Fly2Mars

Developer and Investor on Cardano Ecosystem!

[Alex｜美股投资 @AlexResearchX]

微软账上躺着现金躺了几十年，这几天它的CFO第一次比较明确地说：条件合适的话，会考虑用别的资金来源。债务市场，"尚未"动用。

一家最不缺钱的公司开始为借钱铺路，背后一定是生意本身变了。

先看收入，这一季没什么好挑的。Azure 增长 48%，下季度指引还要更快。管理层反复说需求超过供给，有算力就能卖出去。光看这些，好像一切都很顺。

但往下翻就不一样了。这一季的资本开支几乎等于收入的一半，自由现金流只剩收入的两成左右。微软过去是写一次代码、卖一万次许可证的生意，现在越来越像一家重资产的基础设施公司，先盖房子，再收租。

更有意思的是 Office 这条线。很多人担心 OpenAI 和 Anthropic 会颠覆 Office，我觉得"颠覆"这个词不太准。Office 不会消失，真正的风险是被架空：员工每天打开的是 Claude 或者 ChatGPT，Word 和 Excel 退到后台，只用来存文件。最讽刺的是，微软自己最新的 Copilot Cowork 用的就是 Anthropic 的模型和那套智能体框架，核心能力是借来的。

那微软还剩什么？剩下的恰恰是企业最舍不得的东西：所有邮件、会议、文档都在它的系统里，员工身份、权限、合规、审计也都在它手上，还有十几年的采购合同。智能体再聪明，进了企业也得有人管它能看什么、做过什么。微软现在拼命在这一层圈地。

但这块地不是微软一家的。Palantir 做这件事比微软早得多，它把企业的订单、产线、供应链数据整合成一套业务模型，智能体在上面干活，每一步都有权限和审计。Databricks、Snowflake 从数据平台往上打，OpenAI 也推出了自己的企业智能体平台，想坐在所有系统之上。微软这次财报里提到的 Fabric IQ 本体、派工程师驻场帮客户落地，说实话都有点 Palantir 的影子。我的感觉是，办公协作这块微软几乎没有对手，但一深入到生产和运营这类业务流程，Palantir 更强。

好在这些模型公司很多都在租 Azure 的算力，谁的模型赢了，微软多少都能收一份房租。

所以微软不会被颠覆，但它在 AI 时代的钱会比软件时代难赚。一方面，智能层的利润要分给模型公司。另一方面，按人头卖席位的逻辑在松动。这季 M365 席位只增长了 5%，管理层说是销售执行问题，不一定是真的。如果一个智能体真能顶几个人的活，人头本身就会变少。

不过微软其实已经在转了。GitHub 先改成了按用量收费，M365 也加了按量计费，但现在大头还是按席位卖。按量这块能不能长到足够大，补上席位放缓留下的缺口，是接下来最关键的问题。

接下来几个季度可以观察：云业务毛利率还会不会往下走，按量收入能不能接住席位的放缓，还有这么大的资本开支什么时候开始变成利润，而不只是折旧。

我认为微软大概率会向PLTR的方向靠拢，这条路走好了，未来不会太差。

[Fly2Mars @BuildVWorld]

@AlexResearchX Cool！

[Shan Zhang @sz8ng]

We’re a few good moves away from being a top name in the industry

We’re seeing growth across liquidity, TVL, users, investors, deals, and now token expansion

And we’re having all the fun in the world doing it

Thanks to everyone joining us for the ride

Higher 🫡

[Alex｜美股投资 @AlexResearchX]

树大招风，说的就是现在的英伟达。

上周末，英伟达董事会宣布对黄仁勋启动独立调查。起因是有报道称他可能成了外国情报行动的目标。公司表示没有证据显示机密泄露，黄仁勋继续担任CEO。事情最后怎么收场还不好说，但站得越高，盯着你的人就越多。

