[
 {
  "t": "The State of AI 2025",
  "org": "Bessemer Venture Partners",
  "au": "Kent Bennett, Talia Goldberg 等",
  "date": "2025-08-13",
  "url": "https://www.bvp.com/atlas/the-state-of-ai-2025",
  "kind": "报告",
  "gist": "研究 20 家高增长 AI 初创，分两类：Supernovas 第 1 年约 $40M ARR、第 2 年约 $125M，约 1.5 年到 $100M，但毛利仅约 25%（常为负）；Shooting Stars 第 1 年约 $3M、第 4 年约 $103M，毛利约 60%，提出 Q2T3 增长基准。",
  "topic": "增长速度",
  "co": [
   "Perplexity",
   "Abridge",
   "Cursor"
  ],
  "stance": "谨慎",
  "top": true,
  "cav": "样本是 Bessemer 自己研究的 20 家，不是全市场；报告未把具体公司对号入座到两类；留存脆弱、'薄包装' 是报告自己列出的风险。"
 },
 {
  "t": "The Cloud 100 Benchmarks Report 2025",
  "org": "Bessemer Venture Partners",
  "au": "BVP 团队",
  "date": "2025-09-03",
  "url": "https://www.bvp.com/cloud100",
  "kind": "报告",
  "gist": "Cloud 100 榜单总估值首次破 $1 万亿；据二手转述，平均上榜公司 7.5 年到 $100M ARR，其中 AI 公司 5.7 年，AI 公司平均收入倍数 24x（非 AI 19x）。",
  "topic": "榜单",
  "co": [
   "OpenAI"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "官方榜单页只确认「首次超 $1 万亿」；7.5/5.7 年等数字出自 Metrics Brothers 播客页与中文转述，未在 BVP 页面直接核到；日期取自二手来源。"
 },
 {
  "t": "Top 100 Gen AI Consumer Apps — 7th Edition",
  "org": "Andreessen Horowitz (a16z)",
  "au": "Olivia Moore",
  "date": "2026-10-05",
  "url": "https://a16z.com/100-gen-ai-apps-7/",
  "kind": "报告",
  "gist": "首次加入 YipitData 美国信用卡消费榜：前 1% 付费者占消费 AI 支出 19.5%，超过后 50% 的总和（16.6%）；前 1% 月均 $903，中位付费者 $25；仅 4.5% 合格消费者持有 ChatGPT/Gemini/Claude 付费订阅。",
  "topic": "榜单",
  "co": [
   "ChatGPT",
   "Claude",
   "Gemini",
   "Perplexity",
   "Canva",
   "Notion",
   "Suno"
  ],
  "stance": "谨慎",
  "top": true,
  "cav": "仅美国消费者、信用卡面板样本，a16z 明说不能当作公司总收入；a16z 是 OpenAI 的投资者。"
 },
 {
  "t": "Top 100 Gen AI Consumer Apps: March 2026",
  "org": "Andreessen Horowitz (a16z)",
  "au": "Olivia Moore",
  "date": "2026-03-09",
  "url": "https://www.a16z.news/p/top-100-gen-ai-consumer-apps-march",
  "kind": "报告",
  "gist": "ChatGPT 网页流量约为 Gemini 的 2.7 倍、移动 MAU 约 2.5 倍；Claude 与 Gemini 美国付费订阅增长快；榜单扩到 CapCut、Canva、Notion 等加了 AI 的老产品。",
  "topic": "榜单",
  "co": [
   "ChatGPT",
   "Gemini",
   "Claude",
   "Midjourney",
   "Canva"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "这一版只按流量排名（Similarweb/Sensor Tower），没有收入数据；作者提示流量榜会低估 Claude Code 这类重度用户产品。"
 },
 {
  "t": "2025: The State of Generative AI in the Enterprise",
  "org": "Menlo Ventures",
  "au": "Tim Tully, Joff Redfern, Deedy Das, Derek Xiao",
  "date": "2025-12-09",
  "url": "https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/",
  "kind": "报告",
  "gist": "企业生成式 AI 支出 2025 年 $37B（2024 年 $11.5B）：应用 $19B、基础设施 $18B；编程 $4.0B，垂直 AI $3.5B（医疗约 $1.5B）；称是软件史上增长最快的品类。",
  "topic": "市场规模",
  "co": [
   "Anthropic",
   "OpenAI"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "495 名美国企业 AI 决策者问卷（2025-11-07 至 25）加自下而上建模；不含芯片、推理服务和嵌入既有软件的 AI 功能；Menlo 是 Anthropic 的投资方。"
 },
 {
  "t": "Indexing the AI economy",
  "org": "Stripe",
  "au": "Stripe（页面未署名）",
  "date": "",
  "url": "https://stripe.com/guides/indexing-the-ai-economy",
  "kind": "数据",
  "gist": "Stripe 上收入前 100 的 AI 公司，达到 $1M 年化收入的中位时间 11.5 个月，比增长最快的 SaaS 公司早 4 个月；2020-2023 年成立的新一批达到重要里程碑的速度约为老一批的 3 倍。",
  "topic": "增长速度",
  "co": [
   "Abridge",
   "Harvey",
   "Midjourney"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "只覆盖用 Stripe 收款的公司；页面未标发布日期，SaaS 对照是文中「早 4 个月」反推，完整方法在需下载的 PDF 里。"
 },
 {
  "t": "Stripe says AI startups are growing faster than SaaS ever did, and calling them wrappers 'misses the point'",
  "org": "TechCrunch",
  "au": "Mary Ann Azevedo",
  "date": "2025-02-27",
  "url": "https://techcrunch.com/2025/02/27/stripe-ceo-says-ai-startups-are-growing-faster-than-saas-ever-did-and-calling-them-wrappers-misses-the-point/",
  "kind": "博客",
  "gist": "Stripe 年度信：2024 年收入前 100 的 AI 公司 24 个月达到 $5M 年化收入，2018 年前 100 的 SaaS 公司要 37 个月；信中举 Cursor 收入破 $1 亿、Lovable 3 个月 $17M ARR、Bolt 2 个月 $20M ARR。",
