Pinned Tweet
Fuck this Chinese job market.
Young people are getting absolutely fucked. No real opportunities, shit pay that barely covers rent, and you're expected to grind 996 like a dog. Overtime is mandatory, weekends are optional, and if you complain you get replaced by the next desperate grad.
Everyone's killing each other for scraps. Involution is out of control — 300 people fighting for one mediocre position, then the winner gets paid like shit and worked to death. Graduates with master's degrees delivering food or sitting on their parents' couch.
Companies treat you like disposable labor. No raises, no promotions, just more work and the constant threat of being laid off. The economy's trash, private sector is dying, and the only "stable" path is fighting millions for a government job that might not even exist by the time you get it.
It's a fucking meat grinder. Opportunity? What opportunity. Just endless competition and poverty wages.
我操,一条他妈的零出处帖子,不到一天就骗了84万曝光。
“深圳已经开放外网,能直连TikTok和X了”
假的,纯纯瞎鸡巴传。我把能查的都翻了:
深圳市政府、工信部网站,屁都没有,一条“开放外网”的公告都查不到。
法规白纸黑字还钉在那儿。《计算机信息网络国际联网管理暂行规定》第六条写得清清楚楚:直接国际联网,必须走国家公用电信网的国际出入口信道。这条他妈的没改过。
被拿来瞎传的是前海那条“国际数据专用通道”。深圳市前海管理局2025年1月原文写得很明白,是给前海区域的企业客户开的国际通信服务,让企业连香港和境外快点。不是给全深圳市民用的,别他妈自己加戏。
还有人扯“试点”。2024年10月工信部启动的是北京、上海、海南、深圳四地“增值电信业务扩大对外开放试点”,放开的是外资可以在这几个地方办数据中心、内容分发这些业务,跟普通人手机上网没半毛钱关系。
所以别再他妈的转了。一条情绪拉满、零出处的帖子,几十万人看,几百人在评论区“实测”,最后就是一场集体高潮。
你们在深圳的,实话说,昨天有没有被这条骗到?
Holy shit. It took about 24 hours for Kylian Mbappé to disappear from China.
Sunday: 100th goal for Real Madrid.
Monday: Ballon d'Or runner-up. Same day, an old clip goes viral of him laughing at a meme of himself as China's founding leader.
Tuesday: Dongpeng, Mengniu and Milkground all drop him, all saying "zero tolerance." State TV halts La Liga games. La Liga apologizes. By evening, so does he.
That's why Western stars tiptoe around China. The money's there, the fans are there, and the rules aren't written in Madrid.
In your country, what's the one joke that kills a celebrity's sponsorships overnight?
三、AI 商业化与资本
Anthropic 招股书关键数字: 拟在纳斯达克上市,代码 ANT。2026 年上半年营收 162.9 亿美元(去年同期 11.6 亿);净亏损 807 亿美元,其中约 724 亿是可转债和衍生品公允价值变动造成的非现金损失,第三方测算 GAAP 经营亏损约 6.9 亿美元。SEC 原文、原帖、Luminix 拆解
最值得看的是算力承诺:
对 Google 至少 1145 亿美元,到 2033 年;
