SpaceX / 全球AI竞争主要集中在几个方向 SpaceX / Global AI competition mainly concentrated in a few directions
你好，希望你今天一切都好。
Hello, hope you're having a great day.
本季度多家科技巨头交出了远超预期的成绩单，从云计算到航天基础设施，AI浪潮正在以惊人的速度重塑整个产业格局。以下为本期市场简报，涵盖微软财报解读、SpaceX最新营收展望、AST SpaceMobile竞争格局变化，以及投资心态分享。
This quarter, several tech giants delivered results that far exceeded expectations, and the AI wave is reshaping entire industries at a remarkable pace, from cloud computing to space infrastructure. This briefing covers Microsoft's earnings breakdown, SpaceX's latest revenue outlook, a shift in the competitive landscape around AST SpaceMobile, and a note on investing mindset.
微软：稳健全面超预期
Microsoft: A Broad, Steady Beat
纳德拉带领的团队，用这份财报向市场证明了一件事：过去外界对微软的看空判断是错的。
Satya Nadella's team has used this earnings report to demonstrate one thing to the market: the bearish case against Microsoft was mistaken.
本季度整体表现节奏稳健，从营收到利润各项指标全线走强。
This quarter's performance was steady across the board, with every metric from revenue to profit trending stronger.
具体数据方面：营收达到900亿美元，同比增长18%，超出市场预期；经调整后每股收益为4.74美元，高于华尔街此前预测的4.22美元；营业利润达406亿美元，高于预期的391亿美元；Azure云业务增长43%，年化收入首次突破1000亿美元大关，创下历史新高；商业积压订单同比飙升84%，达到6780亿美元。
The specifics: revenue reached $90 billion, up 18% year-over-year and above market expectations; adjusted EPS came in at $4.74, above Wall Street's forecast of $4.22; operating income hit $40.6 billion, above the expected $39.1 billion; Azure grew 43%, with annualized revenue crossing the $100 billion mark for the first time — an all-time high; and commercial backlog surged 84% year-over-year to $678 billion.
其中一项数据尤其值得关注：Copilot付费用户已突破3000万，而且这些都是真正在日常工具中使用的活跃付费用户——Word、Excel、Outlook、Teams等，他们要的是切实的生产力提升。
One data point in particular stands out: Copilot's paid user base has surpassed 30 million, and these are genuinely active paying users embedded in daily workflow tools — Word, Excel, Outlook, Teams — looking for real productivity gains.
这一数字较上季度增长约50%。
That figure is up roughly 50% quarter-over-quarter.
资本支出计划维持不变，节奏保持稳定。
Capital expenditure guidance remained unchanged, holding a steady pace.
市场反应也在意料之中，花旗、富国银行、Piper Sandler等多家机构相继上调了目标价。
The market response followed accordingly, with multiple institutions — including Citi, Wells Fargo, and Piper Sandler — raising their price targets.
SpaceX：从火箭公司到AI基础设施平台
SpaceX: From Rocket Company to AI Infrastructure Platform
马斯克在最新一次财报电话会议上，给出了极具野心的营收展望：SpaceX目前预计到2030年营收将达到1万亿美元，并表示在2029年实现这一目标"存在一定概率"。
At SpaceX's latest earnings call, Musk laid out an ambitious revenue outlook: the company now projects reaching $1 trillion in revenue by 2030, and said there is "some chance" of hitting that figure as early as 2029.
他同时透露，得益于近几周新增的60亿美元算力合同，到今年年底，公司的年化经常性收入（ARR）有望达到1000亿美元，较6月份的104亿美元大幅提升。
He also disclosed that, thanks to $6 billion in new compute contracts signed in recent weeks, annual recurring revenue (ARR) is expected to reach $100 billion by year-end, up sharply from $10.4 billion in June.
马斯克在电话会议中表示："我们预计今年年底算力将超过2吉瓦，到明年年底，累计在线算力将大幅提升，接近10吉瓦，而非原先预期的5吉瓦。"
On the call, Musk stated: "We expect to exceed 2 gigawatts of compute capacity by the end of this year. By the end of next year, cumulative online compute will increase substantially, approaching 10 gigawatts rather than the previously expected 5 gigawatts."
展望未来的技术路线，公司已决定全面采用英伟达平台进行开发，理由是团队认为Vera架构是目前最优的AI计算方案，双方合作关系将持续深化。
Looking ahead on the technical roadmap, the company has decided to fully adopt NVIDIA's platform for development, citing the view that the Vera architecture is currently the best AI computing solution available, with the partnership set to deepen further.