  "topic": "增长速度",
  "co": [
   "Cursor",
   "Lovable",
   "Bolt"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "媒体对 Stripe 年度信的转述；两组年份不同（AI 2024 vs SaaS 2018），按收入前 100 取样，有幸存者偏差。"
 },
 {
  "t": "Founders are using creative accounting to boost lofty 'ARR'",
  "org": "Fortune",
  "au": "Allie Garfinkle",
  "date": "2025-09-28",
  "url": "https://www.fortune.com/2025/09/28/how-is-arr-calculated-startups-venture/",
  "kind": "博客",
  "gist": "六位 VC 称创始人把试点、一次性合同、未激活合同算进 ARR：一家国防科技公司报 $32.5 万 ARR，实为两周试点加口头承诺；Cluely 据称一周内 ARR 翻倍到 $700 万；'vibe revenue' 不是真正的 recurring。",
  "topic": "收入口径",
  "co": [
   "Cluely",
   "Cursor",
   "Midjourney"
  ],
  "stance": "唱反调",
  "top": true,
  "cav": "匿名 VC 引述，没有披露具体违规公司的审计数据；文中并列 Cursor、Midjourney 的快速增长作对照。"
 },
 {
  "t": "Clouded Judgement 3.22.24 - ERR vs ARR and the Conundrum of AI Revenue Streams Today",
  "org": "Altimeter（Clouded Judgement）",
  "au": "Jamin Ball",
  "date": "2024-03-22",
  "url": "https://cloudedjudgement.substack.com/p/clouded-judgement-32224-err-vs-arr",
  "kind": "博客",
  "gist": "提出 ERR（Experimental Runrate Revenue，实验性年化收入）：大量 AI 收入来自还在试功能、试厂商的买家，留存不具备真 ARR 的特征；建议创始人先分清 ARR 与 ERR 再做招聘和融资计划。",
  "topic": "收入口径",
  "co": [],
  "stance": "唱反调",
  "top": true,
  "cav": "2024 年 3 月的文章，较早；是定性论证，没给全市场的 ERR 占比数字。"
 },
 {
  "t": "A Sequoia partner says a lot of the AI industry's revenue is 'experimental'",
  "org": "Business Insider",
  "au": "Chong Ming Lim（引 Sequoia 合伙人 Alfred Lin 在 Sourcery 播客）",
  "date": "2025-10-29",
  "url": "https://www2.businessinsider.com/sequoia-partner-ai-gold-rush-experimental-revenue-alfred-lin-2025-10",
  "kind": "播客",
  "gist": "Alfred Lin：AI 热潮带来大量「实验性收入」，有创始人把当月试点收入乘 12 就叫 recurring；'我宁要增长慢的优质收入，也不要增长快的劣质收入'，看留存和公司速度而不只是收入增速。",
  "topic": "收入口径",
  "co": [
   "Replit"
  ],
  "stance": "唱反调",
  "top": false,
  "cav": "是 BI 对播客的转述，Lin 未回应 BI 置评；日期取自 BI 链接与搜索元数据（播客周一发布）。"
 },
 {
  "t": "What's a Young AI Startup Earning? It's Unclear, Venture Capitalists Say",
  "org": "The Wall Street Journal",
  "au": "WSJ Pro",
  "date": "2025-11-27",
  "url": "https://www.wsj.com/articles/whats-a-young-ai-startup-earning-its-unclear-venture-capitalists-say-a0ce4b6f",
  "kind": "博客",
  "gist": "多种新定价模式让 AI 初创的顶线数字飘忽，VC 说要加班才能分清哪些收入是真的，才敢投。",
  "topic": "收入口径",
  "co": [],
  "stance": "唱反调",
  "top": false,
  "cav": "WSJ 付费墙，只核到标题与副标题（来自搜索引擎收录的页面摘要）；正文细节未能打开。"
 },
 {
  "t": "The SaaS Retention Report: The AI churn wave",
  "org": "ChartMogul",
  "au": "ChartMogul 驻场分析师",
  "date": "2025-12-09",
  "url": "https://chartmogul.com/reports/saas-retention-the-ai-churn-wave/",
  "kind": "数据",
  "gist": "3,500 家软件公司（约 200 家 AI 原生）：AI 原生公司毛收入留存 GRR 中位 40%、净留存 NRR 48%，对比 B2B SaaS NRR 82%；月费 >$250 的 AI 产品 GRR 70%/NRR 85%，<$50 的只有 23%/32%；GRR 年内从 27% 升到 40%。",
  "topic": "留存",
  "co": [],
  "stance": "唱反调",
  "top": true,
  "cav": "样本由 AI 抓取官网分类，AI 原生仅约 200 家；只含 ARR ≥ $25 万的公司，且按定价档位分化很大。"
 },
 {
  "t": "The ICONIQ Pacesetter Index: The AI-Era Benchmark for Growth, Unit Economics, and Operating Leverage",
  "org": "ICONIQ",
  "au": "ICONIQ Growth",
  "date": "2026-09-17",
  "url": "https://www.iconiq.com/growth/insights/the-iconiq-pacesetter-index",
  "kind": "数据",
  "gist": "Pacesetter（三年增速前四分之一）增速是大盘的 3-5 倍，有的越大越快；因算力成本高，毛利往往偏低；净美元留存是衡量采用是否加深的关键。",
  "topic": "增长速度",
  "co": [],
  "stance": "乐观",
  "top": false,
  "cav": "官方页没有给出 $1M 到 $100M 的具体年数；对象是被挑出来的最强一档，不是平均值。"
 },
 {
  "t": "What's Truly 'Great' Now in B2B + AI Per ICONIQ? 115% Growth at $100M+, 55% Gross Margins, and $655K in Revenue Per Employee",
  "org": "SaaStr",
  "au": "Jason Lemkin",
  "date": "2026-09-17",
  "url": "https://www.saastr.com/whats-truly-great-now-in-b2b-ai-per-iconiq-115-growth-at-100m-55-gross-margins-and-655k-in-revenue-per-employee/",
