Podcast
Marisa Wu · AI Strategy
Hosted by Marisa Wu · AI Strategy · ZH
Marisa Wu is the China Insider who decodes the global AI power game. With 18 years inside China's tech ecosystem (primarily Alibaba), an EMBA from the University of Hong Kong, and a network of 2,900+ frontline AI engineers across China's top tech companies, she brings a perspective that most Western analysts simply cannot access. She analyzes the AI strategies of Google, Microsoft, OpenAI, Anthropic, Meta, Amazon, Nvidia, and Apple through the lens of capital flows and infrastructure control — cutting through the noise to reveal what's really happening beneath the surface.
ngmarisa.substack.com
ngmarisa.substack.com
33episodes
Episodes
Newest firstAll episodes
如果一个AI方案听起来让你觉得"公司要脱胎换骨"，那它大概率会让你"伤筋动骨"
May 1200:03:02Tap to summarize大多数老板不是不知道AI能做什么，而是不知道从哪里开始不会死。这篇文章不谈模型参数，不谈算力成本，只谈一件事：为什么你身边没有人能用你的语言帮你做这个决定，以及真正的AI落地应该是什么感觉——第二天上班，大家觉得活儿轻了，但谁也没觉得天塌了 This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit ngmarisa.substack.com
Transcribe →
Where Is the Line? When AI Search Becomes Advertising Disguised as Knowledge
May 600:03:48Tap to summarizeSomething is happening quietly, and almost nobody is talking about it clearly. In the age of AI search, advertising has found a new disguise — it's showing up dressed as knowledge. When you ask Doubao, Kimi, or DeepSeek,openai "which is better?", the "objective analysis" AI delivers might be content that was deliberately engineered by a brand. This isn't a conspiracy theory. It's the result of a strategy called GEO — Generative Engine Optimization — being systematically abused. Advertising is a drink. Knowledge is clean water. The problem is that someone has started pouring drinks into clean water bottles. And when you drink it, you can't tell the difference. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit ngmarisa.substack.com
Transcribe →
GEO的边界在哪里：当知识变成新型广告，我们如何分辨？
May 600:02:54Tap to summarize有一件事正在悄悄发生，而且几乎没有人在谈论它。AI搜索时代，广告找到了一种新的伪装方式——它开始穿着知识的外衣出现在你面前。当你问豆包、Kimi、DeepSeek"哪个更好"，AI给出的那个"客观分析"，背后可能是一家品牌精心工程化的内容。这不是阴谋论，这是一个叫GEO（生成式引擎优化）的策略被大规模滥用的结果。广告是饮料，知识是纯净水。问题是，现在有人开始往纯净水的瓶子里装饮料。你喝的时候，看不出区别。 This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit ngmarisa.substack.com
Transcribe →
The $40 Billion Illusion: Why Google's "Bet" on Anthropic is Actually a Rent Collection
Apr 2800:04:14Tap to summarizeOn April 24th, Google announced an investment of up to $40 billion in Anthropic — $10 billion in immediate cash, with $30 billion tied to performance milestones. Amazon followed with $25 billion. But read the contract terms: Anthropic will use this capital to purchase compute from Google Cloud and AWS. The money flows back. Google invests with the left hand, collects cloud revenue with the right. More absurd: Google is simultaneously building Gemini, Claude's direct competitor. This isn't a bet. It's the most expensive customer acquisition strategy in history. Whatever model wins the AI race, the compute has to run through Google. Models are tenants. Infrastructure is the landlord. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit ngmarisa.substack.com
Transcribe →
硅谷最贵"回旋镖"：Google砸下400亿，钱却根本没离开自家银行？
Apr 2800:03:50Tap to summarize4月24日，Google宣布向Anthropic投资最高$400亿，先付$100亿现金，剩余$300亿绑定里程碑。同期Amazon追加$250亿。但合同条款揭示了这笔交易真正的逻辑：Anthropic拿到钱后，要用来购买Google Cloud和AWS的算力——巨头左手投钱，右手收租。更荒诞的是，Google同时在开发Gemini，Claude的直接竞争对手。这不是押注，这是锁定AI时代最大付费客户的最贵方式。无论哪个模型赢，算力都必须从Google这里过。模型是租客，算力才是地产。 This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit ngmarisa.substack.com
Transcribe →
Carbon Out, Silicon In: How Big Tech is Swapping 16,000 Roles for AI Infrastructure
Apr 2500:04:21Tap to summarizeMicrosoft launched its first-ever large-scale buyout program in 51 years, covering roughly 7% of US employees (~8,750 people). Meta simultaneously confirmed layoffs of 8,000 employees (10%) effective May 20th. Both companies explicitly tied these cuts to AI infrastructure spending. The Big Four's 2026 capex will exceed $630 billion — up ~70% from 2025. This isn't an industry contracting. It's an industry swapping budgets — moving headcount costs into GPU racks and liquid-cooled compute. Salary lines are exiting. Power bills and silicon are taking their seat at the table. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit ngmarisa.substack.com
Transcribe →
“岗位换算力”时代开启：微软 Meta 1.6 万人“毕业”，只为给下一代 GPU 腾地方？
Apr 2500:03:11Tap to summarize微软51年历史上首次推出大规模买断计划，覆盖约7%美国员工（约8,750人）；Meta同步确认5月20日起裁员8,000人（10%）。两家公司都把这轮裁员与AI基建支出直接挂钩。四大科技公司2026年Capex合计超$6,300亿，较2025年增长约70%。这不是行业在收缩，而是行业在换血——把人力预算腾出来，换成GPU机架和液冷算力。薪资预算在退场，电力与硅片正在上桌。两种组织的竞争力差距，最终不是效率，是认知速度。 This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit ngmarisa.substack.com
Transcribe →
The $1.75 Trillion Disconnect: Why SpaceX is Hedging the Very Future it Sells
Apr 2300:03:41Tap to summarizeIn public, Musk calls orbital data centers "the ultimate future of AI infrastructure." In SpaceX's IPO filing, the same ambition is described as relying on "unproven technologies" that "may never become commercially viable." Same person. Two completely different audiences. This isn't contradiction — it's the oldest art in capital markets: vision leverage to inflate valuation, legal fortress to isolate liability. The gap between the two is the narrative arbitrage space — and the real boundary of the valuation premium. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit ngmarisa.substack.com
Transcribe →
SpaceX罗生门：马斯克卖给世界一个梦想，却在招股书里藏了一道免责声明
Apr 2300:03:17Tap to summarize马斯克公开演讲里，轨道数据中心是"AI基础设施的终极未来"。SpaceX的IPO招股文件里，同一件事被描述为"依赖未经验证的技术，可能永远无法商业化"。同一个人，公开场合一套，法律文件一套。这不是矛盾，这是资本市场里最古老的一门艺术：愿景杠杆，用来拉升估值；法律堡垒，用来规避责任。两者的落差，就是叙事套利空间——也是估值泡沫的真实边界 This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit ngmarisa.substack.com
Transcribe →
Stop Treating AI Like an Employee. It's Becoming the CEO's Second Brain.
Apr 2200:03:43Tap to summarizeOn April 15th, OpenAI updated its Agents SDK with sandboxing, configurable memory, and self-correction capabilities. Gartner forecasts 40% of enterprise apps will embed AI agents by end of 2026. AI is no longer just executing tasks — it's entering the decision layer. The question is no longer “which employees will AI replace?” It's “how deeply can AI participate in a CEO's judgment?” That's a different conversation entirely. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit ngmarisa.substack.com
Transcribe →