发布会·中文精读

马斯克公布SpaceX Flight 15计划及2026 Q3财务与运营业绩 [10/20/2026]

🇺🇸 Elon Musk Unveils SpaceX Flight 15 Plans and Q3 2026 Financial & Operational Results [10/20/2026]

中文精读由 AI 基于原始视频与文稿改写,非官方翻译,可能存在偏差或错误;表述以原视频为准。说明

马斯克在电话会快结束时扔出一个数字:到 2027 年底,SpaceX 的年度经常性收入会落在 5000 亿到 1 万亿美元之间。

同一场会的前半段,首席财务官 Bret Johnsen 报的是另一个数:第三季度营收 146 亿美元。

从单季 146 亿,到一年 5000 亿以上的经常性收入,中间只隔了大约十二个月。这个跨度让总裁兼 COO 格温·肖特维尔(Gwynne Shotwell)当场只能说一句"在这个疯狂的数字上",马斯克接了句"我不想让 Bret 心脏病发作"。

这场 2026 年 10 月 20 日的财报会,直播地点选在得州巴斯特罗普的星链工厂,没放在火箭工厂。选址本身就把重心交代清楚了。

一年之内,从亏 11 亿到赚 44 亿

先把财务底子摆清楚。第三季度营收 146 亿美元,比去年同期的 56 亿涨了 161%。净利润 44 亿美元,去年同期是净亏损 11 亿,一年翻转 55 亿。调整后 EBITDA 88 亿美元,同比涨 306%。

三个分部里,太空业务 18 亿,连接业务 52 亿,超级智能(Super Intelligence,以下简称 SI)76 亿。

SI 一个分部就超过了另外两个加起来。它上个季度只有 26 亿,一年前只有 9 亿。决定这条曲线形状的,是电力接入的速度,产品本身反倒是次要变量。

再看几个容易被忽略的指标。全公司第三季度资本支出 196 亿美元,第四季度指引直接抬到 350 亿到 400 亿。截至 9 月 30 日积压订单 630 亿美元,其中 175 亿已确认为递延收入,约 420 亿预计在一年内确认。2027 年第一季度的营收指引是 330 亿到 370 亿。

1000 亿美元的年度经常性收入目标,Bret 说按 12 月单月年化计算,年底前能到。

Flight 15 想一次抓住两截

Starship 的第十四次飞行把飞船送进了轨道,第一次部署了可用的星链 V3 卫星。这是马斯克口中的关键里程碑。

下个月要飞的第十五次,目标更狠:同时用发射塔的机械臂接住助推器和飞船。

如果成功,这将是有史以来第一次实现轨道级运载器的完整回收。这算得上巨大的技术里程碑。(马斯克)

他给了个概率判断:比不成功更可能成功,但首飞没有保证。就算第十五次砸了,年底前还有一连串试飞,明年加速。

FAA 那边,格温说沟通富有建设性,第十五次飞行的许可证预计会包含返回发射场和双塔捕获。

为什么这件事值得单开一段?马斯克自己解释了:

我们要把意识持续扩展到地球之外,唯一的办法就是完全可复用的轨道运载器。所以我把它看成文明史和意识史上一次深刻的技术里程碑。(马斯克)

他补了一句意思相近的话:大多数人没有意识到它有多重要,但事后回看会一目了然。

节奏上,明年上半年目标是每周一发,下半年每周两发,年底前可能更密。完全复用先会有少量翻修,最终要做到像飞机一样,着陆就能再起飞。

路易斯安那的太空港也公布了。马斯克预计那里最终会有 20 座以上的发射塔,每年向轨道及更远处运送数百万吨质量。首飞内部目标是 2028 年,2029 年大概率规模化发射。

半载荷的 Starship,等于十次猎鹰 9

Flight 14 部署的 26 Tbps 星链带宽,相当于十次猎鹰 9 发射的总和。而那次飞行大约只装了一半载荷。

三年内的规划是:星链星座总带宽提升约 100 倍。路径是双重的,单颗卫星带宽提升一个数量级,卫星数量也提升一个数量级,大概 10 万颗,每颗带宽超过现有系统的十倍。

用户侧的体验拐点在哪儿?马斯克回答得很实在:需要几百颗 V3 卫星,用户才会真正感觉到变化。卫星太少的时候,你会体验到几分钟的高速,十分钟后就没了。这个节点他判断在明年年中。

连接业务的数字:星链消费者订阅 1360 万(格温的口径是"接近 1400 万"),第三季度单季净增 160 万,覆盖 174 个市场,新开越南和乌干达。消费者 ARPU 环比涨 5% 到每月 69 美元。连接业务整体毛利率 55%,其中约 3 个百分点来自关税退款的一次性收益,剔掉之后调整后 EBITDA 利润率同比扩张约 2 个百分点。

企业和政府这块同比涨 91%,环比涨 18%。星盾(Starshield)对美政府业务在放量。

航空是最有画面感的一块。星链已经签约约 1.2 万架飞机,接近可用机队总量的四分之一。美国四大航空公司签了三家。汉莎、瑞士、Copa、大韩、奥地利、韩亚都在本季度飞了首班配备星链的飞机。航空终端的产能年初以来翻了四倍,年底前还要再翻一倍。