对亚马逊至少 1100 亿美元,到 2036 年,这两笔用不用都得付;
对 SpaceXAI(原 xAI)最多 1205 亿美元,到 2029 年,过了短暂的不可撤销期后,提前 90 天通知就能退出。
Meta 最多融资 950 亿美元: 先发约 350 亿(225 亿普通股加 125 亿强制可转换优先股),之后视情况按市价分批增发(ATM)最多 600 亿。英伟达认购 30 亿、AMD 认购 10 亿。截至 9 月底,Meta 现金约 824.5 亿美元,长期债务约 809.2 亿,未来支出承诺约 8250 亿。Meta 公告、CNBC、英伟达公告
值得警惕:芯片商入股大客户,客户再拿这笔钱买它们的芯片,这是典型的循环融资。
投入和回报的缺口: Apollo 首席经济学家 Torsten Slok 测算,标普 500 剔除七巨头后的 493 家公司,经通胀调整的人均营收约 52.5 万美元,2022 年以来基本没动;七巨头接近 80 万美元。Apollo 原文、TechCrunch。讽刺的是,给 SpaceX 提供 GPU 融资的正是 Apollo。
一级市场: Hex 融资 1 亿美元、估值 12 亿;General Compute 融资 1 亿、估值 8 亿;Infisical 融资 4000 万、估值 6 亿(都来自 X 汇总:Hex、General Compute、Infisical)。「Instinct 估值冲 300 亿」只是传言,未核实。
判断: 营收确实在爆发,Anthropic 一年涨了约 14 倍,但算力账单按十年算,AI 公司以外的企业还没见到回报。缺口靠股权、供应商担保和私募信贷三种钱在填,任何一环收紧,最先出事的是信用最弱的新云厂商。
AI 产业深度日报|2026-10-28
今天只有一条主线:AI 的钱从哪来,出了事谁兜底。过去 48 小时,Anthropic 递交了上市申请(S-1 招股书),Meta 宣布最多融资 950 亿美元,博通为 Anthropic 的 630 亿美元芯片融资做担保,SpaceX 签下 400 亿美元的 GPU 租赁。模型发布反倒成了配角。
一、大模型与产品
OpenAI 发布 GPT-6.1 Terra。 主打操作电脑(看懂屏幕、跨应用执行任务)的快速低价模型。官方称它在 Agents' Last Exam 测试上和旗舰 GPT-6 Astra 打平,价格是后者的十分之一,速度快三倍以上;API 每百万输入 token 1 美元、输出 5 美元。目前只在 ChatGPT Work 和 Codex 里提供。OpenAI 博客、原帖
这一代的竞争重点,已经从谁最聪明转到谁能用十分之一的价格干完八成白领杂活。压力主要落在 Anthropic 的中端价位:OpenAI 拿 Claude Opus 5.5 对比,称单任务成本低约 9 倍,但这是自测,要等第三方复核。
OpenAI 公布三项 AI 产出的科学成果: 一是号称找到 3-SAT 问题的次指数时间算法,成立的话就推翻了计算复杂性里的「指数时间假设」(ETH);二是调和分析中的局部光滑猜想;三是包括 Yang-Mills 在内的量子场论存在性证明。OpenAI 不要学术署名,交给外部专家组审核。OpenAI 说明
目前都还只是「声称」。第三方追踪指出,Lean 形式化证明覆盖的界比论文正文弱(分析)。X 上说 Claude 也独立做出同类结果,没找到可靠来源,先别当真。
DeepMind 发布 Genie 4 世界模型: 实时生成可交互的写实场景,带声音和对话,先给受邀测试者,之后开放给 AI Ultra 订阅用户。DeepMind 原帖、TechCrunch。最先受益的是游戏、影视预演和机器人仿真训练。
Anthropic: 和美国总务署(GSA)签了新协议,Claude 给各级政府打六折(原帖);公布与六家药企的研发合作,并开始自己做药物发现(原帖)。「Fable 5.5 即将发布」只是 Azure 模型目录的线索加社区猜测,没有官方确认。
国产模型:
DeepSeek V4.1-Flash-1017 连续三周在 OpenRouter(汇集多家模型的调用平台)周调用量第一,10/19 至 10/25 处理了 33.6 万亿 token,环比增长 801%。环球时报、第一财经。不过另有数据显示开源模型只占付费推理支出的 5% 到 10%(X 汇总),调用量第一不等于赚钱第一。
月之暗面 10/23 发布 Kimi K3.1 Pro,承诺 11/4 前开放权重。官方博客
百度开源 Unlimited OCR 2,30 亿参数、MIT 协议,一次能解析几百上千页文档。演示帖
蚂蚁开源 Ling-3.1-flash-1019。X 汇总
智谱「GLM-5.5」只是一个零调用量的目录条目,没有官方消息。
二、算力与基建:谁受益,谁承压
博通为 Anthropic 的 630 亿美元融资兜底。 美银、花旗、摩根士丹利、黑石出资成立特殊目的公司(SPV),买博通的定制 AI 芯片和网络设备再租给 Anthropic,2027 年开始交付。博通提供残值担保:Anthropic 违约时,博通要补足「未付租金的 72.5%」和设备变卖价之间的差额,最高敞口约 430 亿美元,五年内降到零。博通公告、8-K 摘要。据报道谈判中银行一度想退出,是 CEO 陈福阳接受更高利率才谈成。Yahoo Finance
