新的ABC交易? AI BABA China

一觉起来，外资那边也炸了...昨晚财报 业绩会发酵了一波之后，外资开始大肆吹阿里（甚至上周关于ant的小作文，也有人邮件里提了一嘴...）...

比如下面高盛的这篇，现在昨晚的阿里业绩会后，有些说法开始把“ABC”塑造成“A for AI, B for BABA, C for China”

图片

外资吹起来，比我们还猛 ...慢慢的，可能老外也要FOMO了...

这篇里面问的是下一步能不能看20x PE, 我简单摘抄一些；

图片

昨天另外比较收到关注的是微软CEO Satya Nadella的一个podcase，里面说到的很多东西，和阿里管理层颇为相似（比如云计算是目前最明显的AI货币化模式）。我这里用AI简单翻译一些。

1/ 纳德拉认为云计算公司将是AI的赢家（"在此之后，局面变得有些模糊。有人会说赢家通吃——但我并不这么认为"）。潜台词是微软并不需要依赖模型本身，真正的变现能力在于基础设施。

He views Hyper-scalers as the winner in AI world (‘Then after that, it becomes a little fuzzy. You could say, hey, there is a winner-take-all model- I just don't see it.’).  The subtext is MSFT doesn't need the models but infrastructure is the true monetization.

2/虽然大模型日益同质化，但AI对算力的"饥渴"不会消退——不仅训练需要算力，推理阶段和Agents的指数级应用扩张同样需要。模型仍需运行在需要状态存储 存储 计算的超大规模算力上。你需要在边缘部署"舰队式"分布式数据中心，因为"不可能所有服务都从得克萨斯州提供"。这对微软、亚马逊和Cloudflare（NET）等类型公司构成利好。

LLMs are getting more commoditized but AI remains ‘hungry’ for compute workloads, not just training but test-time compute and then Agents expand use 'exponentially'.  Models will still need to run on hyper-scale compute which requires state storage compute.  You need a 'fleet' of distributed Data Centers at the edge because you 'can't serve everything from Texas'.  Bullish for MSFT, AMZN & NET types.

3/ 最犀利的观点："宣称达到某个AGI里程碑只是无意义的基准测试投机。真正的基准是推动全球增长 10%"。纳德拉认为可能迎来新一轮工业革命，推动发达市场增长 7-10%（或经通胀调整后 5%）。最大赢家将不限于科技公司，而是更广泛采用这种新型AI的行业。

This is the fire quote:  'Claiming some AGI milestone is nonsensical benchmark hacking.  The REAL benchmark is a world growing 10%'.  Nadella sees another Industrial revolution possible that drives 7-10% growth in developed markets or 5% inflation adjusted.  And the big winners will NOT just be Tech companies but broader industries that are using this new abundant intelligence commodity.  Bullish for AI Adopters MSXXAIAD.

4/ 微软明年资本支出将超千亿美元，但相比Sam、马斯克、扎克伯格和孙正义等"大力出奇迹"派，纳德拉的表态仍最理性。此前微软放弃"星际之门"预训练项目已见端倪。他承认超前投资必要性，但也强调最终需供需匹配，若供给端过度炒作却无法转化为客户价值就会"脱轨"。

MSFT spending $100B capex next year but Nadella still sounds most rational on capex relative to Sam Altman, Elon, Zuck and Masa types in the build build build camp.  We saw this with MSFT passing on the Stargate pre-training build so not a surprise.  Nadella says they've invested a lot ahead of demand but at some point you have to match supply and demand and you can 'go off the rails if you're hyping yourself up' on the supply side without translating that to customer value.

5/ 过度建设风险？他明确使用这个词："会出现过度建设...所有人都在赛跑"，虽然微软仍需"可能指数级"投入，但他对2027-2年前后产能上线后成为"出租方"的前景感到兴奋——这意味着算力价格将下降。 (180K: 我们本土现在摸着北美投资节奏过河，ROI会更高一些，也会少一些浪费）

Overbuild risk?  He uses that word, saying ‘there will be overbuild .. everybody’s going to race’, and though MSFT will still need to invest, 'maybe exponentially', he's 'so excited to be a leaser' with capacity coming online in 2027-2, meaning prices will come down for compute.

6/ 尽管与OpenAI关系出现裂痕，但大语言模型不会赢家通吃。他认为消费市场可能赢家通吃，但企业客户必然需要多供应商，模型领域也将如此。

LLMs are not a winner-take-all despite the (fraying?) OpenAI relationship.  He says consumer markets can be winner-take-all but anywhere where the buyer is an enterprise they will want multiple suppliers so he thinks that will happen on the model side.

7/ 杰文斯悖论："任何突破（如deepseek的能效边界突破）改变单位token性能曲线时，需求反而会激增"。

Jevon’s Paradox:   ‘Anytime there's a breakthrough, like even what DeepSeek did, with the efficient frontier of performance per token changes, the curve gets bent, and the frontier moves. That just brings more demand.’