Emergency Ep: The SpaceX Earnings Call Was Insane (Tesla Merger News) ⚡️

以下是根据您提供的要求完成的简体中文翻译： 基于 2026 年 10 月 21 日发布的播客节目文字记录，以下是主持人迪兰（Dylan）提出的每一条新闻、事实和观点的总结： ### 整体情绪与市场论点 * **主持人视角：** 迪兰对太空探索技术公司（SpaceX）极其看好（bullish），将其目前的定位和领导地位比作特斯拉的早期阶段，但他同时指出，公众和绝大多数人对太空、算力、电力、冷却、连接或系统集成（SI）一无所知，完全不知道 SpaceX 在未来三到五年到七年内会做什么。 * **埃隆·马斯克的重心：** 埃隆高度专注于建设和技术，似乎远离了政治，并急于阐明团队的看好立场。 * **估值观点：** 虽然聪明的朋辈可能会认为 SpaceX 今天的估值高得可笑，但迪兰认为，放眼未来 5 到 10 年，该公司的估值被严重低估了——这并非基于当前的营收或盈利能力，而是基于 SpaceX 即将要做的事情。他表示（不构成财务建议），在 2030 年以 2 万亿美元的估值买入 SpaceX，会让人觉得像是明抢。 ### SpaceX 算力、基础设施与 TerraFab * **基础设施优势：** SpaceX 声称其能够比其他任何人快 12 到 18 个月将电力、冷却和网络投入运行。这是因为 SpaceX 可能是地球上最好的铸造公司，能够自己制造涡轮机的内部管道和叶片。 * **巨像 2 号 / 孟菲斯（Colossus 2 / Memphis）：** 埃隆·马斯克将其绝大部分时间（一周四天，日夜兼程）花在孟菲斯的巨像 2 号项目上。 * **部署速度：** 他们今年有望实现每月投入运行 200,000 个 GB300 的目标，计划明年继续保持这一速度，并在 2027 年看到相似或更快的部署速度，届时将转向维拉·鲁宾（Vera Rubin）产品线。 * **吉瓦与营收预测：** * 埃隆顺口提到，到 2027 年底将有 10 吉瓦的算力投入运行（实际数字可能会显著更高）。 * 算力租赁协议价格处于每瓦 30 美元到 50 美元的高端（迪兰采用了较为保守的每瓦 50 美元）。 * 以每瓦 50 美元的价格出租 10 吉瓦算力，仅这单一业务线就能带来 5000 亿美元的营收。 * SpaceX 可能在今年年底前拥有 3 吉瓦算力投入运行，明年新增 10 吉瓦（到 2027 年底总计 13 吉瓦）。将这 13 吉瓦的算力 100% 出租，仅从算力租赁协议中就能获得 6500 亿美元的营收。 * 埃隆表示，他们预计 2027 年结束时的年度运行率（run rate）将在 5000 亿美元到 1 万亿美元之间，并在 X 上发文称，SpaceX 在 2028 年实现 1 万亿美元营收已不再是不可能的任务。 * **算力策略与租赁：** * 理想的目标是 50% 用于出租，50% 用于内部项目和推理，从而赋予他们追求最高利润率选项的灵活性。 * SpaceX 想要短期租赁协议（而非长期），因为他们确信无法餍足的需求将会增长，而且埃隆认为，由于每瓦能够获得更高的智能和效用，租赁价格将会上涨（而不是下跌）。H-100 的租赁价格实际上比该芯片刚问世时还要高。 * 每家寻求算力的大型云厂商（hyperscaler）以及数十家小型公司都在对 SpaceX 穷追不舍。他们本可以把手头的所有算力租出去好几次。 * **电力、冷却与网络：** 限制因素是电力、冷却和网络，而不是算力本身。团队高度关注这一点，并正力求在 2027 年之前投入运行 20 吉瓦的电力、冷却和网络，以拥有富余产能。如果到 2027 年底 SpaceX 投入运行的总算力低于 13 吉瓦，埃隆会感到惊讶。 * **团队与执行力：** SpaceX 将负责将全球大约 90% 的质量送入轨道（全球市场份额）的工程师团队调配到了地面算力项目上。埃隆将其形容为纽约洋基队去打少儿棒球联盟，是“不公平的”，这使得他们能够实现无人能及的部署速度。 * **资本支出（CapEx）：** 布雷特·约翰逊（Brett Johnson）表示，仅第四季度单季，SpaceX 预计的资本支出就在 350 亿美元到 40 亿美元之间。明年的投资回收期可能会从 12 个月缩短至 6 个月。第一季度的营收预计将在 400 亿美元左右，这很大程度上归功于清晰的算力前景。 * **TerraFab：** * 埃隆表示，TerraFab 产出的价值将使其仅凭单一建筑就能跻身全球国家 GDP 的前 20 名。 * 他们希望德州超级工厂（Giga Texas）的研究工厂（ResearchFab）产出的第一颗芯片（用于测试生产线的 SRAM 类型芯片）能够在明年 6 月份实现功能正常并下线。 * 吉本斯克里克（Gibbons Creek）主 TerraFab 厂房进行量产的内部目标是 2028 年第三季度（尽管迪兰对能否赶上这一时间表表示怀疑）。 ### 人工智能、Grok 与 GrokBot * **维拉·鲁宾训练集群：** 到明年第二季度，他们预计将拥有世界上基于维拉·鲁宾（Vera Rubens）构建的最强大的训练集群（约有 440,000 个），这将有助于 Grok 成为世界上最好的模型（或者至少极具竞争力）。Grok 有望在明年年中达到前沿水平。 * **GrokBot 的增长：** 埃隆表示，GrokBot 的增长速度是他一生中从未见过的。 * **GrokBot 的功能与迪兰的使用案例：** 迪兰使用 GrokBot“一次性”为谷歌浏览器（Google Chrome）构建了一个定制的、高度实用的阅读模式，该模式可以处理格式、图像、字幕、要点列表、可自定义的颜色/字体，并且即使在页面刷新后也能保存高亮标记。他指出，这比现有的扩展商店阅读模式要好得多，并且可以实现商业化。迪兰强调，现在任何人都可以使用 GrokBot 构建定制软件工具。 * **GrokBot 企业版与订阅用户：** * GrokBot 正在开始其企业版推广，这将通过企业每季度在 Token 上花费数亿美元来推动业务增长。 * 埃隆表示，从长期来看，GrokBot 可能会占其系统集成（SI）算力的大多数。 * 格温·肖特韦尔（Gwen Shotwell）报告称，截至本月初，GrokBot 拥有 80 万付费周活用户，比前一个月（40 万）翻了一番。增长主要由 B2B / 企业业务驱动。 ### 星链（Starlink）与星链移动业务（Starlink Mobile） * **星链消费端指标：** 星链本季度新增了 160 万新消费者，使零售用户总数达到 1360 万订阅用户。 * **每用户平均收入（ARPU）：** 布雷特·约翰逊报告称，每用户平均年收入（ARPU）上升至每月 69 美元（高于上一季度的 66 美元），打破了其随时间推移会呈下降趋势的预期。 * **星链移动业务 / 地面网络：** 埃隆在 X 上表示，蜂窝通信行业的资深人士在观察这盘棋局时看不到解决方案，是因为他们不理解 SpaceX 正在开发完全颠覆战略格局的全新棋子。SpaceX 将准备一份演示文稿来解释这个新网络，这是团队看到星链年营收达到数千亿美元有清晰路径的主要原因之一。 ### 特斯拉与 SpaceX 合并 * **当前状态：** 这打消了近期会有任何宣布的念头。SpaceX 和特斯拉都已成立特别委员会来进行谈判，但埃隆是利益冲突方，不能参与谈判。 * **埃隆的立场：** 埃隆曾两次确认，在他看来，合并这两家公司是个好主意。 * **迪兰的时间表/预期：** 迪兰表示，如果由于委员会无法达成一致而导致交易未能发生，他会感到失望。他指出，合并并非板上钉钉的事，目前尚未达成。他推测在 2027 年之前不会有任何公告发布，这给特斯拉的 Robotaxi（无人驾驶出租车）留出了规模化扩张并在投票前看到股价拐点的时间，但他依然相信合并终究会发生。 ### 迪兰的个人投资组合与结语 * **特斯拉投资者：** 迪兰对那些因为关注点和兴奋点都在 SpaceX 身上而感到沮丧的特斯拉投资者表示同情，但他相信特斯拉的高光时刻将通过 Robotaxi 和 Optimus（擎天柱机器人）到来。 * **迪兰的持仓：** 迪兰将其个人可投资资产的大约 97% 持仓于特斯拉，大约 3% 持仓于 SpaceX，他强调了自己对这场游戏的热爱，并对能够报道这些技术表示感激。

Based on the transcript from the podcast published on 2026-10-21, here is a summary including every single news item, fact, and opinion presented by the host, Dylan: ### Overall Sentiment and Market Thesis * **Host Perspective:** Dylan is extremely bullish on SpaceX, comparing its current positioning and leadership to the early days of Tesla, though noting that the general public and most people know nothing about space, compute, power, cooling, connectivity, or SI, and have no idea what SpaceX will do over the next three to five to seven years. * **Elon Musk’s Focus:** Elon is laser-focused on building and technology, seemingly removed from politics, and eager to articulate the team's bullish position. * **Valuation Opinion:** While smart friends might argue SpaceX is laughably overvalued today, Dylan argues it is severely undervalued looking at the next 5 to 10 years, not based on current revenue/profitability, but on what SpaceX is about to do. He states (not financial advice) that buying SpaceX today at a $2 trillion valuation in 2030 will feel like highway robbery. ### SpaceX Compute, Infrastructure, and TerraFab * **Infrastructure Advantage:** SpaceX claims it can bring power, cooling, and networking online 12 to 18 months faster than anyone else. This is because SpaceX is potentially the best casting company on planet Earth, making its own veins and blades for turbines. * **Colossus 2 / Memphis:** Elon Musk is spending the majority of his time (four days a week, day and night) in Memphis at Colossus 2. * **Deployment Rates:** They are on track to bring 200,000 GB300s online every single month this year, plan to continue that pace next year, and will see a similar or accelerated pace in 2027 moving to the Vera Rubin line. * **Gigawatt and Revenue Projections:** * Elon threw around having 10 gigawatts of compute online by the end of 2027 (could be meaningfully higher). * Compute leasing agreement rates are seen on the high end of $30 to $50 per watt (Dylan uses a conservative $50 a watt). * Renting out 10 gigawatts at $50/watt equals $500 billion in revenue from that single business line. * SpaceX could have 3 gigawatts online by the end of the year and an