  "kind": "博客",
  "gist": "解读 ICONIQ Pacesetter：$100M+ ARR 的中位增速 115%（前四分之一 165%）；毛利中位 <$10M 为 55%，$25M-$100M 升到 80%，$100M+ 为 75%；NRR 105%→130%→115%。",
  "topic": "毛利",
  "co": [],
  "stance": "乐观",
  "top": false,
  "cav": "SaaStr 页面无明确日期，取 ICONIQ 原文日期；数据是 2024 至 2026 Q2 的公开软件公司加 ICONIQ 自家组合，且小基数会放大早期增速。"
 },
 {
  "t": "2026 State of AI Report: The Builder's Economy",
  "org": "ICONIQ",
  "au": "ICONIQ（约 300 名软件高管问卷）",
  "date": "2026-07-01",
  "url": "https://www.iconiq.com/growth/reports/state-of-ai-2026",
  "kind": "报告",
  "gist": "AI 产品毛利 2025 年 45%、预计 2026 年 53%、2027 年 59%；按用量计价从 35% 升到 42%，按结果计价从 18% 升到 23%；平均每家混用 1.7 种定价。",
  "topic": "毛利",
  "co": [],
  "stance": "谨慎",
  "top": true,
  "cav": "发布日期官方页只写 July 2026（日取 1 号占位）；调查对象是 ICONIQ 网络里的软件公司，毛利为自报。"
 },
 {
  "t": "The Builder's Economy: 10 Metrics from ICONIQ's Newest 2026 State of AI Report",
  "org": "SaaStr",
  "au": "Jason Lemkin",
  "date": "2026-07-01",
  "url": "https://www.saastr.com/the-builders-economy-10-metrics-from-iconiqs-newest-2026-state-of-ai-report/",
  "kind": "博客",
  "gist": "ICONIQ 报告十条：AI 收入占比 32%（2025）→42%→53%（2027）；84% 的按用量计价公司把部分 token 成本转嫁给客户，只有 15% 全部自己吞。",
  "topic": "定价",
  "co": [],
  "stance": "谨慎",
  "top": false,
  "cav": "SaaStr 对 ICONIQ 的二手解读，日期同 ICONIQ 原报告（月份）。"
 },
 {
  "t": "VCs, Superhuman Mail founder share red flags about product-market fit",
  "org": "TechCrunch",
  "au": "Dominic-Madori Davis",
  "date": "2026-10-23",
  "url": "https://techcrunch.com/2026/10/23/vcs-superhuman-mail-founder-share-red-flags-about-product-market-fit/",
  "kind": "博客",
  "gist": "TechCrunch Disrupt：ICONIQ 的 Anu Bharadwaj 说拿试点容易、从试点到生产才难；有 AI 创始人把短试点客户算进年化；红旗是收入集中在少数客户、客户易切换；Superhuman 创始人只数高参与度的 'delighted ARR'。",
  "topic": "收入口径",
  "co": [
   "Superhuman Mail"
  ],
  "stance": "唱反调",
  "top": false,
  "cav": "是会议讨论的转述，没有统计数据；ICONIQ 本身也是 AI 公司投资方。"
 },
 {
  "t": "AI's $1.5T Question",
  "org": "Sequoia Capital（个人 Substack）",
  "au": "David Cahn",
  "date": "2026-07-08",
  "url": "https://dcahn.substack.com/p/ais-15t-question",
  "kind": "博客",
  "gist": "按 Nvidia 数据中心年化收入翻倍再翻倍，2026 下半年 AI 需要约 $1.5T 终端收入才能覆盖一年的数据中心资本开支（2026 年超大规模厂商资本开支约 $7500 亿）；自 ChatGPT 发布累计收入缺口约 $3T；Anthropic 传闻 ARR $600 亿。",
  "topic": "市场规模",
  "co": [
   "Anthropic",
   "OpenAI",
   "Nvidia"
  ],
  "stance": "谨慎",
  "top": false,
  "cav": "个人 Substack，不是 Sequoia 官方；没给全生态收入总数，$1.5T 是推算需求，不是实际收入；Anthropic $600 亿为传闻。"
 },
 {
  "t": "AI's $600B Question",
  "org": "Sequoia Capital",
  "au": "David Cahn",
  "date": "2024-06-20",
  "url": "https://www.sequoiacap.com/article/ais-600b-question/",
  "kind": "博客",
  "gist": "把 Nvidia 年化数据中心收入翻倍两次（覆盖总资本开支和终端 50% 毛利），得出 AI 需要约 $6000 亿年收入；缺口从 $1250 亿升到约 $5000 亿；OpenAI 年化收入约 $34 亿。",
  "topic": "市场规模",
  "co": [
   "OpenAI",
   "Nvidia"
  ],
  "stance": "谨慎",
  "top": false,
  "cav": "2024 年旧文，后续数字已被更新版本取代；推算框架，不是实测收入。"
 },
 {
  "t": "Random thoughts while gazing at the misty AI Frontier",
  "org": "Elad Gil（个人博客）",
  "au": "Elad Gil",
  "date": "2026-04-20",
  "url": "https://blog.eladgil.com/p/random-thoughts-while-gazing-at-the",
  "kind": "博客",
  "gist": "OpenAI、Anthropic 各约 $300 亿年化，若年底都到 $1000 亿，AI 年化收入约占美国 GDP 的 1%；但一部分公司靠补贴推理获客，收入可能建立在补贴算力上；多数 AI 公司应在 12-18 个月内考虑退出。",
  "topic": "增长速度",
  "co": [
   "OpenAI",
   "Anthropic"
  ],
  "stance": "谨慎",
  "top": false,
  "cav": "文章是观点加传闻数字（'rumored'），不是披露口径；BI 对其有转述报道。"
 },
 {
  "t": "The Leonis AI 100: Benchmarking the Most Important AI Startups of 2025",
  "org": "Leonis Capital",
  "au": "Leonis Capital 团队",
  "date": "2025-11-05",
  "url": "https://www.leoniscap.com/research/the-leonis-ai-100",
  "kind": "报告",
  "gist": "索引 1 万多家 AI 初创选出最快增长的 100 家：很多公司 18 个月内到 $10M ARR；Midjourney 约 40 人 ARR 约 $2 亿（人均约 $500 万），Lovable 约 45 人 ARR 约 $1 亿；估算 AI 公司人均收入是顶级 SaaS 的 3-10 倍。",
