AI时代的反共识:为什么AI时代，要做千年不变的生意？The AI-Era Contrarian Take: Build Businesses That Never Change
Translated from the Chinese original, first published on WeChat「世像」on April 12, 2026.本文 2026.04.12 首发于微信公众号「世像」。
导读
AI产业高度技术集中。
真正超额利润在模型层，算力层，芯片层，底层infra全是工程主导。
α很可能被压缩。
进入门槛会越来越低，竞争极度激烈，同质化严重。
AI的风险在于：技术进步压缩商业护城河
回顾
去年伦敦马拉松，第40公里。
双腿灌铅，呼吸像拉风箱，每一步都在和身体对抗。但就在那个瞬间，脑子里突然蹦出一个问题：所有人都在焦虑AI会取代什么，但有没有人想过——什么东西AI永远做不到?
AI可以生成一份完美的马拉松训练计划，可以根据你的体能数据优化配速策略，甚至可以模拟撞墙后的生理反应。现在甚至有公司在做VR马拉松——你戴上头盔，跑步机会模拟波士顿的赛道起伏，屏幕里有虚拟的观众为你欢呼。
法国哲学家鲍德里亚管这个叫"拟像"——虚拟的东西越来越逼真，甚至比真实还要"真实"。
但我后来想明白了一件事：再逼真的拟像，也骗不了你的身体。你的肌肉知道这不是真的。你的意志知道这不是真的。真正的马拉松，是你的大腿在撕裂，是你想放弃但咬牙坚持，是你冲过终点线那一刻，瘫在地上动不了的那种真实的疼痛和释放。这种身体的真实体验，是AI的最后边界。
那个瞬间我想明白了一件事：体育如此，金融亦如此。
我是一个每年跑马拉松的人，也是一个正在做Fintech创业的人。过去的无数个 moment 和碎片，我意识到一件事：它们好像不是两件事，它们是同一件事的两个侧面。底层逻辑完全一样：技术会变，但人类的底层需求千年不变。
这篇文章，是我这三年在跑道上、球场上、创业路上想明白的一些事。
以下为正文，enjoy：
01 跑道上的商业课
我其实忘记了我是什么时候开始爱上体育的。
但我后来复盘的时候，发现了热爱体育的其中一个原因：我从小其实是一个很没有自信的人。唯一有自信的时候就是在运动场和田径场上。这个时候，我会觉得整个天下都是我的。
多年以来，竞技体育给我最潜移默化的影响是：凡事都要拼尽全力，不到最后一刻出结果绝不认输和放弃。这个在某种程度上构成了我的性格底色。
我的梦想一直是当一名运动员。熟悉我的人都知道，我是一名狂热的运动爱好者：篮球、跑步、网球、羽毛球、乒乓球。主流的运动几乎都看，都玩。
虽然一直未能成为真正的运动员，但是体育的确赋予了我自信。并让我明白：有胆量的人不是光有胆量，而是连目标都和其他人不一样，因为他们想要在同样的时间内获取更大的回报，才决定了他们在行动上要杀伐果断。
别小看死磕，这是心力的重要组成部分，无论是带领球队赢球还是交易达成，核心就是死磕，学一堆套路最终都比不过死磕的效果，因为战场上、牌桌上没有人是傻子。最狠的心理武器是你强烈的企图心，想要把这件事做好做完。
近几年，时间和精力逐渐从球类运动转向了田径。跑步是所有运动里最考验绝对天赋的，也是最容易向内求的运动。
跑步更像一种打猎，仔细观察自己的短板，有效击破，快速提升。这和人生以及商业上是一样的：打好基础，快速迭代，懂得试错。
通过体育，尤其是跑步，我学会了团队合作中的奉献，学会了从他人经历中汲取经验。我对体育的热爱使我能够在工作中保持激情，保持谦逊并帮助他人。
没有一个跑步爱好者是无缘无故踏上跑道的，每一个跑者背后的诉求都各不相同：更好的身材，更强健的体魄，更愉悦的心情，更强大的自我。奔跑的过程，也是我们挑战极限，自我蜕变的过程。
有部日本片子，叫《百元之恋》，不是讲跑步的，讲拳击的。女主在影片的最后说："可是…真的好想赢一次啊。"
跑步，让很多人第一次发现自己的身体里有那么多可供挖掘的潜能；第一次让我意识到，我可以跑得比自己想象得更多，更远。
（图：原文此处有配图——I'm running ALONE, because we're in this TOGETHER. London Marathon）
我们生活的大多数时刻比短视频还要快速、重复、肤浅、不可控和无意义，记不住什么东西。只有在进行读书、学习、运动或创造的时候，才会感觉时间被拉长，记忆有被丰富。一部分是因为刻意训练带来的痛苦，一部分是因为有投入和反馈的心流，还有一部分是因为进度可控，而不是被别人的时间进度条拖着往前走。
时间其实是不可被浪费的，但你的身体和记忆会。
如果你有时间去担心你无法控制的事情，你就会分心。内心的平静来自于专注于正确的事情。如果你适度地痴迷于某件事，就算它是断断续续的，只有专注的时间周期足够长，你不断地为解决难题而努力，你会跌跌撞撞地得到一个答案。这是人生的半个秘方。
02 NFL教给我的东西
作为一个体育狂热者，我花了大量时间研究美国体育产业的商业逻辑。从NFL、NBA到NCAA，从媒体转播权到博彩合法化，这个产业给了我关键的商业课：内容永远是王道，但价值在于生态位。
2023年，Google以每年20-25亿美元的价格拿下NFL Sunday Ticket的转播权。这个数字让所有人震惊，这是一笔明显"亏本"的买卖。
但Google不是傻子。它要的不是转播权本身，而是用户增长引擎。
对硅谷科技巨头而言，NFL是无与伦比的工具，用来解决最核心的增长焦虑：
-用户获取：Amazon用《周四橄榄球之夜》驱动Prime会员注册
-生态黏合：在Amazon上看球的观众，会顺手用Alexa查数据，在Prime上购物
-客厅入口：谁掌控体育直播，谁就在未来的客厅战争中占据地形
对华尔街而言，NFL则是估值之锚。Disney的ESPN、Fox Sports、NBC的Peacock，它们的估值很大程度上取决于是否拥有NFL转播权。NFL合约的续签或终止，能让一家媒体公司的股价在一天内波动10%-15%。
这给我的启发是：价值不在于你做什么，而在于你在生态系统中的位置。
如果你仔细观察，有体育背景的人在欧美金融/科技领域有结构性优势，而不仅仅是压力管理、竞争意识、团队协作这种表面解释。但更深层的优势在于：体育背景的人天然理解"生态位>单次交易"。
NFL教会的不是"如何赢一场比赛"，而是"如何成为文化标准"。这种思维在金融里是稀缺的：普通banker：做deal赚fee；体育背景的banker: 做deal是为了成为某个垂直领域的"trusted advisor"
欧美很多公司的真正护城河并不是技术本身，而是来自文化资本的时间积累。
体育训练教的就是这个：肌肉记忆需要10000小时；王朝球队不是靠一个球星，而是几代人传承的system和culture；顶级运动员理解:今天的训练，三年后才见效。
这在金融/tech里是反直觉的：大多数创始人追求"技术突破"、"快速增长"，但真正的护城河是成为标准和积累信任： Visa 70年没出大事故，这种信任不是一个新公司花钱能买的。