马斯克还提到，这正是英伟达创始人黄仁勋多年来一直传达的理念：数据中心运营商真正关心的是总体拥有成本，而不仅仅是芯片单价，而英伟达在这方面始终提供最优方案。
Musk also noted that this echoes what NVIDIA founder Jensen Huang has long emphasized: what data center operators actually care about is total cost of ownership, not just chip price — and NVIDIA has consistently delivered the best value on that front.
他进一步表示，星链未来将承载全球大部分互联网服务。SpaceX首席财务官Bret Johnsen也表示，受益于AI算力交易，公司有望在今年年底前实现1000亿美元的年化经常性收入。
He further stated that Starlink is positioned to eventually carry the majority of global internet traffic. SpaceX CFO Bret Johnsen echoed this, saying that AI compute deals are putting the company on track to reach $100 billion in ARR by the end of this year.
财报电话会议上还提到，第三季度前几周，公司已新签一份为期6个月的云服务收入合同，将于今年10月起逐步生效。
The call also disclosed that in the first few weeks of the third quarter, the company signed an additional six-month cloud services revenue contract, set to phase in starting this October.
在星链卫星升级方面，马斯克表示，星链V3卫星的性能比已经相当出色的V2卫星还要高出一个数量级，公司计划发射的V3卫星数量也将比V2多出一个数量级，这意味着交付带宽将实现约两个数量级的提升。他补充道，即便单比特的变现效率下降10倍，星链的整体收入仍将实现约10倍的增长。
On Starlink satellite upgrades, Musk said the V3 satellite is roughly an order of magnitude more capable than the already highly capable V2, and the company plans to launch roughly an order of magnitude more V3 satellites than V2 — implying close to a two-order-of-magnitude increase in delivered bandwidth. He added that even if per-bit monetization dropped by a factor of 10, Starlink's overall revenue would still grow roughly tenfold.
除营收目标外，多个关键项目也传来积极进展：星舰隔热罩问题已基本解决；AI数据中心的资本支出预计一年内即可收回成本；到2027年底，数据中心容量计划扩大5倍甚至更多；英伟达将继续作为SpaceX地面与太空数据中心的独家AI芯片合作伙伴。
Beyond the revenue targets, several key projects also showed positive progress: the Starship heat shield issue has largely been resolved; AI data center capex is expected to be recouped within about a year; data center capacity is planned to expand fivefold or more by the end of2027; and NVIDIA will remain the exclusive AI chip partner for both SpaceX's ground-based and space-based data centers.
马斯克表示，建造数据中心远比建造火箭简单，公司正把火箭制造中积累的工程经验应用到AI数据中心建设中，这意味着其扩张速度有望大幅超越传统数据中心运营商。
Musk said building data centers is far simpler than building rockets, and the company is applying engineering know-how accumulated from rocket manufacturing to AI data center construction — meaning its expansion pace could significantly outstrip traditional data center operators.
在卫星互联网业务上，他预计星链的商业业务规模最终将远超消费者业务，并有望承载全球大部分互联网连接需求。
On the satellite internet business, he expects Starlink's commercial segment to eventually far exceed its consumer segment in scale, and to carry the majority of global internet connectivity demand.
对于AI的发展节奏，马斯克依然保持高度乐观：他预计到2027年底，AI将能够完成数字世界中几乎所有类型的工作；而随着自动驾驶汽车、人形机器人等物理AI形态的快速普及，全球对网络连接的需求也将迎来爆发式增长。若AI、机器人、自动驾驶与卫星互联网同步进入高速增长阶段，那么算力、数据中心、卫星通信与高速连接，将共同构成下一轮AI基础设施扩张的核心引擎。
On the pace of AI development, Musk remains highly optimistic: he expects that by the end of 2027, AI will be capable of handling nearly all forms of digital work, and as physical AI forms — autonomous vehicles, humanoid robots — see rapid adoption, global demand for network connectivity will see explosive growth. If AI, robotics, autonomous driving, and satellite internet all enter high-growth phases in sync, compute, data centers, satellite communications, and high-speed connectivity will together form the core engine of the next wave of AI infrastructure expansion.
数据中心只是开端：真正的AI基础设施竞争正走向太空
Data Centers Are Just the Start: The Real AI Infrastructure Race Is Moving Into Space
马斯克提出了一个颠覆传统认知的观点："与火箭和星链卫星相比，数据中心显得微不足道。"
Musk has put forward a view that challenges conventional thinking: "Compared with rockets and Starlink satellites, data centers are insignificant."