  "topic": "增长速度",
  "co": [
   "Cursor",
   "Midjourney",
   "Lovable"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "收入多是 Sacra、ARR Club 等的估算，报告自己说可能有误差；数据截至 2025-09。"
 },
 {
  "t": "The high costs and thin margins threatening AI coding startups",
  "org": "TechCrunch",
  "au": "Marina Temkin",
  "date": "2025-08-07",
  "url": "https://techcrunch.com/2025/08/07/the-high-costs-and-thin-margins-threatening-ai-coding-startups/",
  "kind": "博客",
  "gist": "AI 编程助手可能亏钱：Windsurf 毛利'非常负'，推理成本高、被迫上最新模型、竞争激烈；Cursor（Anysphere）6 月 $5 亿 ARR 也面临类似压力；Anthropic、OpenAI 又是直接对手。",
  "topic": "毛利",
  "co": [
   "Windsurf",
   "Cursor",
   "Anthropic",
   "OpenAI"
  ],
  "stance": "唱反调",
  "top": false,
  "cav": "2025 年 8 月旧数据，之后各家毛利有变；多为匿名知情人转述。"
 },
 {
  "t": "The summer of vibe coding is over — How reasoning models broke the economics of AI code generation",
  "org": "CB Insights",
  "au": "CB Insights Research",
  "date": "2025-08-28",
  "url": "https://www.cbinsights.com/research/reasoning-effect-on-ai-code-generation/",
  "kind": "报告",
  "gist": "推理模型让输出 token 量约增 20 倍，推理成本最多涨 20 倍；原本 80-90% 的企业合同毛利可能被压成深度为负；Anthropic 曾因部分用户在 $200 套餐上月耗算力超 $1 万而对 Claude Code 限流。",
  "topic": "毛利",
  "co": [
   "Cursor",
   "Lovable",
   "Replit",
   "Anthropic",
   "OpenAI"
  ],
  "stance": "唱反调",
  "top": false,
  "cav": "2025 年 8 月报告，后续模型降价与自研模型可能改变结论；价格提高幅度为报告的归纳。"
 },
 {
  "t": "Trends – Artificial Intelligence (May '25)",
  "org": "BOND（Mary Meeker 等）",
  "au": "Mary Meeker, Jay Simons, Daegwon Chae, Alexander Krey",
  "date": "2025-05-30",
  "url": "https://www.bondcap.com/reports/tai/",
  "kind": "报告",
  "gist": "339 页 AI 趋势报告：ChatGPT 到 2025 年 4 月约 8 亿用户；引 The Information 等估算 OpenAI 收入从 2022 年的 0 到 2024 年约 $37 亿；基础模型公司年化收入对比累计融资差距很大（OpenAI 约 $92 亿 vs 融资约 $639 亿，2025 年 4-5 月估算）。",
  "topic": "市场规模",
  "co": [
   "OpenAI",
   "Anthropic",
   "xAI",
   "Perplexity"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "官方页是 PDF 阅读器、正文无法直接读取；数字取自 BOND 官方 PDF 搜索摘录与媒体转述；收入多为 The Information、Bloomberg 等的估算，报告自己标注为 estimates。"
 },
 {
  "t": "State of AI Report 2025 — AI adoption becomes visible in spending and revenue",
  "org": "Air Street Capital",
  "au": "Nathan Benaich",
  "date": "2025-10-09",
  "url": "https://www.stateof.ai/2025-explore/business",
  "kind": "报告",
  "gist": "16 家精选 AI-first 公司 2025 年 8 月年化收入合计 $185 亿；另有 a16z 数据称企业与消费 AI 应用第一年中位 ARR 分别超过 $200 万和 $400 万。",
  "topic": "市场规模",
  "co": [
   "Midjourney",
   "Cursor",
   "Lovable"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "官方说明这是精选 16 家的年化 run rate，不是全行业收入，也不是完整财年；网页版为 2026-10-10 制作的历史快照。"
 },
 {
  "t": "Anthropic and OpenAI's Share of AI Startup Revenues Rises to 89%",
  "org": "The Information",
  "au": "The Information",
  "date": "2026-05-17",
  "url": "https://feature.theinformation.com/articles/anthropic-openais-share-ai-startup-revenues-rises-89",
  "kind": "数据",
  "gist": "The Information 生成式 AI 数据库里 34 家头部 AI 初创年化收入合计近 $800 亿（月均 $66 亿），半年增 112%；Anthropic 与 OpenAI 合占 89%（半年前约 84.5%）。",
  "topic": "榜单",
  "co": [
   "Anthropic",
   "OpenAI",
   "Cursor",
   "Cognition",
   "ElevenLabs",
   "Perplexity"
  ],
  "stance": "谨慎",
  "top": false,
  "cav": "付费墙；摘要与多家媒体转述互相印证。两家数字可能偏高：Anthropic 对 Amazon、Google 渠道按总额确认，OpenAI 要给微软分 20%；应用层公司还要向两家付模型费，存在重复计算。"
 },
 {
  "t": "How Sales-Led B2B AI Companies Are Reaching $100M in Revenue So Fast",
  "org": "The Recursive",
  "au": "Seva Ustinov（Plurio 创始人）",
  "date": "2026-08-25",
  "url": "https://www.therecursive.com/how-sales-led-b2b-ai-companies-are-reaching-100m-in-revenue-so-fast/",
  "kind": "博客",
  "gist": "分析 17 家销售驱动型 B2B AI 公司，部分 2-3 年到 $100M ARR：Glean 约 3 年，Harvey 36 个月 $1.95 亿，Abridge 约 30 个月从 $600 万到 $1 亿以上。",
  "topic": "增长速度",
