Elon & Jensen Just Teamed Up: SpaceX Is Now THE Company ⚡️

以下是SpaceX电话会议的主要新闻和更新摘要的中文翻译： **整体情绪与财务状况：** * 此次电话会议的整体氛围极其乐观。 * 尽管承认高资本支出和大量净亏损，发言人表示，考虑到当前情况，现在过于关注财务数据是徒劳的。 * 许多资本部署的回报期约为一年（例如，预计20亿美元的数据中心投资将在一年内通过营收收回）。 * SpaceX的净亏损已实现同比减半。 **星舰与发射：** * 星舰第14次飞行将首次部署V3卫星进入轨道。 * SpaceX希望在今年内实现助推器和飞船的回收。 * 飞船的回收最快可能在第14次飞行中实现。 * 埃隆认为隔热罩问题已基本解决，目前正等待对在澳大利亚的S40进行检查以最终确认。 **算力与云服务协议（CSAs）：** * SpaceX目前的核心发展是快速高效地部署算力，这推动了云服务协议（CSAs）的增长。 * 截至第二季度末，SpaceX已上线约1.4千兆瓦（GW）的算力（高于第一季度末的1 GW）。 * 他们预计到2026年底，上线算力将超过2 GW。 * 他们预计到2027年底，上线算力将达到5到10 GW之间。 * 埃隆之前提到的15-20 GW是指电力和冷却设备，他们正在超额采购。 * 约120万个GPU大致相当于2 GW的算力。 * 从长期来看，SpaceX的算力容量中只有约10%用于训练Grok。 * 上线算力中的90%很可能将用于云服务协议（与Anthropic、谷歌等公司的合作）。 * 这些CSA交易的经济效益“越来越有利可图”，对EBITDA利润率贡献巨大。 * 预计算力市场的供需不平衡将持续存在。 **营收预测：** * SpaceX预计到今年12月底，年化营收将达到1000亿美元（埃隆认为甚至会更高）。 * SpaceX去年的营收不到190亿美元。 * 如果他们能达到10 GW的在线算力（预计明年底实现），即使星链或发射服务没有增长，这也能带来约5000亿美元的营收。 * 内部预测的1万亿美元营收目标已从2031年提前到2030年，甚至不排除2029年实现的可能性。 **星链V3与用户增长：** * 星链V3卫星的能力比V2“提升了一个数量级”。 * SpaceX预计将发射“数量级更多”的V3卫星。 * 随着他们扩展到费率较低的新国家，每用户平均收入（ARPU）预计会下降，但由于性能改进和用户数量增加，总收入仍将上升。 * 第一季度，星链新增了140万客户。 * 第二季度，星链新增了170万客户。 * 随着V3的部署，客户增长预计将加速。 * SpaceX预计星链将成为全球互联网流量的重要组成部分。 * 据报道，星链正在改变消费者行为，人们乘坐更短的航班以确保星链连接。 * 埃隆估计，大约需要1000颗V3卫星才能看到消费者性能的显著改善，这可能在明年第二季度实现。 **星链移动（Direct to Cell，直连手机）：** * 直连手机服务预计将实现显著增长。 * 用于该服务的V2卫星预计明年发射（与消费者宽带V3卫星不同）。 * 目前的系统通过当地电信运营商使用5兆赫的带宽。 * 新购买的EchoStar频谱将提供65兆赫的带宽，能力将大幅提升。 * 肖特韦尔点名了Verizon、AT&T和T-Mobile（总营收达6500亿美元），SpaceX预计通过消除盲区来获取其大量客户。 **星智AI卫星与地面AI1：** * 星智（Starmind）AI卫星将完全采用英伟达算力构建，埃隆表示这是“简直最好的架构”。 * SpaceX预计明年开始发射AI1卫星。 * 这些AI1卫星也将部署在地面。 * 星智V1卫星设计（减去太阳能电池板和散热器）将被部署在数据中心的地面上。 * 这种设计将大幅提高数据中心效率，简化NVL72机架，成本更低，效果更好。 * 埃隆与英伟达的黄仁勋合作，这可能意味着SpaceX能获得更好的芯片分配。 **星链企业与政府业务：** * 企业和政府业务营收同比增幅超过108%。 * 这部分营收被描述为“非常具有粘性和持久性”。 * SpaceX“从未流失任何一个企业客户”。 * 星链在航空领域的渗透率仅为10%，表明有巨大的增长空间。 * 企业营收最终预计将大幅超过消费者营收。 **Grok开发：** * Grok 4.6预计下周推出。 * Grok 4.7预计大约在一个月后推出。 * Grok 5有望在今年年底前推出。 * Grok 5的模型训练中应该会包含整个SpaceX数据语料库，有望使其成为工程师的最佳模型。 * Grok将集成到特斯拉车辆中，并作为擎天柱（Optimus）机器人的“大脑”。

Here's a summary of the key news and updates from the SpaceX conference call: **Overall Sentiment & Financials:** * The conference call was overwhelmingly positive. * While acknowledging high capex and a significant net loss, the speaker suggests focusing on financials now is a "fool's errand" given the context. * Many capital deployments have a payback period of roughly one year (e.g., $2 billion for data centers expected back in revenue within a year). * SpaceX has already narrowed net losses by half year-over-year. **Starship & Launch:** * Flight 14 of Starship will be the first to deploy V3 satellites into orbit. * SpaceX hopes to catch both the booster and the ship this year. * The catch of the ship could happen as soon as Flight 14. * Elon believes the heat shield problem is largely solved, awaiting inspection of S40 in Australia for final confirmation. **Compute & Cloud Service Agreements (CSAs):** * The primary story at SpaceX is the fast and efficient deployment of compute, which drives CSAs. * At the end of Q2, SpaceX had about 1.4 gigawatts (GW) of compute online (up from 1 GW at the end of Q1). * They expect to end 2026 with over 2 GW of compute online. * They expect to end 2027 somewhere between 5 and 10 GW of compute online. * Elon's previous comment about 15-20 GW referred to power and cooling equipment, which they are sourcing in excess. * 1.2 million GPUs roughly translates to 2 GW of compute. * Longer term, only about 10% of SpaceX's compute capacity will be needed to train Grok. * 90% of compute brought online will likely be for CSAs (deals