🔋𝓣𝓮𝓼𝓵𝓪4680🔋

Tesla owner since 2021, Husband, Father, Stand with Elon.

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

I promise： I'll buy a tesla when the stock hits $600 （PT of @Garyblack00 ），M3 or MY would be OK，before that I should deal with my Toyota.

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@cnfinancewatch 狮子的存在只是为了限制羚羊等动物的数量，避免植物被吃光。狮子也并不知道这个事情。

[海伦子Hellen @peng_hellen]

非常精彩的一个访谈，基本概括了所有Rocket Lab收购Iridium投资者们关心的问题。Peter Beck和COO Adam Spice齐上阵。

精简几个重点：1，在技术和应用场景上，Starlink和Rocket lab的卫星有明显差异。

Starlink等宽带星座使用的频段在恶劣天气下稳定性较弱，而Rocket Lab的L波段具备真正的全球覆盖能力，信号穿透性强，抗干扰，不受恶劣天气影响，也是“无GPS信号区域”的替代方案，所以它的客户是以稳定信号为刚需的群体，比如军事国防，航空航海，工业物流，精准农业等等。

Iridium的L波段频谱，也是唯一一个在全球所有地区都能落地的。

2，火箭运力极度紧缺。SpaceX的拼车发射订单已经排到了2028年以后，直接推动了电子与未来中子火箭的需求。

即使未来有新的火箭投入使用，首要任务也是先发射自家的卫星。（重点的话重复说：拥有自主发射能力很重要，否则卫星做的再好也极其被动。）

3，Peter Beck和Adam Spice都不喜欢债务，但为了避免过度稀释现有股东的股权，还是采用了股票+现金的方式收购，所以公司有了36亿美元的bridge loan（快短贵的贷款方式），��气是Iridium稳定且未来强劲的现金流。

但是不排除将来股价高位的时候将部分债务转化为股权。

4，收购Iridium让中子火箭“落地即跑”，跳过了耗时的商业计划，几乎瞬间让公司规模翻倍，立即带来现金流。（复制了SpaceX最高明的星链+猎鹰九号互惠双赢的模式）。

同时Iridium借用Rocket Lab的高度垂直整合和发射能力，可以让原本搁置的创新想法迅速孵化。

5，完美的时机，Rocket Lab刚好利用之前的股价大涨，以个位数级别的股权稀释完成了这比交易。

如果是在一年前，Rocket Lab可能没有足够的实力支撑这次收购。

而Iridium过去一年的市值涨的离谱，因为市场开始意识到了频谱资源的稀缺性，SpaceX开始接触EchoStar，亚马逊找上了Globalstar，频谱市场就是“抢椅子游戏”。如果收购再晚一年，Iridium会更贵， RKLB也许会错失这个机会。

总而言之，就像Peter Beck自己说的，这是一笔1+1等于3的收购， Rocket Lab也终有实力拿下Iridium。

Iridium是初代太空开创者，比SpaceX更早的实现了星座盈利。

但访谈中Peter Beck强调了RKLB并不想一夜之间成为大众互联网提供者，公司的重心依然是成为中立供应商，成为整个太空行业的支撑。

Talking Rocket Lab's New Era with Sir Peter Beck and Adam Spice https://t.co/qG0phqLHGp via @YouTube

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@zhongwen2005 @CharlesXBoy @TroyTeslike 2个多月没更新X了，这是什么情况。🤣

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@usakes_zhang 这个AJ投资特斯拉很多年了。

[qinbafrank @qinbafrank]

再聊聊RKLB的并购策略，通过收购完成了哪些业务的拼图？这两天太空板块热度开始起来，龙头RKLB的主要催化事件则是来自并购Mynaric切入雷射通讯，以及推出Gauss电推进系统。作为个人的心头好，RKLB还是关注了蛮久。其实RKLB这几年来通过不断并购收购来完成自己的业务布局，这里聊聊它的并购策略。关于RKLB具体业务模式，转发的长文里有详细分析介绍。