体育背景的人理解这个时间维度：早期投入决定长期粘性
1.Winner-takes-all环境的生存法则：NCAA篮球64强锦标赛,只有冠军被记住。这培养的不是"做到前50%就行"，而是"必须top 1"的心态 — 这在VC(只有top 1%的portfolio公司贡献90%回报)、investment banking pitch(只有lead banker拿大头)里是核心竞争力。
2.系统性优势 vs 单次运气：顶级体育项目(NFL/NBA)教的是:dynasty不是靠一场比赛的运气,而是draft strategy + salary cap management + coaching system的持续优化。
3.早期锁定的网络效应：早期粉丝的lifetime value远超后期转化的粉丝。
体育背景的人理解这个逻辑: 青训体系(La Masia培养梅西)的投入,10年后才见回报。但一旦形成文化认同，转换成本极高：巴萨球迷很难变成皇马球迷，就像Visa用户很难切换到新支付系统。
体育IP的马太效应极其明显。前20强体育IP的年平均增长率为42%，在过去十年为13%。但除此之外，500强体育IP中剩下的480个，年平均增长率仅有1%。
更可怕的是，顶级体育IP连培养下一代粉丝都占据优势：它们比其他体育项目高出31%的"14岁前培养"粉丝比例。这些粉丝的参与度也更高，每天观看体育节目的可能性高出92%，在体育节目上的支出比其他粉丝高出88%。早年的亲近感会转化成习惯性的高频互动，这种复合效应使得顶级IP更容易收获高粘性粉丝。
（图：原文此处有配图——洛杉矶湖人17次夺冠海报）
法国社会学家布迪厄有个概念叫"文化资本"——就像金融资本可以积累一样，文化也可以积累。而且文化资本有个特点：需要时间，无法快速复制。
NFL的价值不是比赛或者技术本身，而是文化资本：一代代美国家庭围坐在电视前看超级碗的集体记忆，父亲带儿子去现场看第一场比赛的成人礼，感恩节必看的传统。这些东西，你没法用钱买，只能用时间攒。
AI可以一夜之间生成10万条短视频，但AI没法一夜之间让人们产生"我爷爷、我爸爸、我，三代人都看这个比赛"的情感连接。这种文化资本，才是真正的护城河。
这个逻辑在金融领域同样成立。Visa、Mastercard、PayPal，一旦成为支付标准，网络效应和转换成本会让它们的地位越来越稳固。
visa建立了一个全球商户和银行都信任的清算网络。这种信任，也是一种资本——"信任资本"。这个系统可靠"的集体认知，不是一个新公司花钱就能买来的。你必须真的做几十年，真的没出过问题，人们才会信你。
Stripe成立于2010年，短短12年就成为开发者支付的事实标准。为什么？不是因为它的API写得多漂亮，而是因为它解决了一个千年问题：如何让中小商家以最低成本接入全球支付网络。
这个逻辑开始在我脑海中形成一个雏形：真正的护城河不是技术创新，而是成为某个生态系统中不可或缺的信任节点。技术可以被复制，而且在AI 时代迭代的速度只会越来越快，但信任的建立和积累，都需要经年累月的积累。
这三堂课，让我对"什么样的生意是好生意，什么样的生意值得做"有了更清晰的判断：
1.不要只看交易本身，要看生态位
2.不要只看技术壁垒，要看结构性护城河
3.不要只看增长速度，要看是否能成为标准
03 两个世界的裂缝
2025年，我在湾区和几个国内创业的朋友吃饭。桌上七个人，三个做游戏，两个做电商相关，一个做短视频，一个做社交。没有一个做B2B，没人SaaS，更没有人做Fintech。
我困惑了很久：美国创业，没什么人做电商，游戏，社交/社区，几乎全是SaaS，fintech；国内没有fintech 的创业，saas 也基本上都死的差不多了。后来发现这不是文化问题，而是国情结构问题。
从需求侧来看：经济发展阶段决定了创业方向。
美国的经济属于存量经济优化效率。GDP per capita为$76,000（2023）；企业IT支出占营收的4-6%；核心痛点是人力成本太高（$60K/年），必须用软件替代人。
结果是雇一个财务大概$60K/年，买一个财务软件大概$5K/年，企业必然选择买软件，这就是为什么QuickBooks、Salesforce、Workday这些公司能存在：软件ROI极其明显。
而中国目前还属于增量经济抢占市场。GDP per capita是$12,700（2023）；企业IT支出占营收为 0.5-1%，核心痛点是市场增长太快，必须快速获客和规模化。雇一个财务大概¥10万/年，买一个财务软件花费¥10万/年，企业当然选择雇人（更灵活，能处理各种例外情况）
但电商/直播/社交不一样：这些是直接产生GMV/revenue的工具，不是成本优化工具，企业愿意为"增长"付钱，不愿意为"效率"付钱。
供给侧来看：线下基础设施的成熟度不同。
美国的国情是线下太强，电商是补充。Walmart 1962年成立，是零售基础设施，1970年代就全国覆盖了，Costco、Target、Home Depot全国连锁网络极其成熟，电商渗透率只有18%左右（中国是47%）
为什么线下这么强？美国先城镇化（1950年代郊区化完成），后电商化（2000年代），相隔了近半个世纪，线下连锁在互联网出现前就建立了全国网络，而且美国配送成本高，地广人稀，最后一公里成本是中国的3-5倍。结果就是Amazon只是"补充"，不是"主战场"，电商创业很难，因为你要和Walmart这种巨头竞争，它们的供应链效率极高，除非你做垂直品类（Warby Parker眼镜、Casper床垫），否则很难存活。
中国线下弱，电商反而是主战场：城镇化和电商化同步，1990年代城镇化刚起步，2000年代淘宝/京东出现，电商和城镇化是并行的，电商直接成为主流消费渠道。
同时线下连锁没有建立，中国没有像Walmart这种全国性的零售巨头（大润发、家乐福都是外资，且覆盖有限），小商贩、夫妻店占主导，效率低；最后一公里成本也很低，人口密度高，配送成本是美国的1/3
从资本结构，退出渠道，市场成熟度来看，也是很重要的影响因素。
美国VC的LP构成：大学捐赠基金（Harvard、Yale）、养老金（CalPERS）、家族办公室。这些钱的期限是20年、30年，甚至更久。它们可以承受10年不回报的项目。
中国VC的LP构成：政府引导基金、上市公司、高净值个人。这些钱的期限是3年、5年，还带对赌和回购条款。它们需要快速退出。
结果：美国VC可以投Stripe（2010年成立，2021年才盈利），中国VC只能投拼多多（2015年成立，2018年上市）。
美国：IPO、战略并购、PE接盘，三条路都走得通。一个B2B SaaS公司，ARR到5000万美元，就会有Oracle、Salesforce来谈收购。收不成，还有Vista、Thoma Bravo这样的PE专门买SaaS公司。
中国：IPO是唯一出路。而IPO的门槛是"最近三年累计净利润超过3000万"。一个SaaS公司要做到这个利润，得多少年?