  "co": [
   "Glean",
   "Harvey",
   "Abridge",
   "Sierra"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "作者自己做 AI 创业；17 家公司没点名、抽样方法没写；Harvey 数字文中标为'available data'，未验证。"
 },
 {
  "t": "Inference Is the Most Important Market in Software",
  "org": "Theory Ventures",
  "au": "Tomasz Tunguz",
  "date": "2026-10-05",
  "url": "https://tomtunguz.com/inference-is-the-most-important-market-in-software/",
  "kind": "博客",
  "gist": "AI 推理市场预计 2027 年约 $3500 亿，超过约 $1610 亿的数据库市场；每个应用都会变成推理的转售商；二手摘要称传统软件毛利约 72%，推理占收入过半会把它拉向 30-50%。",
  "topic": "毛利",
  "co": [],
  "stance": "谨慎",
  "top": false,
  "cav": "官方页只核到标题和开头摘录；72%→30-50% 的毛利推论来自 Data Driven VC 等二手摘要，未在原文核对。"
 },
 {
  "t": "Global AI market could hit $6 trillion annually by 2031 (Global Technology Report 2026)",
  "org": "Bain & Company",
  "au": "David Crawford、Anne Hoecker（引语）",
  "date": "2026-09-29",
  "url": "https://www.bain.com/about/media-center/press-releases/2026/global-ai-market-could-hit-%246-trillion-annually-by-2031-through-unlocking-value-and-innovation--bain--cos-7th-global-technology-report/",
  "kind": "报告",
  "gist": "按资本开支约占行业收入 25% 估算，撑起 2031 年算力建设需要 AI 行业年收入约 $6 万亿；现有消费与企业 AI 应用到 2031 年合计约 $1.2–1.8 万亿，缺口约 $4.2 万亿，要靠搜索替代加广告、自动驾驶、实体 AI 和尚不存在的新产品补。报告称基础设施建设「远超需求曲线」。",
  "topic": "算力投入",
  "co": [
   "OpenAI",
   "Anthropic"
  ],
  "stance": "谨慎",
  "top": true,
  "cav": "$6 万亿是 Bain 按 25% 资本开支占比倒推的假设值，不是预测；2031 年的 $1.2–1.8 万亿也是区间估计。已读新闻稿原文；完整报告正文未读。"
 },
 {
  "t": "$2 trillion in new revenue needed to fund AI's scaling trend (Global Technology Report 2025)",
  "org": "Bain & Company",
  "au": "David Crawford、Anne Hoecker（引语）",
  "date": "2025-09-23",
  "url": "https://www.bain.com/about/media-center/press-releases/20252/$2-trillion-in-new-revenue-needed-to-fund-ais-scaling-trend---bain--companys-6th-annual-global-technology-report/",
  "kind": "报告",
  "gist": "到 2030 年算力需求约 200GW，需要约 $5000 亿年资本开支、$2 万亿年新增收入；即使把销售、营销、客服、研发里 AI 省下的钱全部再投入，仍缺约 $8000 亿。领先企业靠 AI 拿到 10%–25% 的 EBITDA 提升，但多数公司仍停在试验阶段。",
  "topic": "算力投入",
  "co": [],
  "stance": "谨慎",
  "top": false,
  "cav": "这是 2025 版；2026 版把口径改成 2031 年 $6 万亿，两版口径不同，不要直接比较。缺口 $8000 亿的前提是美国企业把本地 IT 预算全部迁到云。"
 },
 {
  "t": "An update on AI's most important number",
  "org": "Epoch AI（Gradient Updates）",
  "au": "Josh You、Lynette Bye",
  "date": "2026-08-27",
  "url": "https://epochai.substack.com/p/an-update-on-ais-most-important-number",
  "kind": "文章",
  "gist": "OpenAI 年化收入一年内从约 $130 亿涨到 $400 亿以上；Anthropic 2025 年从 $10 亿到 $90 亿，2026 年 Q1 单季翻了三倍多，据报道 7 月底约 $650 亿。两家合计约 $1050 亿，较年初约 $300 亿涨 3.5 倍，整个生成式 AI 市场约 $2000 亿。作者提醒 2026 年内增速可能已在放缓。",
  "topic": "增长速度",
  "co": [
   "OpenAI",
   "Anthropic"
  ],
  "stance": "乐观",
  "top": true,
  "cav": "全是年化运行率，多数来自媒体报道，作者自己说不宜字面理解；Anthropic 按总额记账、OpenAI 经云厂商的部分只记自己那份，两家口径不一致。该栏目是观点性通讯，不代表 Epoch 官方立场。"
 },
 {
  "t": "Anthropic could surpass OpenAI in annualized revenue by mid-2026",
  "org": "Epoch AI",
  "au": "Luke Emberson、Yafah Edelman",
  "date": "2026-02-19",
  "url": "https://epoch.ai/data-insights/anthropic-openai-revenue",
  "kind": "数据洞察",
  "gist": "Anthropic 自 $10 亿年化起增速约 10 倍/年（2025 年 7 月后放缓到约 7 倍），OpenAI 约 3.4 倍/年；按趋势外推，两条线在 2026 年 8 月前后相交，约 $430 亿。",
  "topic": "增长速度",
  "co": [
   "Anthropic",
   "OpenAI"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "外推结果，90% 置信区间从 2026 年 2 月到 2027 年 4 月；事后看（见 Epoch 8 月更新）Anthropic 实际更快。"
 },
 {
  "t": "Introducing the AI Companies Data Hub",
  "org": "Epoch AI",
  "au": "Epoch AI 团队",
  "date": "2025-09-30",
  "url": "https://epoch.ai/blog/introducing-the-ai-companies-data-hub",
  "kind": "文章",
  "gist": "Epoch 推出前沿 AI 公司的收入、融资、估值、人数、算力开支数据库。当时数据：OpenAI 与 Anthropic 合计收入自 2024 年初涨约 10 倍；OpenAI 年化 2025 年 8 月 $130 亿（年初 $50 亿），Anthropic 七个月内从 $10 亿到 $50 亿。",
  "topic": "榜单",