SpaceX 签约约 400 亿美元 GPU 融资租赁: Apollo 管理的实体出资采购英伟达 Vera Rubin 芯片,租期约五年,2027 年上半年起租。SpaceX 公告、SEC 8-K。马斯克借机说,只靠出租算力的新云厂商在债市越来越难融到钱。X 汇总
新云厂商分化: Nebius 拿到花旗 35 亿美元、期限 364 天的有担保贷款,又签了 50 兆瓦机房租约,周二涨超 10%。Nebius 公告、DCD。但 364 天的短贷意味着明年要续借,信用风险只是往后推了。
台积电据报考虑把 CoWoS 先进封装涨价 10% 到 20%, 设备交期拉长到 6 到 8 个月,台积电未确认。TrendForce。另据韩媒,阿里、腾讯、字节在和三星、SK 海力士谈锁定到 2035 年的服务器内存长约。X 汇总
小结: 受益方是有涨价能力的台积电、拿到长约的存储厂,以及卖芯片还入股客户的英伟达和博通。承压方是靠借钱扩张的新云厂商,还有替客户兜底的芯片商自己,博通的或有负债在变大。
今日深度:AI 的账单,正在从股东转给债主和供应商
1. 来龙去脉。 2023 到 2025 年,AI 基建主要靠巨头自己的现金流和风险投资。到 2026 年下半年,账单大到这些钱撑不住了。Meta 账上现金和长期债务几乎打平,却背着约 8250 亿美元的未来支出承诺,于是开始大规模发股,还拉芯片供应商当基石投资人。Anthropic 上半年营收 163 亿美元,对 Google、亚马逊、SpaceXAI 的算力承诺加起来却超过 3400 亿美元,只能通过 SPV 租赁拿芯片,而不是直接买。
2. 新变化:供应商成了担保人。 博通的残值担保、Apollo 给 SpaceX 的租赁,本质上都是拿 GPU 当抵押物发债。前提是五年后这些芯片还值钱、承租方付得起租金。博通的保护线是未付租金的 72.5%,设备残值一旦跌破这条线,差额由博通自己扛。
3. 各方观点。
乐观派:陈福阳认为这说明「有资本愿意为增长融资」;SpaceX 称投资回收期不到一年。
谨慎派:Slok 指出科技行业以外还没看到 AI 回报;银行在博通这笔交易里一度要更高利率才肯出钱;马斯克点名新云厂商在债市遇冷。
我的看法:Anthropic 和 SpaceXAI 的协议有「90 天通知即可退出」的条款,说明连买方自己也在给算力过剩留后路。
4. 对投资的影响(基于公开信息的分析框架,不构成买卖建议)
议价能力最强的是瓶颈环节:台积电 CoWoS 封装(在谈涨价)和存储(在签长约),存储的需求能见度确实在变好。
博通、英伟达的收入质量要多打个问号,一部分需求是它们自己担保或入股催出来的。博通下一份财报(12/9)要盯或有负债的规模。
新云厂商风险最高。Nebius 这类公司拿到融资就大涨,但贷款期限短,再融资利率一上行,压力最先落到它们身上。
Anthropic 上市后的定价,会成为整个 AI 估值体系的锚。
5. 对就业的影响。 像 GPT-6.1 Terra 这种低价干白领杂活的模型,加上 Slok 的数据显示其他企业还没赚到钱,会把企业推向用 AI 压人力成本,而不是用 AI 扩张业务。最先承压的是表格录入、发票处理、初级剪辑这类重复性岗位。对你来说,更实际的方向是把 AI 当内容生产的杠杆,同时守住 AI 复制不了的东西:你自己的受众和独立判断。
2. 奈飞(NFLX):62.52 美元,离高点 -42.0%
估值
过去 12 个月市盈率 18.1 倍,市销率 5.2 倍;远期市盈率 16.6 倍(Yahoo)或 17.6 倍(Superworth)。
和自己比:10 年中位数 48.4 倍,现在处在第 0 百分位,比过去 10 年任何一天都便宜;10 年区间 17.8 到 188 倍。Superworth
和同行比:迪士尼 14.6 倍,规模相近的娱乐公司中位数也是 14.6 倍。
技术面:52 周区间 62.48 到 116.73 美元,收盘就在 52 周低点附近;50 日均线 73.71、200 日均线 82.01 都远在股价上方,近 5 日跌 9.0%。
基本面:营收增长 11.8%,每股收益增长 47.5%,毛利率 50.7%,营业利润率 34.6%(Yahoo);总负债 175 亿美元。43 位分析师的平均目标价 84.76 美元,区间 57 到 126。用户人均观看时长从 66 小时降到 49 小时。X 新闻
结论:估值已进入自身历史偏低区间。
要等的信号:用户活跃度数据止跌;股价在 52 周低点附近企稳,不再创新低。
看多理由:流媒体规模第一,盈利和现金流都在增长,估值是 10 年最低。
看空理由:估值低,是因为市场把它从成长股重新定价成了成熟股;YouTube 在抢用户时间,使用频率下降,这比单个季度的业绩更根本。
3. 耐克(NKE):37.56 美元,离高点 -43.3%
估值
过去 12 个月市盈率 18.1 倍,市销率 1.2 倍;远期市盈率 24.2 倍(Yahoo)或 29.5 倍(Superworth)。
和自己比:10 年中位数 40.9 倍,现在处在第 5 百分位;10 年区间 21.6 到 74.1 倍。Superworth