结果就是美国的创业者可以专注做产品，中国的创业者必须三年内跑出利润。
美国企业的IT预算占营收的比例平均是 4%-6%，中国是0.5%-1%。这不是因为中国企业穷，而是因为中国企业的"人力成本/软件成本"比值太低：雇一个财务10万/年，买一个软件也是10万/年，企业当然选择雇人。而在美国，雇一个财务6万美元/年，买一个软件5000美元/年，企业当然选择买软件。
这是三个结构性差异。
但还有一个更深层的差异，是韦伯在《新教伦理与资本主义精神》中提到的：资本主义的成功不仅是制度问题，更是文化和伦理问题。
韦伯说的"新教伦理"核心是什么？延迟满足、长期主义、容忍失败。你做生意不是为了今年赚快钱，而是为了建立一个能传承的事业。失败不是耻辱，而是上帝对你的考验。
美国VC生态的成功，深处就是这种"新教伦理"：
- LP愿意给GP 10年、20年时间，因为他们相信长期价值
- 投资人愿意给失败的创始人第二次、第三次机会
- 创业者可以说"我要花10年建立一个标准"，而不是"我要3年上市套现"
而中国的创业文化更接近"商业功利主义"：三年必须盈利，五年必须上市，失败就是能力不行，下次没人给你钱。这种文化差异，比制度差异更难跨越。
（图：原文此处有配图——American Express 与 Revolut 卡面）
在研究美国创业生态的过程中，我开始系统性地对比不同行业的特性。这个过程让我对"什么样的行业值得长期投入"有了更清晰的认知。
我把主要行业分成三类：
第一类：技术驱动，α快速压缩：代表：AI、SaaS（纯工具类）、游戏
这类行业的特点是：
- 进入门槛低：云服务、开源框架让技术民主化
- 竞争极度激烈：任何创新三个月就被复制
- 同质化严重：大家都在用GPT-4，差异只在prompt
AI是最典型的例子。真正的超额利润在模型层、算力层、芯片层、底层infra，这些都是工程主导。但应用层的α很可能被压缩，进入门槛会越来越低，但是竞争会越来越激烈，同质化异常严重，因为：
- OpenAI今天发布GPT-5，明天Anthropic就发布Claude 4，后天Google就发布Gemini 2.0
- 所有应用都在用同样的模型，差异只在工程实现
- 护城河极其脆弱：技术进步会不断压缩商业优势
我不是说AI创业没有机会，而是说这个行业的护城河来自持续的技术领先，对于非技术科班出身的人来说，一旦停止创新，几乎立刻被超越。
第二类：监管驱动，α被压缩但稳定：代表：医疗、教育、传统金融。
这类行业的特点是：
- 进入门槛极高：牌照、合规、政策
- 竞争强度中等：护城河是牌照，不是技术
- 利润率被监管限制：医保控费、学费限价、利率管制
传统金融是最典型的例子。银行的ROE长期稳定在8%-12%，不会更高，也不会更低。为什么？因为监管会限制你的风险敞口，也会保护你的市场地位。这类行业适合求稳，不适合追求超额回报。
第三类：信任驱动，α巨大且持续：代表：Fintech基础设施、体育IP、奢侈品
这类行业的特点是：
- 进入门槛极高：不是技术门槛，而是信任门槛
- 竞争强度低：一旦建立标准，极难被替代
- 网络效应强：用户越多越强
- 时间是朋友：越久越值钱
金融的底层逻辑千年不变：信用、清算、资产配置、利率传导、风险管理。这是一套从古代钱庄到现代Stripe都在解决的问题。关键是：这些问题的解决方案，不是技术，而是转换成本和信任。
一旦成为标准，α巨大且持续几十年。而且关键是：Fintech需要的能力：结构理解、法律理解、监管沟通、商业模式设计、信任构建——这些恰恰是商科/经济背景的优势，而不是纯技术背景的优势。
鉴于这个，让我更坚定了做Fintech 而不是跟风做AI的决心。
04 新瓶装旧酒：千年不变的逻辑
所有的风口，最终都会吹回到那些长期坚持的人身上。找一个东西做到底，反复做。
体育的第一个"旧酒"是：人类永恒的情感需求。
如果说AI正在"重塑一切"，那么体育是少数几个不能被直接取代的行业之一。体育满足的不是"信息需求"或"效率需求"，而是人类最原始的情感需求：竞争、英雄、归属感。
麦克卢汉有句名言："媒介即信息"（The Medium is the Message）。他的意思是，真正改变社会的不是内容本身，而是传播内容的方式。
从收音机到电视，从电视到流媒体，体育内容的传播方式一直在变。按照麦克卢汉的理论，每一次媒介革命都应该改变体育本身。
但有趣的是，在体育领域，媒介可以变，但"信息"（为什么需要体育）千年不变。
1950年代，人们围在收音机前听世界杯；1980年代，人们围在电视前看NBA；2020年代，人们用手机看TikTok上的集锦。媒介变了三次，但人们为什么看体育这个问题的答案，一个字都没变：我们需要看到人类极限被突破的瞬间，我们需要英雄，我们需要和别人一起为同一支球队欢呼。
AI正在改变"媒介"：如何制作集锦、如何推荐内容、如何个性化观赛。但AI改变不了"信息"：人类需要体育的那个底层原因。这就是麦克卢汉理论在体育领域的反例：有些"信息"是不会被"媒介"改变的。
AI可以生成文章、写代码、设计海报，但AI永远无法复制2010年NBA总决赛第7场的最后20秒。那个瞬间的心跳、呼吸、观众的尖叫、球员的眼神——这些都是不可编程的。
第二个旧酒是：竞争的偶然性：比赛永远不可预测。体育的魅力在于不确定性。
你可以用AI预测比赛结果，可以分析球员数据，可以模拟战术效果。但你永远无法预测：
- 库里最后一投会不会进
- 梅西会不会在第90分钟绝杀
- 费德勒会不会在赛点救回来
这种不确定性，恰恰是体育最大的价值。如果比赛结果可以被AI 100%预测，那比赛也就失去了意义，博彩公司就会成为100% 的公司。
德国哲学家本雅明有个概念叫"光韵"（aura），那种只有在"此时此地"才能感受到的独特气息，机械复制永远无法捕捉。
AI可以复制库里的投篮动作，可以生成完美的比赛集锦，可以模拟现场的声音和画面。但AI复制不了"光韵"：那种你坐在现在，身在球馆里，看到库里在你眼前投进绝杀，全场爆发的那一刻，只属于那个时空的震撼。
我有个朋友花了8000美元去现场看超级碗，有人问他："你不会是疯了吧？在家看直播不是一样吗？画面还更清楚。
他说："完全不一样。直播可以看到球的轨迹，但你感受不到全场几万人同时屏住呼吸的那种压迫感，感受不到球进那一刻，地板的震动传到你脚底的那种冲击。那个瞬间的'光韵'，是花多少钱都买不回来的。这就是为什么体育门票越来越贵，但现场观众越来越多。人们要的不是"看到比赛"，而是体验那个无法复制的"此时此地"。
AI可以改变内容生产的方式（自动生成集锦、实时数据分析、个性化推荐），但AI无法改变人们为什么需要体育这个问题的答案。
答案永远是：我们需要看到人类极限被突破的瞬间，我们需要在不确定的比赛中找到确定的情感寄托。而这个"瞬间"，必须是真实的、不可预测的、带着"光韵"的。
（图：原文此处有配图——运动员与金钱拼贴图）
第三个是英雄的崇拜。
有个有趣的现象：AI时代，科技公司给顶级AI科学家的签约奖金是1亿美元，而顶级运动员的合同也是1亿美元起步。
为什么运动员的价值没有被AI稀释？因为顶尖运动员除了稀缺之外，价值不是"完成某个任务"，而是"成为英雄"。这是AI永远做不到的。
你可以让AI写出完美的投篮动作分析，但你无法让AI替你崇拜一个英雄。你可以让AI生成科比的比赛集锦，但你无法让AI替你感受看到科比绝杀时的那种激动。
人类需要英雄，这是千年不变的需求。从古代的角斗士到现代的NBA球星，英雄的形式在变，但需求没变。
第四个是归属感：2024年超级碗，1.23亿美国人观看直播，创下历史新高。这不是一场比赛，而是一个集体仪式。全家人围在电视前，朋友聚会，酒吧爆满，整个国家在这一天都在谈论同一件事。
斯坦福社会学家格兰诺维特有个著名理论："弱连接的力量"。他发现，陌生人之间的弱连接，在某些场景下比亲密关系的强连接更有价值。
找工作靠的不是你最好的朋友，而是你认识但不太熟的人——因为他们在不同的圈子里，能给你不同的信息和机会。体育赛事就是"弱连接"发挥最大作用的场景。
超级碗那天，你和全国1.23亿陌生人同时看同一场比赛。你们不认识，可能永远不会见面，但在那一刻，你们有共同的话题、共同的情绪、共同的记忆。第二天你上班，遇到电梯里的陌生同事，可以聊"昨晚那个绝杀"。你去咖啡店，和barista可以聊"那个四分卫太牛了"。你打开社交媒体，全世界都在讨论同一件事。
AI可以让每个人看到不同的内容——根据你的喜好推荐视频，根据你的算法定制信息流。但这种个性化有个代价：你的推荐算法，和别人的完全不同。你活在一个信息茧房里，和外界的"弱连接"越来越少。
而"弱连接"恰恰是社会凝聚力的来源。超级碗让1.23亿美国人有了共同话题，这就是一次大规模的"弱连接"建立。这种连接，AI的个性化算法做不到，也不想做——因为个性化的目标是让你只看你喜欢的，而不是让你和别人有共同语言。
AI可以提供个性化推荐，可以让每个人看到不同的内容。但AI无法提供这种"我和1亿人在看同一场比赛"的集体归属感。
体育的"旧酒"就是这几个不可编程的情感需求。AI可以改变"瓶子"（直播技术、数据分析、内容分发），但改变不了"酒"（情感需求）。
05 Fintech 的旧酒：千年金融逻辑
Fintech的"旧酒"是什么?