  "co": [
   "OpenAI",
   "Anthropic"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "数据库汇总公司披露和媒体报道，口径是年化运行率，不是审计收入；数字已被后续更新取代。"
 },
 {
  "t": "Anthropic and OpenAI earn more revenue per employee than major public tech companies",
  "org": "Epoch AI",
  "au": "Luke Emberson",
  "date": "2026-05-08",
  "url": "https://epoch.ai/data-insights/revenue-per-employee-ai-companies",
  "kind": "数据洞察",
  "gist": "人均年化收入：Anthropic 约 $900 万（$300 亿 / 3328 人），OpenAI 约 $560 万（$250 亿 / 4433 人），高于福布斯全球 2000 强里所有上市科技公司（Nvidia 约 $440–510 万）。",
  "topic": "收入口径",
  "co": [
   "Anthropic",
   "OpenAI"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "分子是年化运行率、分母部分是外推的人数，且两个数据点时间不对齐，作者自己说是近似值；收入与算力成本无关，不代表盈利。"
 },
 {
  "t": "Exclusive: OpenAI Losses Increased Nearly 8X in 2025, With Spending Hitting $34 Billion",
  "org": "Where's Your Ed At",
  "au": "Ed Zitron",
  "date": "2026-06-15",
  "url": "https://www.wheresyoured.at/exclusive-openai-financials/",
  "kind": "文章",
  "gist": "据他看到并称经 FT 独立核实的财务文件：OpenAI 2025 年收入 $130.7 亿（2024 年 $37 亿），总成本 $340 亿，经营亏损 $209.2 亿（2024 年 $87.8 亿）；账面归属净亏损 $385.3 亿。他认为亏损规模和增速让盈利路径难以看清。",
  "topic": "算力投入",
  "co": [
   "OpenAI",
   "Microsoft"
  ],
  "stance": "唱反调",
  "top": false,
  "cav": "作者是长期看空者；文中自己说评论克制、后续才做完整分析。$130.7 亿是确认收入，与 OpenAI 此前对外说的「年化 $200 亿以上」口径不同。我只读了该文，没有独立拿到原始财务文件。"
 },
 {
  "t": "The month Generative AI lost its mojo",
  "org": "Marcus on AI",
  "au": "Gary Marcus",
  "date": "2026-06-26",
  "url": "https://garymarcus.substack.com/p/the-month-generative-ai-lost-its",
  "kind": "文章",
  "gist": "认为收入增长不等于能盈利、也不等于可持续；列出一个月内 Oracle 约 -22%、Cerebras -32%、微软 -10%、英伟达超 -8%；称 Anthropic 若有首个盈利季度，也靠一次性 SpaceX 补贴，补贴可能比利润还大；LLM 在商品化，大规模堆算力可能成为「最大的财务失误之一」。",
  "topic": "算力投入",
  "co": [
   "OpenAI",
   "Anthropic",
   "Nvidia",
   "Oracle"
  ],
  "stance": "唱反调",
  "top": false,
  "cav": "观点文，文中没有自己的收入数据，引用 Noah Smith 的图和 Ed Elson 的数字；Anthropic 盈利与否在该文中也是未证实。"
 },
 {
  "t": "The GenAI Divide: State of AI in Business 2025",
  "org": "MIT NANDA（MIT Media Lab）",
  "au": "Aditya Challapally、Chris Pease、Ramesh Raskar、Pradyumna Chari",
  "date": "2025-08-19",
  "url": "https://virtualizationreview.com/articles/2025/08/19/mit-report-finds-most-ai-business-investments-fail-reveals-genai-divide.aspx",
  "kind": "报告",
  "gist": "在 $300–400 亿企业生成式 AI 投入下，95% 的组织没有可测量的业务回报，只有 5% 的集成试点产生显著价值；自研工具进入生产的约 5%，外部采购合作成功率约为自建的两倍；员工在 90% 以上公司私用个人 AI 工具（影子 AI）。",
  "topic": "收入口径",
  "co": [],
  "stance": "唱反调",
  "top": false,
  "cav": "日期取转述文章发布日（8 月 19 日）；报告本身是 2025 年 7 月发布。链接是 Virtualization Review 对报告的转述，我没打开 MIT 原 PDF。样本为 52 次访谈、153 份问卷、300 多个项目，业界对「95%」口径有争议；它衡量企业试点回报，不是供应商收入。"
 },
 {
  "t": "Cracks in the AI Thesis, Part 3（Ramp AI Index 2026 年 10 月）",
  "org": "Ramp Economics Lab",
  "au": "Ara Kharazian",
  "date": "2026-10-14",
  "url": "https://ramp.com/data/ai-index-oct-2026",
  "kind": "数据月报",
  "gist": "9 月美国企业 AI 付费渗透 55.9%（Anthropic 43.8%、OpenAI 40.6%）；新增客户放缓，作者认为已到 S 曲线顶部；约 80% 的 OpenAI+Anthropic 支出来自 1% 的企业；头部 1% 每人每月 $9568，自 7 月高点回落 3.1%；token 支出 8–9 月几乎不涨，因为降价抵消用量；大企业合同约束不强且常带折扣。",
  "topic": "留存",
  "co": [
   "Anthropic",
   "OpenAI",
   "xAI"
  ],
  "stance": "谨慎",
  "top": false,
  "cav": "样本是 Ramp 客户（偏中小企业、美国），不能代表全市场；是月度更新里的一篇评论，结论是作者个人判断。"
 },
 {
  "t": "Reflections on the State of the Software and AI Market（Redpoint 2026 Market Update）",
  "org": "Redpoint Ventures",
  "au": "Logan Bartlett（与 Adil Bhatia、Lydia Day 合写）",
  "date": "2026-03-26",
  "url": "https://x.com/loganbartlett/status/2037638091671035994",
  "kind": "文章",
  "gist": "反驳「互联网泡沫 2.0」：2000 年光纤利用率峰值低于 3%、收入近零，而 OpenAI、Anthropic 各自年化收入超 $200 亿，新数据中心容量超 90% 开工前已被预订。CIO 调查：45% 的 AI 预算来自现有软件预算而非新增；高增长 B/C 轮私募中位数 61x ARR、ARR 增速中位数 640%，按增长调整后比公开市场便宜 86%。",
  "topic": "算力投入",
  "co": [
   "OpenAI",