不过英伟达真正的长期对手不在华盛顿，而在它的客户名单里。

谷歌、亚马逊、微软、Meta，每年在英伟达身上花上千亿美元，同时每一家都在做自己的芯片。没人愿意把命脉交给一家供应商，更没人愿意一直付那么高的毛利。

很多人觉得英伟达的对手是AMD。但AMD现在更像客户手里的谈判筹码，份额差得还远。

其实看过我之前分析的，各大厂高端自研芯片背后总有博通的影子，他应该是最强的竞争对手了，它不亲自做GPU，而是帮大客户定制专属芯片。谷歌的TPU就是它参与做的，Meta、OpenAI、Anthropic也是它的客户。上一季博通AI芯片收入接近170亿美元，已经算不上挑战者了。

英伟达卖的是万能钥匙，什么锁都能开，但贵。博通帮客户打的是专用钥匙，便宜，效率高。等AI从训练走向大规模推理，专用钥匙的优势会越来越明显。再加上网络本来就是博通的老本行，两家在计算和网络两条线上都在正面竞争。

但英伟达不会被打垮。AI的盘子还在变大，份额下降和收入增长可以同时发生；CUDA生态十几年的积累也不是一颗定制芯片替代得了的。

更可能的结局是：英伟达从"几乎垄断"变成"最大的那一家"。

所以如果你担心的是英伟达的份额被抢，博通可能是最好的对冲，因为抢份额最多的就是它。但如果你担心的是AI投资整体降温，买博通就对冲不了，两家会一起跌，博通的客户更集中，可能跌得更多。上周四芯片股大跌那天，博通跌了10%，比英伟达还多。

11月17日英伟达发财报，第二天是国会要求的闭门说明，这一周值得盯着。
$NVDA $AVGO $AMD $GOOGL

[Alex｜美股投资 @AlexResearchX]

I was intrigued by Firefly's $1.3 billion "engine deal," so I went through every document I could find.

First, an interesting detail. When AST previously signed launch agreements with SpaceX, it reported them to the SEC as material agreements and attached the full contracts. This time, for a $1.3 billion deal, AST issued only a press release. The actual contract wasn't filed. As of today, I haven't found any related SEC filings from Firefly either.

In other words, the only basis investors have for valuing this deal is what the two companies chose to say in their press releases.

Here's what we can confirm.

AST is paying approximately $800 million for a perpetual license to manufacture Firefly's Miranda and Vira engines itself, half in cash and half in AST stock. Another $500 million covers 36 engines for the first three rockets, plus three years of engineering support. Firefly retains the intellectual property, and the company says the agreement is "structured to be non-competitive."

But here's the problem: What exactly is Miranda?

It's not some peripheral technology. It's the heart of Eclipse.

Eclipse is the medium-lift rocket Firefly is developing with Northrop Grumman. Its first stage uses seven Miranda engines. In Firefly's long-term growth story, Eclipse is supposed to be the product that significantly expands the company's addressable market.

Now Firefly has sold perpetual manufacturing rights to the engine at the center of that strategy, before the engine has even flown once.

And those three years of engineering support? Essentially, Firefly is teaching AST how to build the engines itself.

AST plans to use these engines for Blacklift, a reusable rocket with roughly 20 tons of lift capacity, larger than Eclipse, with operations targeted to begin in late 2028.

For now, Blacklift is intended to launch AST's own BlueBird satellites, so there's no immediate direct competition.

But two things are hard to ignore.

First, AST is exactly the kind of large-satellite customer that Eclipse should be targeting. It needs launch capacity and could have been one of Eclipse's ideal customers. Now it's building its own rocket.

Second, if Blacklift eventually starts taking commercial launch orders, Firefly could find itself competing against a larger rocket powered by the same engine.

Whether that "non-competitive" restriction is actually written into the contract is something investors currently cannot verify.

The obvious comparison is Blue Origin supplying BE-4 engines to ULA. Blue Origin also sells engines to a potential competitor.

But there's one crucial difference.

Every time ULA launches a Vulcan, it needs to buy another set of BE-4 engines. That creates recurring revenue for Blue Origin.