是金融的底层逻辑：信用、清算、资产配置、利率传导、风险管理。这些逻辑是千年不变的。无论是古代的钱庄、票号，还是现代的Stripe、Plaid，本质都是在解决同一个问题：如何让价值在不同主体之间安全、高效地转移。
（图：原文此处有配图——电影台词截图："我来美联储，只办三件事：降息，缩表，让美联储再次伟大"）
从钱庄到Stripe：本质问题一样，只是"瓶"不同
明清时期的山西票号，解决的是什么问题？一个山西商人在北京做生意赚了钱，要把钱运回山西老家。但带着白银走镖局，路上会被劫匪抢。怎么办？票号的解决方案：你在北京的票号分号存银子，拿一张票据。到了山西，拿票据去总号取银子。这就是最早的"清算网络"。
现代的Stripe，解决的是什么问题？一个开发者在自己的网站上卖软件，要接入全球支付网络。但对接Visa、Mastercard、各国银行，技术门槛太高。怎么办？
Stripe的解决方案：你用7行代码接入Stripe的API，Stripe帮你对接全球支付网络。这还是"清算网络"，只是技术手段不同。本质问题一样：如何安全、高效地转移价值。只是"瓶"从票据变成了API。
（图：原文此处有配图——US Financial Regulators 监管机构图谱）
为什么AI颠覆不了金融?
很多人说AI会颠覆金融。我不这么认为。AI可以改变金融的"瓶"（风控模型、客服机器人、智能投顾），但AI改变不了金融的"酒"（信用、清算、风险管理）。
第一，AI可以做风控模型，但AI承担不了"放错款"的责任。
假设你是一个品牌CFO，要决定给一个创作者放款20万美元。AI告诉你"这个创作者的违约概率是2%"。你敢放吗?
如果创作者真的违约了，AI会赔你100万吗？不会。AI只是一个工具，不是一个责任主体。所以金融的核心不是预测，而是承担风险。AI可以帮你预测，但不能替你承担。
第二，AI可以写合规文档，但AI没法代替人去SEC开听证会。
金融是强监管行业。商业模式是否合规，是SEC、FINRA、各州金融监管局说了算。
AI可以帮你生成一份完美的合规文档，当监管机构质疑你的模型逻辑时，你需要一个真人坐在听证会上解释："为什么你的风控模型是公平的、可解释的、不歧视的。这个解释权，AI无法给于。
第三，AI可以提高效率，但替代不了信任。
Visa成立于1958年，花了60 年能成为全球清算标准，全球商户和银行都相信Visa的清算，这种信任，是60年积累的结果。60年间，Visa处理了数万亿笔交易，几乎没有出过大的安全事故。这种track record，是AI无法替代的。
PayPal成立于1998年，26年了，依然是C端支付标准。Stripe成立于2010年，14年了，已成为开发者支付标准。
信任不是算法，而是时间。
AI应用的护城河是"持续的技术领先"，一旦停止创新，立刻被超越。金融变革的特点是：慢、重监管、网络效应强，一旦成功极难被替代。Finetch一旦成为标准，α巨大且持续几十年。
Fintech的"旧酒"就是这套千年金融逻辑：信用、清算、风险管理。AI可以改变实现方式，但改变不了底层需求。
06 给同路人的地图
如果你对体育产业感兴趣，我之前写过：所有生意都可以AI重塑一遍，唯独体育不行，这里有四条我看好的创业方向。
在非硅谷的复杂环境中，创业者必须做全栈并解决更脏、更重的现实问题。未来5-15年，简单的API已经被建构完毕，机会将向深度垂直和复杂业务流转移，关于Fintech，我自己理解有三个增量方向：
- B2B科技与CFO Stack
机会点：C端支付已经极其丝滑，但B端企业间的资金流转依然像上个世纪。跨国对账、发票自动化、应付/应收账款（AP/AR）的智能对冲，是巨大的蓝海。结合AI，未来的企业账本不仅是记录，而是具备预测和自动决策能力的财务大脑。
代表公司：
- Ramp、Brex：企业费用管理与发卡
- Bill.com：应收/应付账款自动化
- Deel：全球合规发薪
- 嵌入式金融与垂直场景
机会点：金融将不再是一个独立的行业，而是所有垂直SaaS的变现手段。针对特定行业（如医疗、物流、电竞、创作者经济）量身定制的资金流转和信贷方案。谁掌握了特定行业的垂直工作流数据，谁就能给出比传统银行更精准的风控定价。
作为创业者，不做通用工具，而是做针对特定场景的深度解决方案。创作者营销、游戏出海、跨境电商，每个场景都有独特的资金流转逻辑。
- 跨境与全球化合规网络
机会点：随着远程办公和全球化供应链的不可逆转，能够无缝处理多币种、多主体合规、税务和薪酬发放的基础设施（如Deel的进阶版）依然有极大的结构性红利。Deel解决了雇佣问题，但还有大量的"非雇佣关系"的跨境支付场景。
结语
马拉松终点线前的最后 2 公里-100米的距离，很多人双腿已经不是自己的了，意识也开始模糊，但大家还有一个信念：还有 2 公里，还有100米。这个时候，身体在和意志对抗。你可以选择慢下来，走过终点线。但你不会，因为这42公里的意义，就在于这最后100米的坚持。
AI会改变很多东西，但有些东西永远不会变：
- 人类对竞争和英雄的渴望
- 人类对信任和确定性的需求
- 人类对长期价值的追求
体育和Fintech，就是我在AI浪潮中的两个锚点。它们让我知道，无论技术如何变化，有些东西永远值得相信和坚守。
（图：原文此处有配图——"在所有的选择关头，多和人连接"）
人越走到后面，越看精力和体能，二者能让你保持长久的专注和对一件事死磕出结果的心力。你可以白嫖到别人的想法，但无法白嫖到别人想法的思考路径，这是自己要做的功课，没有捷径，也抄不了答案。
AI会改变很多东西，但AI改变不了那种身体和意志的对抗。AI可以告诉你最优配速，可以预测你的撞墙点，可以生成完美的恢复方案。但AI无法替你跑完最后那100米。这就是体育的价值：不可替代的人类体验。
金融也是一样。AI可以做风控模型，可以优化支付流程，可以自动化对账。但AI无法替你建立信任，无法替你承担责任，无法替你成为那个"Source of Truth"。这就是Fintech的价值：不可替代的信任节点。
如果你读到这里，我想你可能和我有一些共同点。
我在寻找这样的人：
第一：相信长期主义——之前汪滔说：很多人混淆这些东西：你是真心看好一个方向，愿意深耕十年;还是风口来了，想发财、创业，顺便用公司的资源给自己铺路。就像选秀节目，谁会直接说"我想出名"?说出来就low了嘛，说的都是"想把快乐带给大家"。
第二：不盲目跟风，有自己的判断。AI很火，但AI的α会被极度压缩。体育很传统，但体育的护城河是文化。Fintech很难，但难才意味着护城河。
第三：热爱你做的事，不单单是为了钱。我跑马拉松不是为了奖金，而是为了证明我可以。
第四：相信"新瓶装旧酒"，相信底层逻辑不会变。技术会变，模式会变，但人类的情感需求、信任需求、价值转移需求，千年不变。找到那个"旧酒"，然后等着时代给你一个新瓶。反复做，等风吹回来。
如果你是这样的人，我很想认识你。
不管你是在做体育创业、Fintech创业，还是在任何一个你认为不会被AI颠覆的领域深耕，我都很想听听你的故事。
因为在这个AI焦虑的时代，我们需要更多相信底层逻辑的人，需要更多愿意长期坚持的人，需要更多知道"什么东西不会变"的人。
Introduction
The AI industry is highly technology-concentrated.
