20 亿美元梦碎后的反思：AI 浏览器的唯一生路，被巨头们"守着金山要饭"给耽误了

引言：一场资本游戏与用户信任的彻底崩塌

2026 年 4 月 27 日，中国国家发改委明确否决了美国 Meta 对新加坡 AI 初创公司 Manus 的收购案，一并击碎了这家由中国工程师团队打造、估值高达 20 亿美元的"造富神话"（TechCrunch，Reuters）。四个月之后，从 Meta 中剥离出来的独立公司选择"重回独立运营"，立刻在 2026 年 8 月 11 日发出通告，要求某些区域用户在 8 月 23 日之前完成数据备份或迁出，并配合大规模的强制退款（Bloomberg，Monica 官方说明，Monica 退款条款）。之后又用"半价恢复入会"的促销收割老用户，而重新加入者赫然发现，拆分后的产品连接不稳、频繁报错，核心智能出现史诗级滑坡。

这场闹剧完全扒下了"套壳投机型 AI 创业"的遮羞布。但更值得全行业深思的是：为什么这样一个技术极度基础、毫无核心大脑的"套壳聚合器"，能在过去两年里吸走成千上万的全球付费用户？

答案很直接：它无意中踩中了 AI 时代最需、最直觉的现成生意——实质意义上的 AI 浏览器。而这个首要的全民商机，正被万亿巨头们用"大企业病"和不切实际的"科幻饼"活活耽误。

附录：Monica / Manus 的"造富与幻灭"时间轴

• 2025 年 3 月：北京蝴蝶效应科技（Butterfly Effect）推出 Manus 智能体，凭借高度震撼的 Demo 视频在全球一夜爆火，邀请码在二手市场炒到上万元人民币（Wikipedia，SCMP）。
• 2025 年末：Meta 与已迁册新加坡的 Manus 达成金额超过 20 亿美元的收购协议，前提是绝对剥离中国股权与运营（TechCrunch）。
• 2026 年 1—3 月：中国商务部与国家发改委对涉及前沿 AI 能力的这笔交易启动国家安全和技术出口管制审查，创始人肖弘（Xiao Hong）等被曝出境限制（Reuters/FT）。
• 2026 年 4 月 27 日：国家发改委发布明确禁令，以国家安全审查和技术出口控制为由，依法叫停并要求彻底撤销这笔 20 亿美元的外资收购（TechCrunch，JD Supra 法律分析）。
• 2026 年 5—7 月：Meta 与 Manus 进行"外科手术式"业务与数据割裂，切断所有跨境系统集成与数据共享（Reuters）。
• 2026 年 8 月 11 日：Manus 宣布恢复独立运营，并行官方通告要求部分区域用户在 8 月 23 日前完成数据迁出、备份，并触发大面积强制退款（Bloomberg，Monica 官方公告）。
• 2026 年 8 月底至 9 月初：肖弘核心团队宣布重组完成、"回归独立运营"，接着在美国等地推出"半价促销"以迅速回笼资金，但因底层链路割裂，产品体验完全崩坏（Yahoo Finance/Reuters）。

一、用户的真实痛点：我不需要科幻代理，我只需要"左边读，右边问"

当下的 AI 行业陷入了一种病态的"技术热衷"。OpenAI、Anthropic 等巨头正砸下数十亿重金，去研发能够控制键盘鼠标、替人类操作电脑、做多层多步任务的"AI 代理（Agent）"。

但他们搞错了真实的大众市场。让 AI 无监督地操作电脑，是给极少数技术精英准备的"高端局"，开发门槛显著，短期内无法大规模落地。

实质占领 80% 全民市场、推广成本最小、最便捷被大众接受的真命题，事实上朴素得不可思议——屏幕左边与右边栏的绝对连接。

用户不需要 AI 跨软件黑进系统，普通人每天最核心的场景是：

• 屏幕左边放着一篇看不懂的科研论文、学校网课，以及繁琐的网页代码；
• 屏幕右边有一个听话、懂内容的 AI 侧边栏；
• 用户在右边问任何问题，AI 都能基于左边的全面上下文即时解答、推导公式、立即把程序编好、把研究报告扩展写出来，继而一键推送到桌面。

假若做不到这一点，右边栏做得再花哨、塞进 50 种奇奇怪怪的绘图或改写功能，也是毫无意义的摆设。如果用户卡壳了，最后还不得不自己去手动"复制、粘贴"到另一个 AI 对话框，那这个边栏的存在就是"脱裤子放屁"，没有任何商业价值。一个只能总结用户"既已看懂、已经读完"的信息的 AI 工具，彻底不配叫智能，它仅是一个奢侈的"电子复读机"。