和同行比:规模相近的可选消费公司中位数 18.7 倍。
技术面:52 周区间 31.97 到 69.14 美元;略高于 50 日均线 37.09,200 日均线 46.47 远在上方。
基本面:上季度营收 112 亿美元,同比下降 4%,其中大中华区下降 26%;公司指引 2027 财年营收下降高个位数,调整后每股收益 1.15 到 1.35 美元,分析师原来预期约 1.67 美元。耐克公告、Yahoo 营业利润率只有 7.9%(Yahoo)。重组计划「Pace」到 2031 财年要省 25 亿美元,但大部分要到 2029、2030 财年才兑现。路透 37 位分析师的平均目标价 36.50 美元,区间 19 到 60,比现价还低。
结论:仍偏贵。 和自身历史比是低的,但营收和利润都在下滑,远期市盈率 24 到 30 倍,比同行中位数贵。
要等的信号:大中华区止跌;运动时尚和乔丹品牌企稳;利润率触底回升。
看多理由:全球运动品牌第一,跑步等专业品类在恢复;股价已经跌了 43%。
看空理由:RBC 分析师说「情况会先变差再变好」;分析师平均目标价低于现价,基本面还在恶化。
4. 希捷(STX):667.85 美元,离高点 -39.0%(盘后约 630 美元,-5.7%)
估值
过去 12 个月市盈率 48.1 倍,市销率 12.5 倍;远期市盈率 11.6 倍(Yahoo)或 18.0 倍(Superworth)。
和自己比:10 年中位数 11.9 倍,现在处在第 75 百分位;10 年区间 7.0 到 116 倍。Superworth
和同行比:规模相近的电脑硬件公司中位数 13.0 倍。
技术面:52 周区间 217.67 到 1145 美元;股价跌破了 50 日均线 809.96 和 200 日均线 687.14,盘后约 630 美元,离 200 日线更远了。
基本面:营收增长 48.5%,毛利率 45.6%,营业利润率 43.1%(Yahoo)。这次财报的每股收益和下季度指引都超预期。Investing.com 24 位分析师的平均目标价 1091.58 美元,区间 700 到 1600。
结论:仍偏贵。 两家数据源的口径差得很大,按 Superworth 在自身历史的第 75 百分位。周期股在盈利高点时市盈率看起来低,恰恰是最危险的时候。
要等的信号:AI 带来的大容量硬盘需求能不能持续到 2027 年;估值回到 10 年中位数附近。
看多理由:AI 数据中心对大容量硬盘的需求是实打实的;业绩连续超预期。
看空理由:好财报却盘后大跌,说明市场在交易周期见顶;从 52 周低点涨了约 3 倍,获利盘很重。
1. Meta(META):620.19 美元,离 2026 年高点 -20.2%
估值
过去 12 个月市盈率 21.2 倍,市销率 6.6 倍;远期市盈率 17.6 倍(Yahoo)或 19.4 倍(Superworth)。
和自己比:10 年中位数 23.6 倍,现在处在第 20 百分位;10 年区间 8.4 到 32.3 倍。Superworth
和同行比:规模相近的通信服务公司中位数 16.1 倍,谷歌 23.4 倍。
技术面:52 周区间 520.26 到 779.82 美元;股价跌破了 50 日均线 665.65 和 200 日均线 632.29,近 5 日跌 10.7%。
基本面:营收增长 26.1%,毛利率 81.0%,营业利润率 35.4%,过去 12 个月自由现金流 168.5 亿美元(Yahoo)。但最近一个季度自由现金流已经转负,2026 年资本开支 1350 亿到 1450 亿美元,股权融资 975 亿美元。X 新闻、路透 58 位分析师的平均目标价 846.79 美元,区间 629.06 到 1050。
结论:合理,略低于自身历史中枢。
要等的信号:融资定价和股本被摊薄的实际幅度;资本开支能不能在 2027 年见顶;自由现金流什么时候转正。
看多理由:广告主业增长 26%,利润率很高;估值在自身 10 年的第 20 百分位。
看空理由:一边高增长,一边要找股东要近千亿美元,说明主业的现金流已经撑不住它的 AI 野心;股价跌破两条均线,技术面偏弱。
周二老美又收了根阳线。
标普涨 0.60%,纳指 100 涨 1.29%,道指基本没动。
VIX 17.52,大饼 8 万出头,10 年期美债 4.99%。
1. 一句话结论:今天是「等」。
纳指 100 的回撤从周一的 -6.25% 收窄到 -5.04%,离 -9% 第一档还差 4.2%,又远了一步。
标普只回撤了 1.2%,VIX 不到 18,都离规则线很远。
个股方面,Meta 新跨过了 20% 线,但它是因为基本面边际恶化跌下来的,不是好公司打折。
2. 核心原因归类(按重要程度排)
① 资本开支和自由现金流(主因)
两天里,同一个逻辑被验证了两次:
Meta 把股权融资加到 975 亿美元,股价再跌 1.6%,跌破了 20% 线。路透
博通拿到 630 亿美元的订单融资,涨了 4%。
「资本开支越大,股价跌越狠;资本支出低于市场预期,增长和营收高于市场预期,自由现金流充沛的,股价就大涨。」
但要多看一层:博通涨,是因为别人花钱买它的东西,可这笔钱是向银行借的,博通还为其中 435 亿美元提供了残值担保。X 新闻 收钱的一方开始替花钱的一方兜底,说明 AI 产业链的现金流链条在拉长、变脆。