The real excess profits sit in the model layer, the compute layer, the chip layer — the underlying infrastructure, all of it engineering-led.
The alpha will very likely get compressed.
Barriers to entry keep falling, competition is fierce, and the sameness is severe.
The risk in AI is this: technical progress compresses your commercial moat.
Looking back
The London Marathon last year, kilometer 40.
Legs like lead, breath heaving like a bellows, every step a fight against my own body. And in that very moment, a question suddenly popped into my head: everyone is anxious about what AI will replace — but has anyone stopped to ask what AI can never do?
AI can generate a flawless marathon training plan. It can optimize your pacing strategy from your fitness data. It can even simulate the physiological response after you hit the wall. There are now companies building VR marathons — you put on a headset, the treadmill mimics the rise and fall of the Boston course, and a virtual crowd cheers you on from the screen.
The French philosopher Baudrillard called this the "simulacrum" — the virtual growing ever more lifelike, until it becomes even "more real" than the real.
But I later understood something: no matter how convincing the simulacrum, it can't fool your body. Your muscles know it isn't real. Your will knows it isn't real. The real marathon is your thighs tearing, is the moment you want to quit but grit your teeth and hold on, is that instant you cross the line and collapse to the ground unable to move — that real pain and release. This bodily reality is AI's final frontier.
In that moment I understood something: what's true of sport is true of finance.
I'm someone who runs a marathon every year, and someone building a fintech startup. Across countless moments and fragments over the years, I came to realize something: they don't seem to be two things — they're two sides of the same thing. The underlying logic is identical: technology changes, but humanity's deepest needs don't change in a thousand years.
This essay is some of what I've worked out over the past three years — on the track, on the court, on the road of building a company.
Here's the main text. Enjoy.
01 A business lesson on the running track
I honestly can't remember when I first fell in love with sport.
But when I later looked back, I found one of the reasons: as a kid I was actually a very unconfident person. The only time I felt confident was on the playing field and the track. There, I felt the whole world was mine.
Over the years, the most subtle thing competitive sport instilled in me is this: give everything to whatever you do, and never concede or give up until the very last second decides the outcome. In a sense, that became the baseline color of my character.
My dream was always to be an athlete. Anyone who knows me knows I'm a rabid sports fan: basketball, running, tennis, badminton, table tennis. I watch and play nearly every mainstream sport.
Though I never became a real athlete, sport truly gave me confidence. And it taught me: people with real nerve don't just have nerve — even their goals are different from everyone else's, because they want a bigger return in the same amount of time, and that is what makes them decisive in action.
Don't underestimate grinding it out. It's a crucial component of mental strength — whether you're leading a team to a win or getting a deal closed, the core is grinding it out. Learn a pile of tactics and in the end none of it beats sheer persistence, because on the battlefield, at the card table, nobody is a fool. The fiercest psychological weapon is your intense will to win — the drive to see the thing done, and done well.
In recent years, my time and energy have gradually shifted from ball sports to track and field. Running is the sport that tests raw talent most of all, and it's also the one that most easily turns you inward.
Running is more like hunting: observe your weaknesses carefully, break through them efficiently, improve fast. It's the same in life and in business: build the foundation, iterate quickly, know how to fail smart.
Through sport, and running especially, I learned the sacrifice in teamwork, and learned to draw lessons from other people's experiences. My love of sport lets me stay passionate at work, stay humble, and help others.
No runner sets foot on the track for no reason; behind every runner is a different want: a better physique, a stronger body, a happier mood, a more powerful self. The act of running is also our process of testing our limits and remaking ourselves.
There's a Japanese film called 100 Yen Love — it's not about running, it's about boxing. At the end, the heroine says: "But… I really just wanted to win, once."
Running lets many people discover, for the first time, how much untapped potential is inside their own bodies; it made me realize, for the first time, that I could run more, and farther, than I ever imagined.
(Figure in original.)
Most of the moments in our lives are faster, more repetitive, shallower, more uncontrollable, and more meaningless than a short video — nothing sticks. Only when we're reading, learning, exercising, or creating does time feel stretched and memory enriched. Part of that is the pain deliberate training brings; part is the flow of investment and feedback; and part is that the pace is under your control, rather than being dragged forward by someone else's progress bar.
Time actually can't be wasted — but your body and your memory can.
If you have time to worry about things you can't control, you'll be distracted. Inner peace comes from focusing on the right things. If you're moderately obsessed with something — even in fits and starts — as long as your stretch of focus is long enough, and you keep working at solving the hard problem, you'll stumble your way into an answer. That's half the recipe for life.
02 What the NFL taught me
As a sports fanatic, I've spent a lot of time studying the business logic of the American sports industry. From the NFL and NBA to the NCAA, from media broadcast rights to the legalization of betting, this industry gave me a key business lesson: content is always king — but the value is in the ecological niche.
In 2023, Google took NFL Sunday Ticket's broadcast rights for $2–2.5 billion a year. The number stunned everyone; it was an obviously "money-losing" deal.
But Google isn't stupid. What it wanted wasn't the broadcast rights themselves — it was a user-growth engine.
For the Silicon Valley giants, the NFL is an unrivaled tool for solving their most fundamental growth anxieties:
- User acquisition: Amazon uses Thursday Night Football to drive Prime sign-ups.
- Ecosystem glue: Someone watching the game on Amazon will casually use Alexa to check stats and shop on Prime.
- The living-room gateway: whoever controls live sports holds the high ground in the coming war for the living room.
For Wall Street, the NFL is the anchor of valuation. The valuations of Disney's ESPN, Fox Sports, and NBC's Peacock hinge to a large degree on whether they hold NFL rights. The renewal or loss of an NFL contract can swing a media company's stock 10%–15% in a single day.
The lesson for me was this: value isn't in what you do — it's in your position within the ecosystem.
If you look closely, people with a sports background have a structural advantage in Western finance and tech, and it goes deeper than the surface explanations of stress management, competitive drive, and teamwork. The deeper edge is this: people with a sports background instinctively understand that "niche beats single transaction."
The NFL doesn't teach you "how to win a game" — it teaches you "how to become the cultural standard." That mindset is scarce in finance. An ordinary banker does deals to earn fees; a banker with a sports background does deals in order to become the "trusted advisor" of some vertical.
For many Western companies, the true moat isn't the technology itself — it's the accumulation over time of cultural capital.
That's exactly what athletic training teaches: muscle memory takes 10,000 hours; a dynasty isn't built on a single star but on a system and culture handed down over generations; the elite athlete understands that today's training only pays off three years from now.
In finance and tech this is counterintuitive: most founders chase "technical breakthroughs" and "rapid growth," but the real moat is becoming the standard and accumulating trust. Visa has had no major failure in 70 years — that kind of trust is not something a new company can buy with money.
People with a sports background understand this dimension of time: early investment determines long-term stickiness.
1. The survival rules of a winner-takes-all environment: In the NCAA basketball tournament's field of 64, only the champion is remembered. What this breeds isn't a "top 50% is fine" attitude but a "must be number one" mentality — and that is a core competency in VC (where only the top 1% of portfolio companies contribute 90% of returns) and in investment banking pitches (where only the lead banker takes the lion's share).
2. Systematic advantage vs. one-off luck: Elite sports (NFL/NBA) teach that a dynasty isn't the luck of one game but the continuous optimization of draft strategy plus salary-cap management plus coaching system.
3. The network effects of locking in early: the lifetime value of an early fan far exceeds that of one converted later.
People with a sports background get this logic: the investment in a youth academy (La Masia produced Messi) pays off only ten years later. But once cultural identity forms, switching costs become enormous: a Barça fan rarely becomes a Real Madrid fan, just as a Visa user rarely switches to a new payment system.
The Matthew effect in sports IP is extreme. The top 20 sports IPs grew 42% a year on average, versus 13% over the past decade. But the remaining 480 of the top-500 sports IPs grew only 1% a year on average.
