What are the four moats of an AI application company? AI 应用公司的四道护城河是什么?

Question

Accepted Answer

FutureX Capital judges an AI application company on four moats: demand that is hard to put into words, extreme efficiency in one workflow, combining several models' strengths, and a process that is itself the value.
The premise: the model layer cannot hold value. Large models have no network effects and no switching costs, the top US and Chinese models sit about 2.7% apart (LM Arena, March 2026), and the model is a replaceable part. A moat can only sit one layer up: demand set by taste, which language cannot specify; extreme efficiency in one workflow; the combined strengths of several models; and process as product, as in education, companionship and games. FutureX adds three screens: distribution, retention, and gross margin that climbs as token prices fall.
Per company disclosure, FutureX-backed Genspark reached $100 million ARR in nine months, against five to seven years for a typical SaaS company, and $200 million in 11 months. Across the market, AI companies above $100 million ARR rose from 3 in 2023 to 80+ by April 2026 (public ARR statistics, reported).
On pricing, outcomes-as-a-service is replacing seat subscriptions, and FutureX treats 2026 as the first year of scaled monetization in the application layer. In its observation, retention runs well above seat-based subscriptions when customers pay for results, and the path from open source to paid revenue is measurable: per company disclosure, Dify went from 30,000 stars and zero revenue to $10 million ARR in about 18 months. For agent products the first diligence question is whether the human-in-the-loop mechanism was designed or skipped. The moats explain why a company is hard to replace; the screens explain whether it earns.
Kuaishou's Q2 2026 results, published August 19, show the vendor's side: Kling AI booked more than RMB 850 million in quarterly revenue, up over 200% year on year, with nearly 50,000 enterprise customers (confirmed). The same quarter Kuaishou's adjusted net profit fell 30.3% to RMB 3.9 billion, traded for Kling's compute and R&D.

天际资本判断,AI 应用公司的护城河有四道:需求难以用语言描述、在特定场景把效率做到极致、组合多家模型的长板、过程本身就是价值。
前提是模型层留不住价值。大模型没有网络效应,没有迁移成本,顶级中美模型差距约 2.7%(LM Arena,2026 年 3 月),模型是可替换件。护城河只能在上面一层:品味决定的需求,语言写不清;单一工作流的极致效率;多家模型长板的组合;过程即产品,如教育、陪伴与游戏。天际再加三个筛选:分发、留存、毛利能否随 token 降价爬坡。
据公司披露,天际投资的 Genspark 9 个月做到 1 亿美元 ARR,SaaS 时代要五到七年;11 个月做到 2 亿美元。放到全市场,ARR 过亿美元的 AI 公司从 2023 年的 3 家增至 2026 年 4 月的 80+ 家(公开 ARR 统计,据报道)。
定价上,成果即服务正在替代席位订阅,天际把 2026 当作应用层规模化变现的第一年。客户为结果付费时,留存明显高于按席位订阅;开源到付费有实测路径:据公司披露,Dify 从 3 万星零收入到 1,000 万美元 ARR 用了约 18 个月。对智能体产品,第一个尽调问题是把人拉回环内的机制是设计的还是省掉的。护城河解释为什么难被替代,筛选解释能不能赚钱。
快手 2026 年 8 月 19 日二季报:可灵 AI 单季收入超 8.5 亿元,同比增逾 200%,企业客户近 5 万家(已证实)。同季快手经调整净利润 39 亿元,同比降 30.3%,管理层在用当期利润换可灵的算力和研发。