② 美债收益率
10 年期 4.992%,2 年期连涨三天到 4.735%。
天花板没动,指数就很难走出大行情。明天凌晨沃什的表态,决定这块天花板是抬高还是压低。
③ 周期
油价回落,能源股垫底;希捷财报超预期,盘后却大跌。
两件事是同一个道理:周期股在高位时,好消息已经被定价,市场关心的是周期什么时候拐头。
④ 估值和风格
10 月防守风格、等权指数、非 AI 股都跑赢了大盘。Kobeissi、Liz Ann Sonders 这是资金在给 AI 估值降温,不是恐慌。VIX 才 17,离他的出手温度还远。
3. 规则线距离表
周二个股回撤线的变化(2026 年高点按最新收盘重新计算)
新跨过 20% 线:Meta(从 -18.9% 到 -20.2%)。
盘后可能跨线:Robinhood 收盘 -24.2%,盘后价约 87.8 美元,相当于 -29.6%,贴近 30% 线;希捷的盘后价相当于 -42.4%。以周三收盘为准。
回到线以内:博通(从 -31.6% 回到 -28.9%,回到 30% 线以内);阿斯麦(从 -20.0% 回到 -17.8%,回到 20% 线以内);SpaceX(从 -30.5% 回到 -29.7%)。
在 20% 到 30% 之间:谷歌 -20.4%、高盛 -20.9%、百事 -21.7%、美国银行 -21.8%、迈威尔 -23.0%、美光 -23.5%、IBM -24.0%、高通 -26.6%、麦当劳 -26.9%、Strategy -27.8%、AT&T -28.0%。
超过 30%:英特尔 -33.5%、诺和诺德 -39.6%、奈飞 -42.0%、耐克 -43.3%、泛林 -44.4%、应用材料 -45.2%、甲骨文 -47.4%、Arm -48.4%、CoreWeave -56.8%,以及存储股和中概股。
4. 值得关注的清单(只是进入研究,不是动手)
纳指 100 / QQQ:离第一档还差 4.2%。明天凌晨有美联储决议和两家云厂商的财报,波动可能放大。宽基是这套框架里最优先的工具。
谷歌:-20.4%。搜索第一,云业务还在追赶,算「第一」不算「唯一」;生意看得懂;增长没坏,但自由现金流在边际变差。明天凌晨的财报看资本开支和现金流。
阿斯麦:-17.8%,回到了 20% 线以内。EUV 光刻机全球唯一供应商,是第一兼唯一,基本面没变差。等它重新回到 20% 线附近再看。
博通:-28.9%,回到了 30% 线以内。营收增长 85%,基本面没坏,但开始替客户兜底芯片残值。先看清这部分风险,看 12/9 的财报会不会披露。
奈飞:-42.0%,估值是 10 年最低。生意看得懂,但它不是「唯一」,用户观看时长在下降,基本面边际在变差。只观察,等活跃度数据止跌。
台积电:-10.1%,AI 链条里最确定的环节,等它到 20% 线。
不碰的清单
Meta:虽然新跨过了 20% 线,但原因是资本开支越来越大、自由现金流转负、要向股东融资 975 亿美元,基本面在边际恶化,不符合「基本面没变差」这个前提。
耐克:营收和利润都在下滑,分析师平均目标价低于现价,重组要到 2029 年以后才见效。跌得多不等于便宜。
应用材料、泛林:跌了 44% 以上,但估值还在自身历史的偏高位置,周期没见底。
甲骨文、CoreWeave:高杠杆、现金流为负,周二的大涨是融资消息带来的反弹,不碰。
特斯拉、SpaceX、Palantir:估值看不懂。
Robinhood、Strategy、Circle 这类加密相关股:看不懂,估值没有锚。
5. 仓位提醒
永不满仓,总仓位最多七成到七成半,至少留三成现金。
从明天凌晨到周五事件密集。就算纳指碰到第一档,也是金字塔式分批买,越跌买得越多,每一笔都留余地,不一把梭哈。
Fuck. 8.5 million entries. That's how big China's court blacklist for unpaid debts was in early 2025.
They call them "laolai". Deadbeats. Get on that list and you can't fly, can't take a fast train, can't book a nice hotel. Your name sits on a public court website for anyone to look up.
And the line never stops. In 2025 alone, courts added people to that list 2.34 million times. In one year.
So are millions of Chinese just scumbags who love not paying? No. That's the lazy story.
Look at what's actually clogging the courts. In 2023, the #1 civil lawsuit in China wasn't divorce. It was loan contracts: 3.99 million cases. Every app on your phone wants to lend you money, and the bill ends up in court.
