The math is brutal: agents are valuable to customers right now, but they use far more compute than chat interfaces. When your revenue model depends on token consumption but your costs scale even faster, something has to give.

While AI safety researchers debate existential risks from superintelligence, the real existential threat facing OpenAI and Anthropic is much more mundane: becoming profitable businesses before investor patience runs out. The industry's focus on apocalyptic scenarios misses the immediate challenge of building sustainable revenue models when your most valuable products are also your most expensive to run.

For developers, this means expect more restrictions, higher pricing, and strategic product cuts as AI companies optimize for profitability over capability demos. The party phase is ending — what remains will be the products that can actually pay for themselves.