A Hacker News thread asking what makes AI a bubble has tech workers airing their skepticism. The top concerns: compute costs are enormous (hyperscalers alone are pouring billions into infrastructure), revenue isn't keeping pace with spending, and valuations seem disconnected from actual commerce.
One commenter claimed OpenAI shut down Sora and canceled a Disney+ integration because of compute costs. That's false. Sora is still in active development, in limited-access red-teaming. OpenAI did demo Sora to Hollywood studios including Disney, but there's no record of a finalized integration deal that got axed. Still, the rumor took off because it confirmed what people already suspect: the math is brutal for video generation models.
Several commenters compared AI to nuclear fusion, always ten years away. Others pushed back, saying the technology already works for real tasks even if the business models haven't caught up. The divide is familiar. Some see genuine utility. Others see a pile of money burning on compute costs like what happened with Microsoft's OpenAI deal with no clear path to profitability.
For anyone building AI agents like OpenAI's Workspace Agents, the bubble question isn't abstract. It affects who gets funded, who gets hired, and which companies survive if the market tightens. The compute cost problem is real regardless of where you land on the bubble debate.