And here's the number nobody puts on a poster: the top court itself told lawmakers in 2018 that in about 43% of cases that reach enforcement, the debtor has literally nothing left to take. Not hiding it. Broke.
To be fair, the blacklist is supposed to be for people who CAN pay and won't. Those assholes exist, and some fake divorces to hide assets. Fuck those guys.
But the top court now says a big chunk of debtors are "honest but unfortunate": got sick and went broke, or their business died. Since 2024, more entries come off the list each year than go on.
Easy loans in. Lawsuits out. Millions of names in between.
What happens in your country when you can't pay back a loan? Can they stop you from boarding a train?
Here's what happens in China if you can't pay back those app loans.
First your credit report gets wrecked. Then the collection calls start. Then they sue you.
Lose in court and still don't pay, and a judge can hit you with a consumption ban. These are real rules from China's Supreme People's Court:
No flights.
No high-speed G trains. Not even the cheapest seat.
No sleeper berths.
No star-rated hotels.
No vacations.
No buying a home. No buying a car you don't need for work.
Your kid can't go to a pricey private school.
If the court decides you can pay but won't, you go on the public blacklist. Your name ends up on a court website for anyone to look up. People call you a "laolai", a deadbeat. And a government job? Forget it.
In 2025, courts added people to that blacklist 2.34 million times.
All this started with two taps on a phone and a "credit line" you never asked for.
Would you ride a slow train for 20 hours because you missed a loan payment?
Remember my brother? This is how it happens.
In China, you don't go to a bank to borrow money. You open Alipay. WeChat. Didi. Ctrip. Even an electronics store app.
5 apps. Each one advertises up to ¥200,000. That's ¥1 million in "limits" sitting in your phone, between your food delivery and your taxi ride.
Nobody explains the interest. Nobody asks why a 22-year-old needs ¥200,000. The button is right there.
Household consumer debt in China, excluding mortgages: ¥20.83 trillion.