Firefly, on the other hand, is giving AST the right to manufacture the engines itself. Once the engines for the first three rockets have been delivered, that revenue stream ends unless the contract contains undisclosed royalty provisions.

So I don't think this deal necessarily means Firefly has lost confidence in Eclipse.

A more accurate interpretation is that Firefly cannot afford to wait. It doesn't have the financial flexibility to slowly turn its engine technology into revenue, one engine at a time.

The company is burning roughly $80 million per quarter, and its stock had already fallen to around $15 before the announcement. For Firefly, $400 million in cash is a meaningful financial lifeline.

Trading a potentially long-lasting engine business for a large upfront payment is understandable. But it shouldn't be presented as nothing more than a "technology victory."

There's another issue worth considering.

Half of the licensing consideration is paid in AST shares, and AST's stock has fallen more than 50% over the past three months.

That $1.3 billion headline figure, much like the $855 million announced when Firefly acquired SciTec, is based on a stock valuation at a particular point in time.

For Firefly's November 10 earnings call, these are the questions I most want answered:

Is the manufacturing license exclusive or non-exclusive?
Can AST offer commercial launch services to third parties using Blacklift?
Are there any ongoing royalties?
When will Firefly actually receive the $400 million in cash?
How will the $800 million licensing fee be recognized as revenue?
Will the full agreement eventually be filed with the SEC?

If most of those answers favor Firefly, this could turn out to be a genuinely excellent deal.

If not, it looks more like pulling future revenue forward at a substantial long-term cost.

For research and discussion only. Not investment advice.
$ASTS $FLY

[Alex｜美股投资 @AlexResearchX]

Firefly can't even launch its own small rocket reliably, and now they're partnering on a 20-ton rocket? On top of that, AST has zero experience building rockets, and the first launch isn't until 2028. This is a really bad call. Going forward, AST's satellites won't just keep getting delayed because launches aren't reliable — we're also going to get a steady stream of bad news about the rocket's first flight slipping again and again. Honestly, this is terrible. Even companies as experienced as RKLB and SpaceX took forever to get there. Are you really naive enough to think you can just throw something together and make it work? Why not partner with RKLB instead? Signing a Neutron launch contract would've been the smart move. Management is just awful!

$RKLB $ASTS $FLY

[Angry Crypto Show @angrycryptoshow]

UPDATE: Charles Hoskinson says $NIGHT will "use Celestia as a data availability layer and Filecoin for archival storage. We have a great relationship with Celestia and Filecoin. There's a lot of data overhead—Nightframes for agents, private payments, the DeFi layer for privacy."

[Alex｜美股投资 @AlexResearchX]

我想明白了这件事情，产能卖到 2029 年，Lumentum 为什么偏偏现在把自己卖掉？

Evercore 在电话会前就问了这个问题：既然产能都卖光了，为什么要卖？管理层的回答是"考虑过其他选项"，以及和 Marvell 一起能更好地做 scale-up、scale-in。这解释了为什么选 Marvell，没有解释为什么是现在。

下面是我的推理。这是推断，公司没有这么说过。

【一、CPO 改变的不是需求，是谁来买】

可插拔模块时代，Lumentum 把 EML 卖给几十家模块厂（旭创、新易盛等）。客户分散，模块厂自己定义不了激光器，议价权在 Lumentum 手里。

CPO 时代，激光器从模块里搬出来，变成外置光源，买家变成少数几个掌握交换芯片或 XPU 的平台方：Nvidia、博通、AMD 和几家云厂商。

这个变化已经在数字里了：Lumentum 最大客户的收入占比，一年内从 15.4% 升到 26.6%。

【二、平台方的议价工具，比模块厂多得多】

· 规格由买方定。CPO 光源采用 ELSFP 标准化可插拔外形，越标准化，越接近按规格采购的零件，换供应商越容易。

· 买方主动扶持第二供应商。Nvidia 3 月同时给 Lumentum 和 Coherent 各投了 $20 亿，拿到的是非独占产能权。Coherent 计划把自有 InP 激光器产能翻一倍，到 2027 年底再翻一倍，超高功率 CPO 光源预计 2026 年第四季度开始量产。