Rocket Lab在2016年前主要专注Electron火箭自主研发，几乎没有公司级收��。真正启动系统性收购战略是从2020年开始，至今共完成7笔公司级收购（全部聚焦Space Systems业务垂直整合），另有1笔资产收购（非公司整体��。
1、按照时间线梳理RKLB过去几年的并购案：
1）Sinclair Interplanetary（加拿大）
时间：2020年4月
业务：卫星姿态控制组件（reaction wheels反应轮、star trackers星跟踪器等）
意义：首笔收购，补齐卫星子系统供应链。

2）Advanced Solutions, Inc. (ASI)时间：2021年10月
金额：4000万美元 + 潜在550万美元earnout
业务：航天器飞行软件、任务仿真、制导导航与控制（GNC）
意义：首次进入软件层。

3）Planetary Systems Corporation (PSC)
时间：2021年12月
金额：约4200万美元现金 + 股份 + earnout
业务：卫星分离系统（separation systems）和分配器（dispensers）
意义：补齐卫星发射后的分离能力。

4）SolAero Holdings, Inc.
时间：2022年1月
金额：8000万美元现金
业务：高性能空间太阳能阵列（solar panels）和精密航空结构件
意义：补齐卫星电源与结构���心硬件。

5）Geost, LLC（及其母公司）
时间：2025年5月宣布，2025年8月完成
金额：2.75亿美元（1.25亿现金 + 1.5亿股票 + 最高5000万美元earnout）
业务：电光/红外（EO/IR）传感器载荷，用于导弹预警、空间域感知等国家安全任务
意义：正式进入卫星有效载荷（payload）领域。

6）Optical Support, Inc. (OSI)
时间：2026年2月26日完成
金额：未公开（规模较小）
业务：高精度光学系统与光机组件（lenses、optomechanical instruments）
意义：强化Geost后的光学能力，整合进Rocket Lab Optical Systems部门。

7）Mynaric AG（德国）
时间：2025年3月宣布，2026年4月14日完成
金额：1.553亿美元（少量现金 + 约227.7万股RKLB股票）
业务：激光光学通信终端（laser optical
communications terminals，CONDOR Mk3等）
意义：首笔欧洲收购，内部化高成本通信硬件，并获得慕尼黑欧洲据点。

额外资产收购（非公司���体）：Virgin Orbit Long Beach工厂及设备（2023年5月/2024年完成）：1610万美元，获得14万平方英尺制造设施，用于Neutron火箭规模化生产（发动机、复合材料等）。这不是完整公司收购，但极大加速中型火箭产能。

2、补齐了哪些短板？
Rocket Lab早期是“纯发射公司”（Electron火箭），Space Systems业务虽有Photon卫星平台，但大量核心子系统依赖外部供应商，导致了几个问题：
1）供应链风险高：交付周期长、成本不可控、容易卡脖子（尤其国防合同对“美国/盟国原产”要求严格）。
2）毛利率被压缩：卫星BOM中，反应轮/星跟踪器、太阳能阵列、分离系统、激光通信终端、光学载荷等高价值部件（合计占卫星成本30-50%）全部外采。
3）集成能力弱：无法作为“Prime Contractor”独立交付完整卫星+载荷+通信解决方案，在SDA、Golden Dome等国防大单中竞争力不足。
4）地域局限：缺乏欧洲/全球供应链布局，难以深度参与国际项目。
5）产能瓶颈：Neutron中型火箭规模化生产需要现成工厂和光学/通信硬件。

而通过过去十年收购精准补齐：
1）2020-2022年四笔收购 → 补全“卫星总线+子系统+软件”基础链条（从姿态控制、软件、分离、电源到结构）。
2）2025-2026年三笔（Geost+OSI+Mynaric） → 补齐“有效载荷+精密光学+激光通信”高端国防/商业短板，毛利率潜力大幅提升（激光终端单件成本占比可达10-20%）。
3）收购Virgin Orbit工厂 → 解决Neutron产能硬件瓶颈。