二、曾经的财富密码：国内人民币成本，美国美金收租

Monica / Manus 们缘由能赚到第一桶金，本质上就是利用了巨头在这一全民刚需上的"缺位"，打了一场跨国成本与汇率的剪刀差战役：

• 成本在内：核心研发和运维全在国内，几十号人，月开销不过百万人民币量级；
• 收租在外：通过 Chrome 插件抢占全球侧边栏的生态位，每个海外用户立即收取 20 美元（约合 140 元人民币）的月租；
• 只要堆起几万个付费用户，每月躺着就是数百万美元的纯利润。

但这种模式毫无技术壁垒。为了不亏本，套壳公司们在后台节俭到了顶点：特意把边栏的自动读取功能做得缺失，逼用户"手动粘贴"，把高昂的 API 流量费（Token 成本）转嫁给用户；他们从未重视老用户，满脑子想的都是把用户当成筹码去套现。当监管一拳砸碎这个美梦，没有核心技术的他们只能瞬间原形毕露、口碑雪崩。

三、巨头的傲慢与退缩：守着金山要饭

最让人感到可笑的是，这场现成的商业大生意，本应是谷歌或苹果闭着眼睛都能赚翻的。

Chrome 浏览器是谷歌的，Gemini AI 也是谷歌的，都在同一个谷歌账户里。假若谷歌把 Gemini 侧边栏做成这种"左边看、右边充分深度联动、立即解决编程和研究工作"的实际 AI 浏览器，全球 20 亿 Chrome 用户里，会有大把人排着队心甘情愿地月付 20 美元。

这是天底下最简单变现、逻辑最顺理成章的 AI 商业化方向。可万亿市值的超级巨头却患上了明显的"大企业病"：

• 法务的万能借口：谷歌总拿版权和侵权当挡箭牌，前怕狼后怕虎，生怕深度读取左边网页会招来官司。当年谷歌做网页搜索、强行抓取全球网站索引时，怎么不见它退缩？谁敢不让谷歌抓取，谁就在互联网上完全消失。如今在 AI 时代，他们却变得扭扭捏捏，用"安全隔离"掩盖功能上的半残。
• 搜索广告的自我阉割：谷歌 70% 以上的利润来自网页搜索广告。它希望你左边看网页不懂就继续在谷歌里搜索、随之点击它的广告。假若右边 Gemini 侧边栏做得太完美——立即帮你把研究做完、代码编好、报告推送到桌面——那用户就不再不需要点击任何网页、再也不用看广告了。让 Gemini 亲自杀死自家的搜索广告帝国，这种自己革自己命的事，谷歌高层没有魄力完全推进。
• 科学家的自负：巨头内部的顶级科学家们看不上这种"写抓取代码、帮普通人导 Word 文件"的苦力活。他们更愿意去赌那些高不可攀的 AGI（通用人工智能）幻想。结果，把眼前最丰饶、最能造血的全民大市场，拱手让给了骗子和投机者。

四、正规军的降维打击：微软与苹果正在终结"套壳时代"

就在这些投机套壳公司忙着炒作概念、摆烂割韭菜的恰逢，实质的行业"正规军"随即开始在这个方向发力，立即进行降维打击：

• 微软 Edge Copilot 的步步紧逼：微软由于没有反垄断的包袱，步子迈得最大。Edge 浏览器内置的 Copilot 侧边栏目前实现了较高程度的左向右联动——用户具备立即对侧边栏说"基于左边的论文给我写个总结并扩展推导"，它不心疼 Token 成本，正在死死粘住切实需要生产力的用户。
• 苹果 Apple Intelligence 的全系统穿透：苹果的系统级 AI 完整打破了浏览器的局限。它拥有的"屏幕感知"（Screen Awareness / Onscreen Awareness）能力，让 AI 具备即时"看懂"用户屏幕上开着的任何软件、网页或 PDF，不需要任何乌七八糟的边栏和插件，直接在系统层提供跨软件的深度连接与扩展能力。
• Perplexity Comet 的原生化尝试：新一代 AI 原生浏览器 Perplexity Comet 一键把"读页面 + 问 AI + 出结果"合成默认工作流，绕开了传统插件的沙箱限制。

