Apollo 经济学家：AI 支出集中度惊人，前 10% 客户占 99.5% 推理支出
Apollo 首席经济学家的数据，前 10% 客户占走 99.5% 推理支出，头部是平均值的 1791 倍，讲清了 AI 和传统软件商业模式的差别。
Apollo 首席经济学家 Torsten Slok 统计发现，AI 领域前 10% 客户占模型推理支出的 99.5%、占 neocloud 支出的 99%。头部客户平均推理支出是普通客户的 1,791 倍，neocloud 差距约 891 倍，而传统非 AI SaaS 和 CRM 仅分别约 101 倍和 48 倍。文章指出 AI 按机器工作量而非席位数扩张，仅统计客户数会严重失真。若普通企业让智能体在客服、编码、销售、后台持续运行，基础设施需求增长将远快于客户数增长。
AI adoption is spreading but AI spending remains concentrated.
"The top 10% of customers (in AI) account for 99.5% of model-serving spend and 99% of neocloud spend, leaving the bottom 90% of firms with 0.5% and 1%."
- Torsten Slok, Chief Economist/Partner of Apollo
The average firm in that top group is spending roughly 1,791x more on inference. For neoclouds, the gap is about 891x. The same calculation is only about 101x for non-AI SaaS and 48x for CRM. Apollo Academy
That is a very different economic shape from normal enterprise software. Software scales mostly with seats.
AI scales with machine work. 2 companies can both count as "AI adopters," while 1 runs a chatbot a few hundred times a month and the other has agents making millions of model calls every day. They look identical in an adoption chart and completely different in infrastructure revenue.
So counting AI customers can now be badly misleading.
So the next big unlock should turning today's huge low-spend tail into persistent machine workloads.
If ordinary companies start running agents continuously across support, coding, sales, operations and back-office work, infrastructure demand can grow far faster than the number of AI customers, because the real unit of growth is no longer the customer in the AI era.
It is the amount of work the machines are doing.