incremental 10 gigawatts next year (13 gigawatts total by the end of 2027). Renting 100% of 13 gigawatts would be $650 billion in revenue just from compute lease agreements. * Elon stated they are expecting an annual run rate exiting 2027 between $500 billion and $1 trillion, and posted on X that $1 trillion in SpaceX revenue in 2028 is no longer impossible. * **Compute Strategy & Leasing:** * The goal is ideally 50% rented out and 50% used for internal projects and inference, giving them variability to pursue highest-margin options. * SpaceX wants shorter-term lease deals (not long-term) because they are confident insatiable demand will increase, and Elon thinks rates will go up (not down) due to getting more intelligence and utility per watt. H-100 leasing rates are actually higher than when the chip first came out. * SpaceX is being hounded relentlessly by every hyperscaler wanting compute and dozens of smaller companies. They could rent out all the compute they have several times over. * **Power, Cooling, and Networking:** The limiting factor is power, cooling, and networking, not the compute itself. The team is hyper-focused on this and trying to bring online 20 gigawatts of power, cooling, and networking by 2027 to have excess capacity. Elon would be surprised if SpaceX brought online less than 13 gigawatts of total compute by the end of 2027. * **Team & Execution:** SpaceX took the team of engineers responsible for getting roughly 90% of all mass to orbit (global market share) and put them on the terrestrial compute project. Elon described it as the Yankees playing Little League and "not fair," enabling deployment rates no one else can touch. * **CapEx:** Brett Johnson said SpaceX's expected CapEx in Q4 is $35 billion to $40 billion for the fourth quarter alone. Payback periods might drop from 12 months down to six months next year. Q1 revenue is expected to be on the order of $40 billion, largely due to the clear line of sight into compute. * **TerraFab:** * Elon stated the value of TerraFab's output will place it in the top 20 of country GDP in a single building. * The first chip out of the ResearchFab in Giga Texas (an SRAM type of chip to test the line) is hoped to be functional and coming off a line in June of next year. * Internal target for volume production from the main TerraFab site at Gibbons Creek is Q3 2028 (though Dylan has doubts on hitting that timeline). ### AI, Grok, and GrokBot * **Vera Rubens Training Cluster:** By Q2 next year, they expect to have the most powerful training cluster online in the world built on Vera Rubens (around 440,000 of them), helping Grok potentially be the best model in the world (or at least competitive). Grok could get to the frontier by the middle of next year. * **GrokBot Growth:** Elon stated GrokBot is growing at a rate he has never seen in his life. * **GrokBot Capabilities & Dylan's Use