with Anthropic, Google, and others). * The economics of these CSA deals are "increasingly favorable," contributing significantly to EBITDA margins. * The supply-demand imbalance for compute is expected to continue. **Revenue Projections:** * SpaceX expects to be at an annual run rate revenue of $100 billion by the end of December this year (Elon believes it will be even higher). * SpaceX did under $19 billion in revenue last year. * If they can hit 10 GW of compute online (expected by end of next year), that could translate to around $500 billion in revenue, even assuming no growth in Starlink or launch services. * Internal projections of $1 trillion in revenue have been pulled forward from 2031 to 2030, with a non-zero chance of 2029. **Starlink V3 & User Growth:** * Starlink V3 satellites are an "order of magnitude more capable" than V2. * SpaceX expects to launch an "order of magnitude more" V3 satellites. * Annual Revenue Per User (ARPU) is expected to come down as they expand into new countries with lower rates, but overall revenue will still climb due to improved performance and user numbers. * In Q1, Starlink added 1.4 million new customers. * In Q2, Starlink added 1.7 million new customers. * Customer additions are expected to accelerate with V3 deployment. * SpaceX expects Starlink to become a significant portion of global internet traffic. * Starlink is reportedly changing consumer behavior, with people taking shorter flights to ensure Starlink access. * Elon estimates around 1,000 V3 satellites will be needed to see improved consumer performance, possibly by Q2 next year. **Starlink Mobile (Direct to Cell):** * Significant growth is expected for the Direct to Cell service. * V2 satellites for this service are expected to launch next year (distinct from consumer broadband V3 sats). * The current system uses 5 MHz of bandwidth through local telco providers. * The new EchoStar spectrum purchase will provide 65 MHz of bandwidth, enabling massive increases in capability. * Shotwell called out Verizon, AT&T, and T-Mobile ($650 billion combined revenue), with SpaceX expecting to acquire many of their customers by eliminating dead zones. **Starmind AI Satellite & Terrestrial AI1:** * Starmind, the AI satellite, will be built exclusively with Nvidia compute, as Elon states it's "simply the best architecture." * SpaceX expects to begin launching AI1 satellites next year. * These AI1 satellites will *also* be deployed terrestrially. * The Starmind V1 satellite design (minus solar and radiator) will be deployed on the ground in data centers. * This design offers a major improvement in data center efficiency, simplifying NVL72 racks, costing less, and being more effective. * Elon is partnered with Nvidia's Jensen Huang, which likely means better chip allocation for SpaceX. **Starlink Enterprise & Government:** * Enterprise and government revenue is up over 108% year-over-year. * This revenue is described as "very sticky" and durable. * SpaceX has "never lost an enterprise customer." * Starlink is only 10% penetrated in aviation, indicating significant headroom for growth. * Enterprise revenue is eventually expected to significantly exceed consumer revenue. **Grok Development:** * Grok 4.6 is expected next week. * Grok 4.7 is expected about a month from now. * Grok 5 is hoped for before the end of this year. * Grok 5 should include the entire corpus of SpaceX data in its model training, potentially making it the best model for engineers. * Grok will be integrated into Tesla vehicles and as the "brain" of Optimus.