供应链风险大幅降低、生产周期缩短、成本可控、毛利率提升，同时打开国防Prime Contractor大门。

3、最终形成了什么样的体系？
Rocket Lab已从“小型运载火箭公司”彻底转型为全球领先的垂直一体化空间基础设施提供商（End-to-End Space Systems Prime），
���心定位是“Space as a Service + 国防Prime”。业务布局有
1）发射层：Electron（小型、高频次，已超75次成功发射）+ Neutron（中型火箭，2026年底首飞在即，载荷能力远超Electron）。
2）卫星制造层：Photon卫星平台 + 全套子系统（姿态控制、电源、分离、飞行软件）。
有效载荷与光学层：EO/IR传感器（Geost）+ 高精度光学/光机组件（OSI）。
3）通信层：激光光学终端（Mynaric）——实现星间/星地高速安全链路。
4）制造与产能层：Long Beach工厂（前Virgin Orbit）+ 全球多地设施，支持规模化生产。
5）附加能力：国防/国家安全重点（SDA 8.16亿美元18颗卫星合同、导弹跟踪、空间域感知），同时服务商业星座。

形成高度垂直整合的产业体系：几乎所有关键环节自产（类似SpaceX闭环，但更聚焦中小型/中型星座和国防市场）。可以实现端到端交付：客户可一站式采购“发射 + 完整��星 + 载荷 + 通信 + 运营”，显著降低集成风险和时间。

双轮驱动：Launch（稳定现金流）+ Space Systems（更高毛利、更高增长，backlog主力，已超18亿美元）。

战略定位：现在RKLB既是发射商，又是“卫星工厂 + 关键组件供应商 + 国防Prime承包商”，在D2D/LEO星座、国家安全领域形成独特护城河。

去年底https://t.co/1g8EPVBmr8 rklb就拿到一笔8亿美金18颗卫星的订单，开始参与太空国防领域的最大项目。

简单说，过去十年7笔收购让Rocket Lab彻底补齐了“从火箭到卫星总线、再到有效载荷、光学、通信”的全链条拼图，形成了从发射到在轨数据全链条自给自足的垂直一体化空间平台，竞争力与估值逻辑已完全不同于十年前的纯发射公司。未来Neutron + 规模化Space Systems将成为主要增长引擎，同时在国防领域作为“ disruptive Prime”与传统巨头竞争。

[Ryan Wang 🇹🇼 @ryanwang]

xAI is undertaking a major overhaul of Grok’s training and inference software stack. The core of this transformation is rewriting the entire system from high-level frameworks into C/C++, drastically eliminating intermediate abstraction layers, and performing exact mapping (exact-map) optimizations tailored to NVIDIA GB300 hardware.

Most current AI models (including previous versions of Grok) are built on high-level frameworks such as Python + PyTorch or JAX. While convenient for development, these frameworks introduce numerous intermediate layers, buffers, and runtime overhead, resulting in low GPU utilization, memory waste, and increased latency and costs. As Elon hinted, the real “massive gains” come from this low-level reconstruction - completely discarding unnecessary software layers and directly implementing the core stack in C/C++.

Since NVIDIA’s CUDA is itself based on C/C++, xAI is pushing further down the stack by writing lower-level CUDA kernels and performing custom tuning for the architecture of the GB300 NVL72 system. Through precise memory allocation management, reduced unnecessary data copying, optimized KV cache, and more aggressive quantization and scheduling strategies, hardware utilization can be significantly improved.

The expected outcomes span multiple dimensions: First, performance gains—inference latency will decrease and tokens generated per second will increase, allowing users to experience faster and smoother responses. Second, memory savings—high-level frameworks often waste large amounts of VRAM; the rewrite enables precise memory control, supporting larger batch sizes or longer context windows on the same hardware. Third, cost and efficiency—GPU utilization is expected to rise from 60-70% to over 90%, enabling the same computing power to serve more users while significantly reducing cloud computing expenses and power consumption. This is critical for xAI’s large-scale deployment.

This reconstruction is not only an upgrade for Grok but will also drive the entire AI industry from a “scale parameters and hardware” paradigm toward a “software extreme optimization” architecture. Amid current shortages of computing resources and memory, it effectively increases effective compute by 30-50%, lowers inference costs, and accelerates the adoption of large models. In the long term, it will elevate the importance of low-level engineering, encourage more companies to invest in infra talent, and shift the competitive landscape - favoring companies with strong systems engineering capabilities. The industry as a whole will place greater emphasis on efficiency rather than pure scale, moving toward a more sustainable and cost-controllable direction.