My brother didn't even spend the money. A "friend" did. The apps still want it back.
Does your Uber app offer you a loan?
Part 4: the numbers behind the rant.
18% of young Chinese in the urban labor force can't find a job. Official figure, students already excluded.
Every summer a new batch of graduates hits the same wall: 14.9% in June, 18.9% in August. Same peak as last year.
Full chart here 👇
Remember my Chinese job market rant? Here's the chart.
18% of young Chinese in the urban labor force can't find a job. That's the official number, from China's own statistics bureau. And it already excludes students.
Look at the pattern. Every summer, a new wave of graduates hits the market and the line explodes. June 2026: 14.9%. August: 18.9%. Same peak as last year. Nothing got better. They just printed a new batch of graduates into the same wall.
Then it "recovers" a bit. Not because jobs appear. Because people give up, take gig work, delivery, livestream selling, or go back home and study for civil service exams again.
12 years of school. 4 years of college. Your parents' savings. And the prize is a 1-in-5 shot at sitting at home.
What's youth unemployment like where you live?
Remember my Chinese job market rant? Here's the chart.
18% of young Chinese in the urban labor force can't find a job. That's the official number, from China's own statistics bureau. And it already excludes students.
Look at the pattern. Every summer, a new wave of graduates hits the market and the line explodes. June 2026: 14.9%. August: 18.9%. Same peak as last year. Nothing got better. They just printed a new batch of graduates into the same wall.
Then it "recovers" a bit. Not because jobs appear. Because people give up, take gig work, delivery, livestream selling, or go back home and study for civil service exams again.
12 years of school. 4 years of college. Your parents' savings. And the prize is a 1-in-5 shot at sitting at home.
What's youth unemployment like where you live?
Fuck this Chinese job market.
Young people are getting absolutely fucked. No real opportunities, shit pay that barely covers rent, and you're expected to grind 996 like a dog. Overtime is mandatory, weekends are optional, and if you complain you get replaced by the next desperate grad.
Everyone's killing each other for scraps. Involution is out of control — 300 people fighting for one mediocre position, then the winner gets paid like shit and worked to death. Graduates with master's degrees delivering food or sitting on their parents' couch.
Companies treat you like disposable labor. No raises, no promotions, just more work and the constant threat of being laid off. The economy's trash, private sector is dying, and the only "stable" path is fighting millions for a government job that might not even exist by the time you get it.
It's a fucking meat grinder. Opportunity? What opportunity. Just endless competition and poverty wages.
Part 3: it's not just jobs. Even your friends will harvest you.
xitter.cf/AkZfl/status/211496275…
Fuck. My little brother just got scammed out of 15,000 yuan (about $2,000). Not by some Myanmar phone call. By a guy he'd known for years.
The pitch: Chinese shopping apps are full of crazy subsidies. Max out your credit, buy iPhones, laptops, bags at the low price, I flip them, pay you back plus a cut.
So he maxed out Douyin, Kuaishou, Huabei, JD. In China, two taps and the app throws a credit line at you like it's fattening a pig.
It worked for two years. Then it blew up overnight. Over a dozen people. More than 4 million yuan gone.
The money's gone. The debt isn't. My brother's paying off credit he never got to spend, and the apps don't give a shit.
China's police say this kind of scam is now the No. 1 fraud in the country: 25% of all online fraud cases.
Friends harvest you like leeks. Apps hook you with credit. Trust is dead. The credit limit is still there.
Ever been scammed by a "friend"?
Fuck. My little brother just got scammed out of 15,000 yuan (about $2,000). Not by some Myanmar phone call. By a guy he'd known for years.
The pitch: Chinese shopping apps are full of crazy subsidies. Max out your credit, buy iPhones, laptops, bags at the low price, I flip them, pay you back plus a cut.
So he maxed out Douyin, Kuaishou, Huabei, JD. In China, two taps and the app throws a credit line at you like it's fattening a pig.
It worked for two years. Then it blew up overnight. Over a dozen people. More than 4 million yuan gone.
The money's gone. The debt isn't. My brother's paying off credit he never got to spend, and the apps don't give a shit.
China's police say this kind of scam is now the No. 1 fraud in the country: 25% of all online fraud cases.
Friends harvest you like leeks. Apps hook you with credit. Trust is dead. The credit limit is still there.
Ever been scammed by a "friend"?