More striking still, the elite sports IPs even dominate at cultivating the next generation of fans: they have 31% more fans "cultivated before age 14" than other sports. These fans are also more engaged, 92% more likely to watch sports programming daily, and they spend 88% more on sports content than other fans. That early affinity converts into habitual, high-frequency engagement, and this compounding effect makes it easier for top IPs to win high-stickiness fans.
(Figure in original.)
The French sociologist Bourdieu had a concept called "cultural capital" — just as financial capital can accumulate, so can culture. And cultural capital has a distinctive trait: it takes time and can't be quickly replicated.
The NFL's value isn't the games or the technology — it's cultural capital: the collective memory of generations of American families gathered around the TV for the Super Bowl; the rite of passage of a father taking his son to his first game; the Thanksgiving tradition of a must-watch game. These things you can't buy with money — you can only build them with time.
AI can generate 100,000 short videos overnight, but AI can't produce overnight the emotional bond of "my grandfather, my father, and I — three generations all watch this game." That cultural capital is the real moat.
The same logic holds in finance. Once Visa, Mastercard, and PayPal become the payment standard, network effects and switching costs make their positions ever more entrenched.
Visa built a clearing network that merchants and banks worldwide trust. That trust is also a form of capital — "trust capital." The collective belief that "this system is reliable" is not something a new company can buy with money. You have to actually run it for decades, actually never have a problem, before people will trust you.
Stripe was founded in 2010 and in just twelve years became the de facto standard for developer payments. Why? Not because its API is beautifully written, but because it solved a thousand-year-old problem: how to let small and medium merchants plug into the global payment network at the lowest cost.
This logic began to take shape in my mind: the real moat isn't technical innovation — it's becoming the indispensable node of trust in some ecosystem. Technology can be copied, and in the AI era it iterates faster and faster, but building and accumulating trust takes years upon years.
These three lessons gave me a clearer judgment about "what makes a good business, and what business is worth doing":
1. Don't just look at the transaction — look at the niche.
2. Don't just look at the technical barrier — look at the structural moat.
3. Don't just look at the speed of growth — look at whether you can become the standard.
03 The rift between two worlds
In 2025 I had dinner in the Bay Area with a few friends who are founders back in China. Seven of us at the table: three in gaming, two in e-commerce, one in short video, one in social. Not one in B2B, no one in SaaS, and certainly no one in fintech.
I was puzzled for a long time: in American startups, hardly anyone does e-commerce, gaming, or social/community — it's almost all SaaS and fintech; in China there are no fintech startups, and SaaS has more or less all died off. I later realized this isn't a cultural problem but a problem of structural national circumstance.
On the demand side: the stage of economic development determines the direction of entrepreneurship.
The American economy is a stock economy optimizing for efficiency. GDP per capita is $76,000 (2023); corporate IT spending is 4%–6% of revenue; the core pain point is that labor is too expensive ($60K a year), so you must use software to replace people.
The result: hiring a finance person costs about $60K a year, buying finance software about $5K a year, so a company will inevitably choose to buy the software. That's why QuickBooks, Salesforce, and Workday can exist: the software ROI is glaringly obvious.
China, meanwhile, is still an incremental economy racing to grab market share. GDP per capita is $12,700 (2023); corporate IT spending is 0.5%–1% of revenue; the core pain point is that the market grows so fast you have to acquire customers and scale quickly. Hiring a finance person costs about ¥100K a year, and buying finance software also costs about ¥100K a year — so of course the company hires the person (more flexible, and able to handle all the edge cases).
But e-commerce, livestreaming, and social are different: these are tools that directly generate GMV and revenue, not cost-optimization tools. Companies are willing to pay for "growth," not for "efficiency."
On the supply side: the maturity of offline infrastructure differs.
America's condition is that offline is too strong, and e-commerce is a supplement. Walmart was founded in 1962 as retail infrastructure and had national coverage by the 1970s. Costco, Target, and Home Depot have extremely mature national chain networks. E-commerce penetration is only about 18% (China's is 47%).
Why is offline so strong? America urbanized first (suburbanization completed in the 1950s) and went e-commerce later (the 2000s) — nearly half a century apart. Offline chains built national networks before the internet appeared, and America's delivery costs are high: sparsely populated over a vast land, with last-mile costs 3–5 times China's. The result is that Amazon is merely a "supplement," not the "main battlefield." E-commerce startups are hard, because you have to compete with a giant like Walmart, whose supply-chain efficiency is extremely high. Unless you do a vertical category (Warby Parker glasses, Casper mattresses), survival is difficult.
In China, offline is weak, and e-commerce is the main battlefield instead. Urbanization and e-commerce moved in step: urbanization was just getting going in the 1990s, Taobao and JD appeared in the 2000s, and e-commerce and urbanization ran in parallel, so e-commerce directly became the mainstream consumer channel.
At the same time, offline chains never got established. China has no national retail giant like Walmart (RT-Mart and Carrefour are foreign-owned and have limited coverage); small vendors and mom-and-pop shops dominate, and are inefficient. Last-mile costs are also very low: high population density means delivery costs are a third of America's.
Capital structure, exit channels, and market maturity are also very important factors.
The LP composition of American VC: university endowments (Harvard, Yale), pensions (CalPERS), family offices. This money has horizons of 20 years, 30 years, or longer. It can tolerate a project that returns nothing for 10 years.
The LP composition of Chinese VC: government guidance funds, listed companies, high-net-worth individuals. This money has horizons of 3 to 5 years, with performance-guarantee and buyback clauses attached. It needs to exit quickly.
The result: American VC can fund Stripe (founded 2010, not profitable until 2021); Chinese VC can only fund Pinduoduo (founded 2015, public by 2018).
America: IPO, strategic M&A, and PE takeover are all viable paths. A B2B SaaS company whose ARR reaches $50 million will get acquisition talks from Oracle and Salesforce. If that falls through, there's still Vista and Thoma Bravo — PE firms that specialize in buying SaaS companies.
China: an IPO is the only way out. And the IPO bar is "cumulative net profit over the past three years exceeding ¥30 million." How many years does it take a SaaS company to hit that profit?
The result: American founders can focus on building the product, while Chinese founders must turn a profit within three years.
American companies' IT budgets average 4%–6% of revenue; China's are 0.5%–1%. It's not that Chinese companies are poor — it's that their ratio of "labor cost to software cost" is too low: hiring a finance person is ¥100K a year and buying software is also ¥100K a year, so of course the company hires the person. In America, hiring a finance person is $60K a year and buying software is $5K a year, so of course the company buys the software.
Those are three structural differences.
But there's an even deeper difference, the one Weber raised in The Protestant Ethic and the Spirit of Capitalism: the success of capitalism is not only a matter of institutions but of culture and ethics.
What is the core of Weber's "Protestant ethic"? Delayed gratification, long-termism, tolerance of failure. You do business not to make a quick buck this year but to build an enterprise that can be handed down. Failure isn't a disgrace but a test from God.
The success of the American VC ecosystem is, at its depths, this "Protestant ethic":
- LPs are willing to give GPs 10 or 20 years, because they believe in long-term value.
- Investors are willing to give failed founders a second and a third chance.
- Founders can say "I want to spend 10 years building a standard" rather than "I want to go public in three years and cash out."
Chinese startup culture, by contrast, is closer to "commercial utilitarianism": you must be profitable in three years, public in five; failure means you weren't capable, and next time no one will give you money. This cultural difference is harder to cross than the institutional one.
(Figure in original.)
In the course of studying the American startup ecosystem, I began systematically comparing the traits of different industries. The process gave me a clearer sense of "what kind of industry is worth investing in for the long term."
I sort the major industries into three types:
Type one — technology-driven, alpha compressed fast. Examples: AI, SaaS (pure tools), gaming.
The traits of these industries:
- Low barriers to entry: cloud services and open-source frameworks democratize the technology.
- Extremely fierce competition: any innovation is copied within three months.