正规军的入场和系统级 AI 的普及，正在以摧枯拉朽之势，最终结那个靠"复制粘贴"和"信息总结"就能糊弄用户的低级套壳时代。

结论：天下没人能吃满全场，务实者才是最后的赢家

天底下没有哪家公司能把所有的市场全吃了。随着地缘监管收紧、投机型套壳插件的集体倒闭，市场正在迎来一场全面的清算。

用户已然醒了。那种"只能总结你已经看懂的信息"的电子复读机插件，不久就会被最终卸载。

实际的 AI 浏览器，并非是一个在旁边凑热闹的聊天框，就是深度融入桌面工作流的__超级副驾驶__。它不需要去跨软件操控电脑，它只需要吃透屏幕、解决现实工作、实现真正的左向右融合。

巨头们假若再不放下自负、不把这桩现成的生意干透，那么这个价值万亿的全民市场，迟早会被下一代实际务实、接地气、敢于打破常规的技术颠覆者最终切走。

Introduction: A Capital Game and a Complete Collapse of User Trust

On April 27, 2026, China's National Development and Reform Commission (NDRC) officially vetoed Meta's $2 billion acquisition of Manus, the Singapore-based agentic-AI startup founded by East Asian engineers (TechCrunch, Reuters). Four months later, on August 11, 2026, the entity spun off from the failed Meta deal announced it was resuming self-governing operations — accompanied by a notice forcing users in designated regions to back up or migrate their data by August 23, and triggering mass automated refunds (Bloomberg, Monica official notice, Monica refund policy). Days later came an urgent half-price re-engagement campaign, merely for returning subscribers to find a split product plagued by connectivity errors and an absolute freefall in intelligence.

This farce has stripped away the facade of wrapper-style (套壳) AI startups. But the industry must ask: how did an extension with zero in-house core technology attract hundreds of thousands of international paying users in the first place?

Because it by chance stumbled onto the central urgent, intuitive demand of the AI era — a true AI Browser. And this large-scale-market opportunity is being squandered by trillion-dollar giants paralyzed by enterprise inertia and obsessed with unachievable sci-fi gimmicks.

Appendix: The Manus / Monica Boom-and-Bust Timeline

• March 2025 — Beijing-based Butterfly Effect Technology launches Manus. refined demo videos make it a worldwide overnight sensation; beta invitation codes trade on secondary markets for thousands of dollars (Wikipedia, SCMP).
• Late 2025 — Meta agrees to acquire Manus (previously re-domiciled in Singapore) for more than $2 billion, conditional on a total exit from Chinese ownership and operations (TechCrunch).
• January–March 2026 — China's Ministry of Commerce and NDRC launch national-security and export-control reviews of the frontier-AI transaction; co-founders Xiao Hong and principal scientist Yichao Ji are reportedly placed under exit bans (Reuters/FT).
• April 27, 2026 — NDRC issues an official prohibition order, legally blocking the $2 billion deal on grounds of technology export control and national security, and demanding a full unwind (TechCrunch, JD Supra legal analysis).
• May–July 2026 — Meta executes a "precise" separation, severing all beyond-border system integrations and data-sharing pipelines with Manus (Reuters).
• August 11, 2026 — Manus announces resumption of self-sustaining operations; users in targeted regions are told to migrate or back up data by August 23, and automated forced refunds begin (Bloomberg, Monica notice).
• Late August – first September 2026 — Xiao Hong's core team declares restructuring final and pivots to aggressive half-price promotions in the US to recover cash flow — but with the underlying architecture broken, the product experience collapses (Yahoo Finance / Reuters).

1. The Real User Pain: We Don't Need Sci-Fi Agents; We Need "Read Left, Ask Right"

The AI industry is trapped in a state of maladaptive tech-fetishism. Giants like OpenAI and Anthropic are pouring billions into "AI Agents" that can control keyboards and mice and execute multi-phase-sequential tasks unsupervised.

However they have misjudged the mass market. self-operating desktop control is a "premium-end game" for a small fraction of engineers, with immense development friction and no near-term path to widespread adoption.

The proposition that commands 80% of the mass market — with the smallest customer acquisition cost — is shamefully simple: seamless connection between the left side of the screen and the right sidebar.

Users don't need an AI to hack across their applications. The core scenario is:

• The western side shows a dense academic paper, an online lecture, or complex source code the user can't parse by themselves.
• The opposite side is a dynamic, context-aware AI sidebar.
• The user asks any question on the right, and the AI — entirely absorbing the left-side context — immediately explains, derives formulas, compiles code, or expands the material into a full research report that is pushed directly to the desktop.

