消费者会为 AI 付费吗：重度用户、真实需求与订阅之外的市场

Beam may become an efficient foundation for enterprise post-training, but selective benchmarks and restricted early access do not establish parity with the strongest competing open-weight models.

Customer concentration creates material revenue exposure, yet spending cuts by rival laboratories are not necessarily representative of demand among ordinary enterprise customers.

A revenue leaderboard can reveal professional utility that traffic rankings obscure, because frequent visits and willingness to pay measure different forms of engagement.

Startups can differentiate through a proprietary model, a multi-model experience, or a narrowly defined audience whose requirements are underserved by general-purpose assistants.

A product's paid appeal may come from entertainment or a physical-world workflow, so a technology observer's unfamiliarity with its audience is not evidence of absent utility.

The top one percent of paying AI customers spend far more than the median payer, but this concentration should not be interpreted as the behavior of the top one percent of all AI users.

Low household subscription penetration is consistent with both an early adoption curve and limited demand for today's products, so it cannot by itself determine the eventual addressable market.

An assistant that reduces administrative friction for frequent travelers may offer little value to other consumers, especially when verifying its work takes as much effort as doing the task directly.

Even if many consumers resist productivity subscriptions, power users may justify higher consumption through positive returns, while entertainment products can monetize motivations unrelated to professional efficiency.

Advertising and underrepresented consumer categories leave room for alternative AI business models, but their absence from today's top rankings is an opportunity hypothesis rather than proof of future winners.