· 买方可以自己做。博通有自己的 InP 晶圆厂，Intel 也在推 InP 硅光。

· 价值重心转移。在 CPO 系统里，价值集中在交换芯片、XPU 和台积电的封装，激光器只是平台方成本清单上的一行。

【三、所以现在是 Lumentum 议价权的顶点】

眼下，CPO 超高功率光源只有 Lumentum 一家通过 Nvidia 认证，产能卖到 2029 年，EML 价格还在涨，估值处在高位（传闻前市盈率约 42 倍预期盈利，5 年均值约 30 倍）。

往后看，第二供应商起来、标准化推进、买家集中度上升，议价权会慢慢向买方转移。量不一定少（scale-up 用光替代的是铜，是新增市场），但单价和利润率的天花板会降下来。

在议价权最强、估值最高的时候，把"供应商"的身份换成"平台"的股权，这是我认为最能解释交易时机的逻辑。

【四、几个需要说清的点】

· 收股票，说明不是套现，而是换位。Lumentum 股东将持有合并后约 33%，继续分享"电+光"平台的成长。这也是现实约束：Marvell 账上只有约 $39 亿现金，$1,150 亿的交易只能用股票。

· 短期市场和长期逻辑在打架。11/2 当天 MRVL 收 $171.41，4.5 股只值约 $771，远低于 $1,132 的标题价，LITE 收盘还比这个价值低约 11%。市场在为稀释、长达 18 个月的交割期和整合风险打折扣。如果整合顺利、CPO 按时放量，这个折价就是长期持有者的机会；反过来，折价也可能继续扩大。

· 时间表有不确定性。主流预期 scale-up CPO 在 2027 年下半年量产、2028 年部署，也有业内观点认为要等到 2030 年以后。如果铜撑得更久，可插拔 EML 的好日子更长，那现在卖就卖早了。

【结论】

这不是"生意变差了才卖"，也不是高位套现，而是在 CPO 把议价权交给少数平台方之前，用稀缺性换一个好价格，并且把自己变成平台的一部分。双方都在赌 CPO 时代到来：

Marvell 赌的是拿到光源后能和 Nvidia、博通平起平坐，Lumentum 股东赌的是平台的位置比供应商的位置更值钱，而且他们选择留在牌桌上。

[Alex｜美股投资 @AlexResearchX]

越研究 AI 时代搜索广告会怎么变，越觉得有意思。

先说我当前的结论（注意是"当前"）：AI 模式一定会降低每次搜索的广告收入。

为什么？以前你搜"女士化妆品"，第一页有十几条结果，中间可以插很多条广告。你每点一次，谷歌就赚一次，点得越多赚得越多。

AI 模式只有一个对话框、一个答案，很难塞进一堆广告。现在谷歌的做法是在对话里放一些商品图片，点击后右侧弹出类似广告的内容。但 AI 对话的特点是，用户会通过多轮问答，一步步缩小自己的选择范围，乱点的机会越来越少。AI 越精准，点击量原则上就越低。

数据上已经能看到迹象：$GOOG 本季搜索查询量增速是五年来最高，但搜索收入增速从年初的 19% 降到了 12%。查询越来越多，每次查询赚的钱在变少。

那为什么还要做 AI 模式？很简单：你不做，生意以后就会被 OpenAI 抢走。ChatGPT 的广告上线不到 200 天，年化收入已经做到 10 亿美元。

那就涨价吧。原则上，用户意图越精准，广告主的付费意愿越强。短期内涨价也许能稳住收入，但能持续吗？

更根本的问题是：用户为什么非要用你的 AI 模式？现在靠的是很多人还习惯用搜索框。

如果以后大家习惯每天打开 ChatGPT、Claude，搜索的主入口变成大模型的对话窗口，嵌在搜索里的 AI 模式还有意义吗？

谷歌目前手里的牌：

1. 苹果新版 Siri 用了 Gemini，但这是后台合作：用户看不到谷歌品牌，查询经过苹果服务器处理，谷歌拿的是授权费，不是搜索流量和广告。苹果还在开发自己的模型，随时可能替换。这个入口是苹果的，不是谷歌的。