If a sidebar can't do this, no polished UI or fifty junk features (meme generators, one-click rewriters) can save it. If the user hits a wall and continuing has to by hand copy-paste text into a separate AI window, the sidebar is redundant. An AI tool that can exclusively summarize information the user has prior read is not intelligent — it's just an expensive digital parrot.

2. The Former Wealth Code: RMB Costs, USD Rents

Startups like Monica made their fortune by exploiting the giants' absence in this precise large-scale-market slot, playing a cross-border cost/currency arbitrage:

• local costs: Core R&D and operations kept in China — a few dozen people, monthly fixed costs of a few million RMB.
• international rent: A Chrome extension that captures the browser-sidebar actual estate; global users charged ~$20/month.
• With a few tens of thousands of paying users, they netted millions of USD in clean profit every month, automatically.

However the model had zero engineering moat. To protect margins, these companies were pathologically tight-fisted on the back end. They purposefully crippled autonomous reading to force "manual pasting," shifting the heavy API token cost onto users. They viewed subscribers as chips to cash out. The moment regulators smashed that dream, their lack of core tech was exposed and their reputation collapsed suddenly.

3. The Arrogance and Timidity of Tech Giants: Starving While Holding a Gold Bowl

The final irony is that this immediate, extremely lucrative business should have been a trivial win for Google or Microsoft.

Chrome and Gemini both belong to Google, tied to the specific unified Google account. If Google turned the Gemini sidebar into a true AI Browser — one that intensely pairs with the left screen and smoothly solves programming and research problems — millions of Chrome users would gladly pay $20/month. However trillion-dollar conglomerates suffer from severe corporate rot:

• The official excuse: Google hides behind copyright and infringement fears, terrified that intensive reading of active web pages will trigger lawsuits. Yet when Google built its search empire by unrelentingly scraping the full global web for indexing, it showed no such cowardice — refuse Google's crawler and you disappeared from the internet. In the AI era, they act timid and use "security sandboxing" to mask half-shoddy features.
• Cannibalization of their internal ad empire: Google earns 70%+ of profit from web-search advertising. It wants users to click into pages with afterward re-search to click ads. If a Gemini sidebar completed the research task and delivered the file to the desktop, users would never need to click a webpage — or an ad — repeatedly. Google's leadership lacks the will to have Gemini kill Google's own ad empire.
• The arrogance of scientists: preeminent researchers inside these giants look down on "grunt work" like writing scraping scripts or exporting Word documents. They prefer chasing the AGI illusion. As a result, the largest productive, cash-generating mass market has been handed to snake-oil salesmen and speculators.

4. The Real Players Enter: Microsoft, Apple, and Perplexity End the Wrapper Era

While hypothetical wrappers were hyping concepts and fleecing users, the real industry titans have started executing a top-down flattening of this market:

• Microsoft Edge Copilot's unwavering aggression — Lacking Google's regulatory baggage, Microsoft has taken the boldest strides. The Copilot sidebar built into Edge currently offers a significant degree of left-to-right integration. Users can command it to summarize and expand on complicated charts or documents accessible on the left. Microsoft doesn't pinch pennies on costly costs; it is vigorously locking in productivity users.
• Apple Intelligence's system-broad penetration — Apple's system-level AI obliterates the browser boundary. Its Onscreen Awareness / Screen Awareness lets the AI see and reason about whatever software, webpage, or PDF is currently on screen — no bloated sidebars, no third-party plugins, deep cross-application synthesis delivered immediately from the OS layer.
• Perplexity Comet's inherent attempt — A new generation of AI-native browsers, prominently Perplexity Comet, folds "read page + ask AI + get result" into the default workflow, bypassing the sandbox limitations of traditional extensions.

The arrival of corporate giants and system-integrated AI is forcefully tearing down the primitive wrapper era.

Conclusion: No One Can Monopolize the Field — the Pragmatists Win

No one company can swallow the total global market. As state-driven regulation tightens and theoretical AI plugins collapse en masse, a brutal market correction is underway.

Consumers have eventually woken up. Digital parrots that can exclusively summarize what you've previously understood will soon be uninstalled entirely.

A genuine AI Browser is not a chat box lingering on the periphery of the screen — it is a super co-pilot intensely woven into the desktop workflow. It doesn't need to automatically control your computer across different apps; it merely needs to comprehend the screen, solve immediate tasks, and achieve genuine left-to-right integration.

If the tech giants refuse to drop their arrogance and execute this ready-made business to its complete limit, this trillion-dollar mass market will certainly be seized by a new generation of pragmatic, grounded, rule-breaking tech disruptors.