   "Anthropic"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "Redpoint 是 AI 投资人，立场偏多；$200 亿以上是 3 月数据，已过时。我读的是 Logan Bartlett 在 X 上的长文；官方 Market Update PDF（69 页）没打开。"
 },
 {
  "t": "How AI Applications Earn Their Premium",
  "org": "Andreessen Horowitz (a16z)",
  "au": "Tugce Erten",
  "date": "2026-11-02",
  "url": "https://www.a16z.news/p/how-ai-applications-earn-their-premium",
  "kind": "文章",
  "gist": "应用应在「能可靠衡量、归因、守住」的最高价值层定价，而不是按 token；信用点（credits）应对应任务而不是价值指标。毛利要算完整交付成本（检索、模型调用、重试、评测、人工复核），并把失败尝试算进每次成功的成本。威胁：客户用通用模型就能复现应用核心产出时，溢价就站不住。",
  "topic": "定价",
  "co": [
   "Harvey",
   "Decagon",
   "Clay"
  ],
  "stance": "谨慎",
  "top": false,
  "cav": "没有毛利或 ARR 数字，是方法论文；引用的客户成果来自厂商自己发布的案例；日期按页面显示。"
 },
 {
  "t": "What \"Working\" Means in the Era of AI Apps",
  "org": "Andreessen Horowitz (a16z)",
  "au": "Olivia Moore、Marc Andrusko",
  "date": "2025-06-06",
  "url": "https://a16z.com/revenue-benchmarks-ai-apps/",
  "kind": "文章",
  "gist": "AI 原生企业应用第 1 年 ARR 中位数超 $200 万（AI 之前最佳约 $100 万），约 9 个月后 A 轮；消费类中位数 $420 万，约 8 个月后 A 轮；公司自报：Lovable 6 个月 $5000 万，Cursor 第一年 $1 亿。",
  "topic": "增长速度",
  "co": [
   "Lovable",
   "Cursor",
   "Gamma"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "样本是 a16z 看到的公司（幸存者偏差）；示例收入为公司自报；文中无毛利数字，也没给留存百分比。"
 },
 {
  "t": "Retention Is All You Need: A New Gut-Check for Measuring AI Companies' Retention",
  "org": "Andreessen Horowitz (a16z)",
  "au": "Santiago Rodriguez、Alex Immerman",
  "date": "2025-09-10",
  "url": "https://a16z.com/ai-retention-benchmarks/",
  "kind": "文章",
  "gist": "建议把留存和获客的基准月从 M0 改成 M3，以过滤注册后几个月就流失的「AI 游客」；主指标 M12/M3；头部自助/月付 AI 公司的先行指标显示长期净美元留存（NDR）有望超 100%；样本为数十家 ARR 超 $100 万的 AI 公司。",
  "topic": "留存",
  "co": [],
  "stance": "谨慎",
  "top": false,
  "cav": "具体基准数值在图表图片里，我没读到；文中没有声称有公司做到 150% NDR，只是提问。"
 },
 {
  "t": "The State of AI Report（Battery Ventures）",
  "org": "Battery Ventures",
  "au": "Dharmesh Thakker、Danel Dayan、Jason Mendel、Sudhee Chilappagari",
  "date": "2025-12-18",
  "url": "https://www.slideshare.net/slideshow/the-state-of-ai-report-battery-ventures/284723367",
  "kind": "报告",
  "gist": "毛利对比：应用层 SaaS 80%+，AI 应用 0–30%；模型推理 30–60%；云基础设施约 60%；芯片 45%→75%。AI 原生目标：ARR 增速 5–10 倍（SaaS 2–3 倍）、毛利 20–40%、毛利留存 80%+，并指出「流失意味着还在试验」；定价从座位费转向「座位费加按动作/按结果付费」。",
  "topic": "毛利",
  "co": [
   "Sierra",
   "Baseten"
  ],
  "stance": "谨慎",
  "top": false,
  "cav": "数字是 Battery 自己的估计，不是公司审计数据；SlideShare 转载页，Battery 官网原页未核；页面末尾 2668 字符未读。"
 },
 {
  "t": "AI inference is becoming a vast business（State of AI Report 2026 商业章）",
  "org": "Air Street Capital",
  "au": "Nathan Benaich",
  "date": "2026-10-08",
  "url": "https://www.stateof.ai/2026-explore/business",
  "kind": "报告",
  "gist": "OpenAI + Anthropic 合计年化收入从年初约 $300 亿到夏末约 $1050 亿；Standard Metrics 二季度初步样本：年化收入 $100 万–$2000 万的 AI 原生公司增速（第 75 百分位）256%，加 AI 功能的现有公司 90%；超 $2000 万档分别为 172% 和 53%。训练数据供应商：Mercor 约 $20 亿、Handshake AI 近 $10 亿（年化，2026 年）。",
  "topic": "增长速度",
  "co": [
   "OpenAI",
   "Anthropic",
   "Mercor",
   "Handshake AI",
   "micro1"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "文中明说各公司口径与日期不同，不能加总；两家年化收入包含订阅、编程产品和 API，不等于推理收入。"
 },
 {
  "t": "AI Harness' ARR Multiples",
  "org": "Theory Ventures",
  "au": "Tomasz Tunguz",
  "date": "2026-08-09",
  "url": "https://tomtunguz.com/ai-harness-arr-multiples",
  "kind": "文章",
  "gist": "Harvey、Legora、Sierra 九个月内相继突破 $1 亿 ARR，估值约 50 倍、56 倍、100 倍 ARR；规模化公司的估值倍数在 25–125 倍，近期更高；增速解释不了倍数差距，品类位置更重要。Ramp（$14 亿）与 Decagon（$3500 万）作对照。",
  "topic": "收入口径",
  "co": [
   "Harvey",
   "Legora",
   "Sierra",
   "Ramp",
   "Decagon"
  ],
  "stance": "谨慎",
  "top": false,
  "cav": "部分收入是估算（Decagon 的 129 倍是上界）；作者自己说看趋势比看精确值更重要。"
 },
 {
  "t": "2026: The State of Consumer AI",
  "org": "Menlo Ventures",