- Severe sameness: everyone uses GPT-4; the only difference is in the prompt.
AI is the most typical example. The real excess profits are in the model layer, compute layer, chip layer, and underlying infra — all engineering-led. But the alpha in the application layer will very likely be compressed; barriers to entry keep falling, competition grows fiercer, and the sameness is severe, because:
- OpenAI ships GPT-5 today, Anthropic ships Claude 4 tomorrow, Google ships Gemini 2.0 the day after.
- All applications use the same models; the difference is only in engineering implementation.
- The moat is extremely fragile: technical progress continually compresses the commercial advantage.
I'm not saying there's no opportunity in AI startups — I'm saying this industry's moat comes from sustained technical leadership, and for those who aren't formally trained engineers, the moment you stop innovating you're overtaken almost instantly.
Type two — regulation-driven, alpha compressed but stable. Examples: healthcare, education, traditional finance.
The traits of these industries:
- Extremely high barriers to entry: licenses, compliance, policy.
- Moderate competitive intensity: the moat is the license, not the technology.
- Margins capped by regulation: insurance cost-control, tuition caps, interest-rate controls.
Traditional finance is the most typical example. A bank's ROE stays stable at 8%–12% over the long run — no higher, no lower. Why? Because regulation limits your risk exposure and also protects your market position. These industries suit those seeking stability, not those chasing excess returns.
Type three — trust-driven, alpha huge and enduring. Examples: fintech infrastructure, sports IP, luxury goods.
The traits of these industries:
- Extremely high barriers to entry: not a technical barrier but a barrier of trust.
- Low competitive intensity: once the standard is set, it's extremely hard to displace.
- Strong network effects: the more users, the stronger you get.
- Time is a friend: the older, the more valuable.
The underlying logic of finance doesn't change in a thousand years: credit, clearing, asset allocation, interest-rate transmission, risk management. It's a set of problems that everything from the ancient money house to the modern Stripe has been solving. The key is: the solutions to these problems aren't technology — they're switching costs and trust.
Once you become the standard, the alpha is huge and endures for decades. And crucially, the capabilities fintech demands — structural understanding, legal understanding, regulatory communication, business-model design, trust-building — are exactly the strengths of a business/economics background, not a pure-tech one.
All of this made me more resolved to do fintech rather than follow the crowd into AI.
04 New bottle, old wine: the logic that doesn't change in a thousand years
Every hot trend eventually blows back to the people who persisted over the long haul. Find one thing and do it to the end — do it over and over.
The first "old wine" of sport is: humanity's eternal emotional needs.
If AI is "remaking everything," then sport is one of the few industries it can't directly replace. Sport satisfies not an "information need" or an "efficiency need" but humanity's most primal emotional needs: competition, heroes, belonging.
McLuhan had a famous line: "The medium is the message." His point was that what truly changes society isn't the content itself but the way the content is transmitted.
From radio to television, from television to streaming, the way sports content is transmitted has kept changing. By McLuhan's theory, each media revolution should change sport itself.
But interestingly, in sport, the medium can change, yet the "message" (why we need sport) doesn't change in a thousand years.
In the 1950s people gathered around the radio to hear the World Cup; in the 1980s they gathered around the TV to watch the NBA; in the 2020s they watch highlights on TikTok on their phones. The medium changed three times, but the answer to "why do people watch sport" hasn't changed by a single word: we need to witness the moment a human limit is broken, we need heroes, we need to cheer for the same team alongside others.
AI is changing the "medium": how highlights are made, how content is recommended, how viewing is personalized. But AI can't change the "message": the underlying reason humanity needs sport. That is the counterexample to McLuhan's theory in the realm of sport: some "messages" won't be changed by the "medium."
AI can generate articles, write code, and design posters, but AI can never replicate the final 20 seconds of Game 7 of the 2010 NBA Finals. The heartbeat, the breath, the crowd's scream, the look in the players' eyes — none of it is programmable.
The second old wine is the contingency of competition: the game is never predictable. Sport's charm lies in uncertainty.
You can use AI to predict a game's result, analyze player stats, and simulate tactical effects. But you can never predict:
- Whether Curry's last shot goes in.
- Whether Messi buries a stoppage-time winner in the 90th minute.
- Whether Federer saves it on match point.
That uncertainty is precisely sport's greatest value. If a game's result could be predicted by AI with 100% accuracy, the game would lose its meaning, and the bookmakers would become 100%-margin companies.
The German philosopher Benjamin had a concept called "aura" — the singular breath that can only be felt "here and now," which mechanical reproduction can never capture.
AI can replicate Curry's shooting motion, generate flawless game highlights, and simulate the arena's sound and picture. But AI can't replicate the "aura": that shock, belonging to that moment in space and time alone, when you sit in the arena and watch Curry sink the game-winner right before your eyes and the whole place erupts.
I have a friend who spent $8,000 to attend a Super Bowl in person, and someone asked him: "Are you out of your mind? Isn't it the same watching the broadcast at home? The picture's even clearer."
He said: "Completely different. The broadcast shows you the ball's trajectory, but you can't feel the crushing pressure of tens of thousands of people holding their breath at once, and you can't feel the shock the moment the ball goes in — the floor's vibration traveling up through the soles of your feet. That moment's 'aura' — no amount of money can buy it back." That's why sports tickets keep getting more expensive yet live crowds keep growing. What people want isn't to "see the game" but to experience that irreproducible "here and now."
AI can change the way content is produced (auto-generated highlights, real-time data analysis, personalized recommendations), but AI can't change the answer to the question of why people need sport.
The answer is always: we need to witness the moment a human limit is broken; we need to find a fixed emotional anchor in an uncertain game. And that "moment" must be real, unpredictable, carrying the "aura."
(Figure in original.)
The third is the worship of heroes.
There's an interesting phenomenon: in the AI era, tech companies offer top AI scientists signing bonuses of $100 million — and top athletes' contracts also start at $100 million.
Why hasn't athletes' value been diluted by AI? Because, beyond being scarce, the top athlete's value isn't in "completing some task" but in "becoming a hero." That is something AI can never do.
You can have AI write a flawless analysis of a shooting motion, but you can't have AI worship a hero on your behalf. You can have AI generate Kobe's game highlights, but you can't have AI feel, for you, that thrill of watching Kobe sink the game-winner.
Humanity needs heroes — a need that doesn't change in a thousand years. From the ancient gladiator to the modern NBA star, the form of the hero changes, but the need does not.
The fourth is belonging: the 2024 Super Bowl drew 123 million American viewers to the live broadcast, an all-time high. That isn't a game — it's a collective ritual. The whole family gathered before the TV, friends' get-togethers, packed bars, an entire nation talking about the same thing on the same day.
The Stanford sociologist Granovetter had a famous theory: "the strength of weak ties." He found that the weak ties between strangers can, in certain settings, be more valuable than the strong ties of close relationships.
Finding a job depends not on your best friend but on the people you know but aren't close to — because they move in different circles and can give you different information and opportunities. Sporting events are exactly the setting where "weak ties" do their most powerful work.
On Super Bowl day, you and 123 million strangers across the country watch the same game at the same time. You don't know each other and may never meet, but in that moment you share a topic, an emotion, a memory. The next day at work, you can chat with the stranger in the elevator about "that game-winner last night." At the coffee shop, you can chat with the barista about "that quarterback was unbelievable." You open social media and the whole world is discussing the same thing.
AI can let everyone see different content — recommending videos to your taste, tailoring your feed to your algorithm. But this personalization comes at a cost: your recommendation algorithm is completely different from everyone else's. You live inside a filter bubble, with fewer and fewer "weak ties" to the outside world.
And "weak ties" are precisely the source of social cohesion. The Super Bowl gives 123 million Americans a shared topic — a massive-scale act of building "weak ties." That connection AI's personalization algorithms can't make, and don't want to — because the goal of personalization is to have you see only what you like, not to give you common ground with others.
AI can provide personalized recommendations and let everyone see different content. But AI can't provide the collective belonging of "I am watching the same game as 100 million other people."