@grok is my understanding of this Elon’s thread right?

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@elonmusk @Tesla_AI Finally, Elon remembered that he also owned a company called Tesla.

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@hanking66 Peter Beck的减持计划好像还没完成。

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@Money_or_Life_X 芯片的原材料是沙子，美光怎么不说这事呢？

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@SawyerMerritt Starmind.

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@Rustallintsla 回国了？

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@Rustallintsla 这…

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@WangNextDoor2 果然不出所料，V3黄了🤣

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

Only Tesla can do.

[Sawyer Merritt @SawyerMerritt]

The Dutch vehicle ​authority (RDW) ‌has released a statement after it received criticism for approving @Tesla's FSD (Supervised).

"The RDW carefully assessed and tested this driver assistance system before approval was granted. As part of this process, we analyzed and evaluated data from vehicles that had been driven in Europe. We also collected and assessed our own data and continue to monitor the performance of this system on the road.

The RDW did not rely solely on Tesla’s data; we also carried out extensive testing ourselves using our own test equipment. Over a period of more than 3,000 hours, testing took place both on test tracks and on public roads, under a wide range of conditions, including complex and busy urban traffic, a broad variety of road types, and different - including extreme - weather conditions. This included the use of data from 1.8 million kilometres driven in Europe with FSD Supervised.

The RDW’s own tests — comprising more than 1,000 test runs — were conducted in accordance with European regulations. The Tesla data used by the RDW also complied with those rules. The RDW independently assessed and validated the process used to collect the data.

Through these tests, the RDW gathered objective information and verified the manufacturer’s data. The RDW does not base its assessment solely on information provided by the manufacturer. We carried out extensive independent investigation. On the basis of this overall process, the RDW concluded that the system meets the applicable requirements. As a result, a European type approval valid for use in the Netherlands was granted.

Safety is the RDW’s foremost priority. This Tesla driver assistance system supports the driver to a greater extent than other systems because, when activated, it takes over multiple driving tasks. The proper use of driver assistance systems makes a positive contribution to road safety by supporting the driver in performing driving tasks; it is intended as an aid to the driver. Owing to the continuous and strict monitoring of the driver in the vehicle, this driver assistance system is at least as safe as other driver assistance systems. We have thoroughly assessed and tested this system over a period of more than 18 months."

Full statement linked in thread below:

[Sawyer Merritt @SawyerMerritt]

Tesla is sending out this email to New Jersey Tesla owners, warning them that NJ could block autonomous vehicles, and to take action.

"Proposed legislation moving through Trenton right now would impose restrictions so severe that true driverless deployment would remain illegal. Your voice is critical to ensure New Jersey remains at the forefront of transportation innovation.

We urge you to contact your state legislators today and tell them to:
• Fix S. 1677 / A. 3968, and
• Support autonomous vehicle policy that encourages safety outcomes, open competition, and expanded access for every New Jersey resident.

Autonomous vehicles are no longer a futuristic concept. The driverless technology exists and it is already making roads measurably safer in states that have welcomed it. Instead of positioning itself as a leader in transportation innovation, New Jersey is about to go in the opposite direction."

You can take action here: https://t.co/99JQca3KQb

[NL F$D Inside @TSLA_inside_]

🚨 BREAKING: Dutch Minister Karremans pushes back hard against the NOS/Reuters narrative on Tesla in Parliament.

“The RDW does NOT rely on Tesla marketing claims.

We conduct our own independent research more than 1.6 million kilometers of real-world testing on Dutch roads.

Safety comes first. Period.”

The Netherlands continues to lead Europe in evidence-based regulation. While critics speculate, the RDW makes decisions based on data, testing, and facts.

This is what taking innovation seriously looks like.
@vincentkar

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@JasonZX 还是Biden walk的级别。

[🔋𝓣𝓮𝓼𝓵𝓪4680🔋 @Roystable]

@AYi_AInotes True.