  "au": "Shawn Carolan、Amy Wu Martin、Sam Borja",
  "date": "2026-09-16",
  "url": "https://menlovc.com/perspective/2026-the-state-of-consumer-ai/",
  "kind": "报告",
  "gist": "消费者 AI 支出 2026 年估算约 $400 亿，是上年 $120 亿的 3 倍多；美国成年人使用率 64%（上年 61%），采用几乎不涨但付费涨；55% 的用户至少付费一款（约 9000 万人）；占付费者 14% 的月花 $100+ 用户贡献 60% 的支出；雇主/学校付费用户月均 $95，自费用户 $55。",
  "topic": "市场规模",
  "co": [
   "OpenAI"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "基于 Morning Consult 对 5067 名美国成年人的 7 月调查，是自报；报告没有给直接留存率；文中「上年约 3% 的用户付费」与「55% 付费」口径看起来相差很大，引用时需回原文核对。"
 },
 {
  "t": "Goldman Sachs 测算：AI 算力资本开支回报需要多少收入",
  "org": "Goldman Sachs Research（经华尔街见闻 / 富途转述）",
  "au": "Eric Sheridan 团队",
  "date": "2026-09-25",
  "url": "https://news.futunn.com/en/post/1000215519/as-ai-capital-spending-surges-toward-1-7-trillion-what",
  "kind": "券商研报",
  "gist": "六大云厂商 2026–27 年 AI 算力资本开支约 $1.73 万亿，要得到 15% ROIC，2028–2030 年需累计收入约 $1.42 万亿（每 GW 每年约 $116 亿）；ROIC 目标 0%–30% 对应 $9080 亿–$1.89 万亿。AWS、Azure、Google Cloud 在手订单约 $1.69 万亿（同比 +152%）；2028–2030 年资本开支再达约 $4.14 万亿。",
  "topic": "算力投入",
  "co": [
   "Microsoft",
   "Alphabet",
   "Amazon",
   "Meta",
   "Oracle"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "我没拿到 Goldman 原报告，链接是华尔街见闻稿经富途的英文转载；高盛自己说框架不含约 $3 万亿表外承诺，且算力价格下行压力未建模。「六大」里包含 SpaceX，系转述所列。"
 },
 {
  "t": "The state of the AI economy",
  "org": "Exponential View",
  "au": "Azeem Azhar",
  "date": "2026-06-25",
  "url": "https://www.exponentialview.co/p/the-state-of-the-ai-economy",
  "kind": "报告",
  "gist": "自下而上、去重后（不含中国）生成式 AI 过去 12 个月收入约 $1100 亿，最近一个月折年约 $1750 亿；2026 年 Q1 季度收入约 $250 亿，连续第二季超过同期芯片与数据中心折旧（约 $210 亿）；AI 收入增速约为互联网、移动浪潮的 3 倍；但折旧仍吃掉收入的三分之二以上，余量很窄。",
  "topic": "收入口径",
  "co": [
   "OpenAI",
   "Anthropic",
   "Cursor",
   "ElevenLabs"
  ],
  "stance": "谨慎",
  "top": true,
  "cav": "摘要段与折旧细节来自作者页面和 Bloomberg/华尔街见闻转述；私营公司数据是建模推算，不是披露值；不含中国市场，也不含推荐系统等内部 AI 增益。"
 },
 {
  "t": "AI in 2026: A Tale of Two AIs",
  "org": "Sequoia Capital",
  "au": "David Cahn",
  "date": "2025-12-03",
  "url": "https://sequoiacap.com/article/ai-in-2026-the-tale-of-two-ais",
  "kind": "文章",
  "gist": "AI 终端收入「几百亿美元/年」，对比未来五年数据中心与能源投资「万亿美元」；两个杀手级应用是编程和 ChatGPT，今年各自逼近或跨过两位数十亿收入；还有近十几家初创在冲 $1 亿；大企业自研 AI 受挫导致疲劳；预测 2026 年会出现「$0 到 $10 亿」俱乐部。",
  "topic": "算力投入",
  "co": [
   "OpenAI",
   "Anthropic"
  ],
  "stance": "谨慎",
  "top": false,
  "cav": "作者估计与预测，不是审计数据；页面元数据日期与署名日期冲突，取署名日 2025-12-03；这是 Sequoia $6000 亿问题的后续观点（2026 年更新版未找到）。"
 },
 {
  "t": "Anthropic 3Q26 Profit Over $1B: The Anthropic IPO Financials Sneak Peak",
  "org": "SemiAnalysis",
  "au": "Joey Brookhart、Crystal Huang、Dylan Patel",
  "date": "2026-07-08",
  "url": "https://newsletter.semianalysis.com/p/anthropic-3q26-profit-over-1b-the",
  "kind": "文章",
  "gist": "标题判断 Anthropic 2026 年 Q3 利润超 $10 亿；开头称 Anthropic 与 OpenAI 合计约 $1000 亿 ARR，Anthropic 以更高毛利、定价权和盈利方式领先，是「首个 $6 万亿公司」的基准情形之一；数据来自其 Tokenomics 模型按 SKU、档位、客户类型自下而上建模。",
  "topic": "毛利",
  "co": [
   "Anthropic",
   "OpenAI"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "正文付费墙，我只读到开头；Anthropic 当时只是保密递交 IPO，无公开财务，全部数字是 SemiAnalysis 模型推算。10 月 S-1 公开后的二手解读（推理毛利 85%、Q2 收入同比 +1368% 等）我没能打开原文核对，未收入。"
 },
 {
  "t": "Head of Growth (Anthropic): \"Claude is growing itself at this point\"",
  "org": "Lenny's Newsletter / Lenny's Podcast",
  "au": "Lenny Rachitsky（嘉宾 Amol Avasare，Anthropic 增长负责人）",
  "date": "2026-04-05",
  "url": "https://www.lennysnewsletter.com/p/anthropics-1b-to-19b-growth-run",
  "kind": "播客",
  "gist": "节目页称 Anthropic ARR 14 个月内从 $10 亿涨到超 $190 亿；Lenny 在 X 上补充：仅 2 月就新增 $60 亿 ARR，而 Palantir、Atlassian 用 15–20 年才到约 $50 亿；增长团队 70/30 押大注，激活被称为 AI 最高杠杆的增长问题。",
  "topic": "增长速度",
  "co": [
   "Anthropic"
  ],
  "stance": "乐观",
  "top": false,
  "cav": "正文付费，我只核到公开简介；$190 亿是 4 月时点数字，已被后续数据（$300 亿、$650 亿等）超过；$60 亿/2 月出自 Lenny 的 X 帖，未在正文核对。"
 }
]