2. 消费端 Gemini 用户量排第二，但很大一部分来自安卓、Chrome、搜索的预装，有多少是用户主动选择的，还不清楚。

3. 企业愿意付钱的市场上，最新数据没有，2025年底Gemini 排第三，我觉得它的份额应该比2025更低了：企业大模型 API 支出中，Anthropic 约 40%，OpenAI 27%，谷歌 21%（2025 年底数据）。

所以，目前谷歌的分发渠道是第一梯队，但模型本身在付费市场上还不是。别人用谷歌的模型，也不等于谷歌守住了入口。

结论：只有 Gemini 真正进入第一梯队，在用户有选择的情况下也被主动选择，谷歌才能把最值钱的商业查询留在自己的 AI 里。否则，谷歌会从一家垄断型的搜索公司，慢慢变成一家以云为主、重资产的 AI 基础设施公司。规模依然很大，但已经不再是那台印钞机。

[Fly2Mars @BuildVWorld]

@AlexResearchX Good Point

[Fly2Mars @BuildVWorld]

@AlexResearchX Thanks

[Alex｜美股投资 @AlexResearchX]

马斯克在买频谱，说要自己当运营商。
贝索斯买下 Globalstar，拿下了苹果。
夹在中间的 $ASTS ，目前手里只有 12 颗卫星。

它会被两个巨头碾碎吗？
$ASTS 的护城河到底是什么？窗口还剩多久？

这期中英文双语视频，用长期视角讲清楚 👇

[Fly2Mars @BuildVWorld]

@AlexResearchX Nice！

[Fly2Mars @BuildVWorld]

@AlexResearchX Good Insight！

[Fly2Mars @BuildVWorld]

I've never seen a team in the Cardano ecosystem as active and dedicated as @strikeperps

They're constantly improving their app, launching new features, and coming up with fresh ideas.

Instead of complaining about a lack of funding or constantly trying to get more money from the Cardano Foundation, they focus on finding ways to build something better.

Cardano needs more outstanding teams like this if the ecosystem is going to thrive.

I really hope @IOHK_Charles pays more attention to them and gives them the recognition and support they deserve.

[Alex｜美股投资 @AlexResearchX]

Why is $RKLB so committed to carbon composites for Neutron?

Because every kilogram saved in structural weight can translate into better payload performance.

And Rocket Lab has done this kind of thing before.

• Electron: The world's first carbon-composite orbital rocket.

• Rutherford: The world's first 3D-printed orbital rocket engine.

• Launch Complex 1: The world's first private orbital launch site.

Each of these breakthroughs required $RKLB to solve problems no one had solved before.

And together, these innovations helped make Electron the most frequently launched small orbital rocket in the world.

That's what makes me confident in Neutron.

Rocket Lab isn't just trying to build another rocket. It's pushing the limits of engineering efficiency, manufacturing speed, and launch economics.

I believe that relentless pursuit of efficiency could eventually make Neutron the most cost-effective reusable medium-lift rocket in the world.

First, they need to prove it.

But Peter Beck and his team have done the seemingly impossible before.

[Fly2Mars @BuildVWorld]

@AlexResearchX Nice！

[Alex｜美股投资 @AlexResearchX]

$RKLB's Launch Services gross margin is now around 43%, up from just 20–30% a year ago. Space Systems remains steady at around 35%.

Even without Neutron, Rocket Lab's existing businesses are already generating a gross margin of approximately 37%.

Backlog stands at $2.36B, up 137% YoY, with about 45% expected to be recognized as revenue within the next 12 months.

This gives Rocket Lab strong revenue visibility through the first half of 2027.

And I expect H1 2027 revenue to significantly exceed the $434M generated in H1 2026.

[Fly2Mars @BuildVWorld]

@AlexResearchX @elonmusk Nice！

[Fly2Mars @BuildVWorld]

@AlexResearchX Nice！