Case:** Dylan used GrokBot to "one-shot" build a custom, highly functional reader mode for Google Chrome that handles formatting, images, subtitles, bullet points, customizable colors/fonts, and saves highlights even after page refreshes. He notes it is far better than existing extension store reader modes and could be monetized. Dylan emphasizes that anyone can now build customized software tools using GrokBot. * **GrokBot Enterprise & Subscribers:** * GrokBot is beginning its enterprise rollout, which moves the needle via companies spending hundreds of millions of dollars every quarter on tokens. * Elon stated that long-term, GrokBot might make up a majority of their SI compute. * Gwen Shotwell reported that GrokBot has 800,000 paid weekly users as of earlier this month, a doubling from the month prior (which had 400,000). Growth was largely driven by B2B/enterprise. ### Starlink and Starlink Mobile * **Starlink Consumer Metrics:** Starlink added 1.6 million new consumers in the quarter, bringing the new retail number to 13.6 million subscribers. * **ARPU:** Brett Johnson reported the annual revenue per user (ARPU) went up to $69 per month (up from $66 last quarter), defying expectations that it would trend down over time. * **Starlink Mobile / Terrestrial Network:** Elon stated on X that veterans of the cellular industry look at the chessboard and don't see a solution because they don't understand SpaceX is developing entirely new pieces that fundamentally alter the strategic landscape. SpaceX will prepare a presentation to explain this new network, which is a big part of why the team sees a clear path to hundreds of billions in annual revenue from Starlink. ### Tesla and SpaceX Merger * **Current Status:** Cold water was poured on the idea of an announcement anytime soon. Both SpaceX and Tesla have formed special committees to negotiate, but Elon is a conflicted party and cannot be involved in the talks. * **Elon's Stance:** Elon confirmed twice that in his mind, it would be a good idea to merge the companies. * **Dylan's Timeline/Expectations:** Dylan expresses disappointment if the deal doesn't happen due to committees failing to agree. He states the merger is not a certainty and hasn't been reached yet. He speculates no announcement will happen until into 2027, giving Tesla's robotaxi time to scale and see an inflection in the stock before a vote, but he still believes a merger will happen eventually. ### Dylan's Personal Portfolio and Closing Thoughts * **Tesla Investors:** Dylan sympathizes with Tesla investors feeling discouraged that attention/excitement is around SpaceX, but believes Tesla's time in the sun will come via Robotaxis and Optimus. * **Dylan's Holdings:** Dylan holds about 97% of his personal investable assets in Tesla and about 3% in SpaceX, emphasizing his love of the game and gratitude for covering these technologies.