这句话的重点并非否定数据中心的价值，而是他认为，未来AI竞争的终局不会仅仅是建设更多地面数据中心，而是谁能够掌握更大规模的算力、能源、通信网络以及真实世界数据。
The point isn't to dismiss data centers' value — rather, his view is that the endgame of future AI competition won't simply be about building more ground-based data centers, but about who can command larger-scale compute, energy, communication networks, and real-world data.
目前，大多数AI公司走的路径是：采购GPU，租用云算力，建设数据中心，训练模型。但马斯克旗下的生态体系正尝试走出另一条路：火箭负责将基础设施送入轨道，星链负责全球高速连接，SpaceX负责大规模制造能力，AI计算中心负责训练模型，特斯拉负责收集真实世界数据，Optimus与自动驾驶负责AI的实际落地——这是一条从"算力"延伸到"现实世界"的完整闭环。
Today, most AI companies follow the same path: buy GPUs, rent cloud compute, build data centers, train models. But Musk's ecosystem is attempting a different route — rockets deliver infrastructure into orbit, Starlink provides global high-speed connectivity, SpaceX provides large-scale manufacturing capability, AI compute centers train the models, Tesla collects real-world data, and Optimus plus autonomous driving handle real-world AI deployment. It forms a closed loop extending from "compute" all the way to "the physical world."
马斯克多次强调，未来AI最大的瓶颈不只是芯片，而是能源、算力规模与数据获取能力。传统数据中心受限于土地、电力、散热与建设周期，而太空算力的长期想象空间在于：太阳能资源更丰富，散热环境不同，能够突破地面能源限制，并可以建立全球化的计算网络。
Musk has repeatedly emphasized that the biggest bottleneck for future AI won't just be chips, but energy, compute scale, and the ability to acquire data. Traditional data centers are constrained by land, power, cooling, and construction timelines, whereas the long-term promise of space-based compute lies in more abundant solar resources, a different cooling environment, the ability to bypass ground-based energy constraints, and the potential to build a globally networked compute system.
这也是SpaceX正在加速扩张AI算力的原因之一。马斯克表示，公司正大规模增加算力资源，并计划依托英伟达平台构建AI计算基础设施。未来的SpaceX或许不再只是一家火箭公司，而可能演变成连接AI算力、卫星通信、太空基础设施与全球数据网络的新型科技平台。
This is one reason SpaceX is accelerating its AI compute buildout. Musk said the company is significantly ramping up its compute resources and plans to build its AI computing infrastructure on the NVIDIA platform. Going forward, SpaceX may evolve beyond a rocket company into a new kind of technology platform connecting AI compute, satellite communications, space infrastructure, and a global data network.
目前全球AI竞争主要集中在几个方向：
The current global AI competition is largely concentrated across a few fronts:
NVDA — GPU与AI芯片
NVDA — GPUs and AI chips
MSFT — 云计算与企业AI
MSFT — cloud computing and enterprise AI
AMZN — AWS算力基础设施
AMZN — AWS compute infrastructure
GOOGL — AI模型与TPU
GOOGL — AI models and TPUs
而马斯克押注的是另一种模式：掌握从能源、硬件、通信网络到真实世界数据的完整AI供应链。这也是为什么市场正在重新评估SpaceX的长期价值——未来AI的赢家，可能不仅是拥有最强模型的公司，而是拥有最大规模算力、最多真实数据以及最完整基础设施闭环的平台。SPCX所代表的，正是这种"AI加太空基础设施"的长期叙事。
Musk, by contrast, is betting on a different model: controlling the entire AI supply chain, from energy and hardware to communication networks and real-world data. This is why the market is beginning to re-rate SpaceX's long-term value — the eventual winners of the AI race may not simply be the companies with the strongest models, but the platforms with the largest compute scale, the most real-world data, and the most complete infrastructure loop. SPCX represents precisely this long-term "AI plus space infrastructure" narrative.
ASTS：竞争格局悄然生变
ASTS: A Quiet Shift in the Competitive Landscape
据披露，Starlink Mobile预计将于2027年底推出，届时将使用EchoStar的频谱资源，并搭载在现有的星链屋顶天线上运行蜂窝无线电，直接与EchoStar展开竞争。
According to disclosures, Starlink Mobile is expected to launch by the end of 2027, using EchoStar's spectrum and running cellular radios installed on existing Starlink rooftop antennas — putting it in direct competition with EchoStar.
这一变化反而提升了ASTS对电信运营商的战略价值：由于T-Mobile的卫星合作伙伴星链正逐渐转变为竞争对手，ASTS依然是50多家运营商可以选择的独立合作方。
This development, in turn, raises ASTS's strategic value to telecom carriers: as T-Mobile's satellite partner Starlink increasingly turns into a competitor, ASTS remains an independent option available to more than 50 carriers.