Sport's "old wine" is these few unprogrammable emotional needs. AI can change the "bottle" (broadcast technology, data analysis, content distribution), but it can't change the "wine" (the emotional needs).
05 Fintech's old wine: the thousand-year logic of finance
What is fintech's "old wine"?
It's the underlying logic of finance: credit, clearing, asset allocation, interest-rate transmission, risk management. This logic doesn't change in a thousand years. Whether the ancient money house and remittance shop or the modern Stripe and Plaid, the essence is solving the same problem: how to move value between different parties safely and efficiently.
(Figure in original.)
From the money house to Stripe: the essential problem is the same — only the "bottle" differs.
What problem did the Shanxi remittance banks of the Ming and Qing solve? A Shanxi merchant made money doing business in Beijing and needed to move the money back to his hometown. But carrying silver overland with an armed escort meant bandits might rob you on the road. What to do? The remittance bank's solution: deposit your silver at its Beijing branch and take a note; arrive in Shanxi and take the note to the head office to withdraw silver. That was the earliest "clearing network."
What problem does the modern Stripe solve? A developer selling software on his own website needs to plug into the global payment network. But connecting to Visa, Mastercard, and banks in every country has too high a technical barrier. What to do?
Stripe's solution: you plug into Stripe's API with seven lines of code, and Stripe connects you to the global payment network. This is still a "clearing network," just with different technical means. The essential problem is the same: how to move value safely and efficiently. Only the "bottle" changed, from a paper note to an API.
(Figure in original.)
Why can't AI disrupt finance?
Many say AI will disrupt finance. I don't think so. AI can change finance's "bottle" (risk models, customer-service bots, robo-advisors), but AI can't change finance's "wine" (credit, clearing, risk management).
First, AI can build a risk model, but AI can't bear the responsibility of "lending wrong."
Suppose you're a brand's CFO deciding whether to lend a creator $200,000. AI tells you "this creator's default probability is 2%." Do you dare lend?
If the creator really defaults, will AI pay you $1 million? No. AI is only a tool, not a party that bears responsibility. So finance's core isn't prediction but bearing risk. AI can help you predict, but it can't bear it for you.
Second, AI can write compliance documents, but AI can't stand in for a person at an SEC hearing.
Finance is a heavily regulated industry. Whether a business model is compliant is up to the SEC, FINRA, and state financial regulators.
AI can help you generate a flawless compliance document, but when regulators question your model's logic, you need a real person sitting in the hearing to explain "why your risk model is fair, explainable, and non-discriminatory." That right to explain, AI cannot provide.
Third, AI can raise efficiency, but it can't replace trust.
Visa was founded in 1958 and took 60 years to become the global clearing standard, trusted by merchants and banks worldwide. That trust is the result of 60 years of accumulation. Over those 60 years, Visa processed trillions of transactions with almost no major security incident. That track record AI cannot replace.
PayPal was founded in 1998 — 26 years now — and is still the consumer-payments standard. Stripe was founded in 2010 — 14 years now — and has become the developer-payments standard.
Trust isn't an algorithm — it's time.
The moat of an AI application is "sustained technical leadership"; the moment you stop innovating, you're overtaken. Financial change has a different character: slow, heavily regulated, strong network effects — once successful, extremely hard to displace. Once fintech becomes the standard, the alpha is huge and endures for decades.
Fintech's "old wine" is this thousand-year logic of finance: credit, clearing, risk management. AI can change how it's implemented, but it can't change the underlying need.
06 A map for fellow travelers
If you're interested in the sports industry, I've written before: AI Can Remake Every Business — Except Sports. There you'll find four startup directions I'm bullish on.
In the messier environments outside Silicon Valley, founders must go full-stack and solve dirtier, heavier real-world problems. Over the next 5–15 years, the simple APIs will have all been built, and the opportunity will shift toward deep verticals and complex business processes. On fintech, my own understanding points to three incremental directions:
1. B2B tech and the CFO stack.
The opportunity: consumer payments are already extraordinarily smooth, but the flow of money between businesses on the B side still looks like it's from the last century. Cross-border reconciliation, invoice automation, and the intelligent hedging of accounts payable/receivable (AP/AR) are a vast blue ocean. Combined with AI, the enterprise ledger of the future isn't just a record but a financial brain with predictive and autonomous decision-making capability.
Representative companies:
- Ramp, Brex: corporate spend management and card issuance.
- Bill.com: AP/AR automation.
- Deel: globally compliant payroll.
2. Embedded finance and vertical scenarios.
The opportunity: finance will no longer be a standalone industry but the monetization method of all vertical SaaS. Money flows and credit solutions tailored to specific industries (healthcare, logistics, esports, creator economy). Whoever holds a specific industry's vertical workflow data can price risk more accurately than a traditional bank.
As a founder, don't build a general-purpose tool — build a deep solution for a specific scenario. Creator marketing, gaming going global, cross-border e-commerce — each scenario has its own logic of money flow.
3. Cross-border and globalized compliance networks.
The opportunity: as remote work and globalized supply chains become irreversible, the infrastructure that can seamlessly handle multi-currency, multi-entity compliance, taxation, and payroll (an advanced version of Deel) still holds an enormous structural dividend. Deel solved the employment problem, but there remain vast "non-employment" cross-border payment scenarios.
Coda
In the final 2 kilometers to 100 meters before a marathon's finish line, many people's legs are no longer their own and their minds begin to blur, but everyone still holds one belief: 2 kilometers to go, 100 meters to go. In that moment, the body fights the will. You could choose to slow down and walk across the line. But you won't — because the meaning of those 42 kilometers lies precisely in the persistence of these last 100 meters.
AI will change many things, but some things will never change:
- Humanity's craving for competition and heroes.
- Humanity's need for trust and certainty.
- Humanity's pursuit of long-term value.
Sport and fintech are my two anchors in the AI wave. They let me know that no matter how technology changes, some things are always worth believing in and holding to.
(Figure in original.)
The further along you go, the more it comes down to energy and stamina — the two things that let you hold sustained focus and the mental strength to grind a thing out to a result. You can freeload other people's ideas, but you can't freeload the path of thinking that produced those ideas. That's homework you have to do yourself; there's no shortcut, and you can't copy the answer.
AI will change many things, but AI can't change that contest between body and will. AI can tell you the optimal pace, predict where you'll hit the wall, and generate a perfect recovery plan. But AI can't run the last 100 meters for you. That's the value of sport: an irreplaceable human experience.
Finance is the same. AI can build risk models, optimize payment flows, and automate reconciliation. But AI can't build trust for you, can't bear responsibility for you, can't become the "source of truth" in your place. That's the value of fintech: an irreplaceable node of trust.
If you've read this far, I suspect you and I have some things in common.
I'm looking for people like this:
First: those who believe in long-termism. Frank Wang once said: many people conflate two things — whether you genuinely believe in a direction and are willing to grind at it for ten years, or whether the trend arrived and you want to get rich, start a company, and incidentally use its resources to pave your own path. It's like a talent show — who's going to say straight out, "I want to be famous"? Say it and you look cheap; what they say is, "I want to bring joy to everyone."
Second: those who don't blindly follow the crowd and have their own judgment. AI is hot, but AI's alpha will be extremely compressed. Sport is traditional, but sport's moat is culture. Fintech is hard, but hard is exactly what makes a moat.
Third: those who love what they do, not just for the money. I don't run marathons for the prize money — I run to prove that I can.
Fourth: those who believe in "new bottle, old wine," who believe the underlying logic won't change. Technology changes, models change, but humanity's needs for emotion, for trust, for the transfer of value don't change in a thousand years. Find that "old wine," then wait for the age to hand you a new bottle. Do it over and over, and wait for the wind to blow back.
If you're that kind of person, I'd very much like to meet you.
Whether you're building a sports startup, a fintech startup, or grinding away in any field you believe AI won't disrupt, I'd love to hear your story.
Because in this age of AI anxiety, we need more people who believe in the underlying logic, more people willing to persist for the long haul, more people who know "what won't change."