格温提到的数据很有说服力:装了星链的航班载客率更高,全程联网正在变成乘客的基本预期,加分项这个说法已经不成立。

手机直连这边,星链移动月活超过 2200 万,环比涨 37%。FCC 已批准发射 1.5 万颗 mobile V2 卫星,公司还宣布收购一组低频段频谱,为在美国成为主流移动运营商铺路。服务开通时间点指向 2028 年初。

带宽需求会涨一千倍,因为 agent 比人多

这块是整场会里最容易被跳过去、但最重要的一段判断。

我认为三到五年内,带宽需求可能是现在的一千倍。因为 AI 或者超级智能 agent,单个 agent 用的带宽远超一个人,而 agent 的数量会远远超过人的数量。(马斯克)

他的推理链条很干净。今天的通信网络是为几十亿人设计的,人类用带宽的方式有物理上限:眼睛能看多少、耳朵能听多少。agent 没有这个上限,它们的数量可以指数级增长,每个都在不停地互相通信、调用工具、交换上下文。

格温说了类似的话:连接汽车、机器人,加上超级智能的普及,通信需求的增长按几个数量级算,百分比这个单位不够用。

这就解释了一个看似矛盾的现象:SpaceX 一边说星链带宽三年涨 100 倍,一边说需求涨 1000 倍。供给追不上需求,成了这家公司所有投资决策的前提。

算力这件事,他们用造火箭的班子来做

马斯克把话题从太空拉到孟菲斯。他个人每周有四天泡在那里,昼夜不停地推 Colossus 2 集群上线。

节奏是:今年剩下的每个月,上线约 20 万块 GB300,明年保持这个速度,主要换成 Vera Rubin。到 2027 年,他有信心新增 10 吉瓦算力。今年年底前会有 2.5 到 3 吉瓦。

这么说吧,如果到明年年底我们上线的算力少于 13 吉瓦,我会感到意外。(马斯克)

他还说,公司明年的目标其实是新增 20 吉瓦的电力与冷却基础设施,让电和冷却的能力远超点亮 GPU 所需。所以明年年底算力超过 20 吉瓦,概率不为零,虽然目前还小。

"以前是零,现在不能再算零了。"这句话他说得很轻,信息量最大。

训练侧,明年第二季度末会上线最强的训练集群:44 万块 Vera Rubin,专为训练优化。

为什么他们能跑这么快?马斯克给了个类比,好笑,但信息量足:

这支团队拿下了地球入轨质量将近 90% 的市场份额,中国大约 6%,世界其他地方加起来包括美国其他公司大约 4%。这支团队做到的事情,是所有太空竞争对手加起来的一个数量级。然后我们把同一支团队放到地面算力的部署上。这有点像纽约洋基队去打少棒联盟。不公平。(马斯克)

说完他补了一句"让这个信息沉淀一会儿",熟悉他的人知道这是老梗。

真正的瓶颈在涡轮叶片的铸造车间

被问到怎么解决电力、冷却和基础设施约束时,马斯克没有讲采购,讲的是铸造。

燃气轮机的供应商告诉他:叶片和叶轮(blades and vanes)的交付,比整机其他部件落后 12 到 18 个月。除了叶片和叶轮,什么都能提前一年半给你,唯独这两样卡在那里,而且情况还在恶化。

SpaceX 恰好有一项别人没有的能力。火箭发动机对铸造的要求,超过地球上任何其他应用。马斯克的说法是,他们拥有地球上最先进的铸造能力,而且是客观上。

于是逻辑就通了:让这帮造火箭发动机涡轮泵的人,顺手铸一批地面燃气轮机的叶片和叶轮,比他们过去干的活简单得多。

这意味着我们能比任何人快 12 到 18 个月把电送上网。这种优势很难复制,除非别人也去建一个先进铸造车间。这块他们没法跟 SpaceX 竞争。(马斯克)

他还提到另一个方向:把推理算力做成模块化产品,电和冷却在工厂里就集成好,然后像集装箱一样分发到全国各地,甚至全世界。

至于具体怎么做到 20 吉瓦,他不肯说。这是高度专有的东西,一讲出去,别人就会照着做。

回本不到一年,H100 越老越贵

Bret 给了一组很硬的数字:按当前定价,新部署的算力回本周期不到一年。算力价格目前在每瓦 30 美元区间的上沿之上,比上个季度讨论的还高。

马斯克补了一层解释,这部分值得细看,因为它反常识。

我们相信每瓦算力的价值会随时间上升,不会下降。原因应该很明显:AI 软件变强,模型变好,每瓦产出的有用智能就更多,而算力的价值与每瓦有用智能成正比。像 H100 这种老机器,租金比它刚出厂时还贵,原因也在这里。(马斯克)

通常硬件贬值,这里升值。这个资产卖的东西已经超出晶体管本身,卖的是智能。模型每提升一档,同一块芯片的产出就贵一档。

基于这个判断,SpaceX 的策略是尽量做短期合约,按现货市场价格出租,不签长约。内外部算力大致五五开,可能 60/40 浮动。

"我们未来会长期面临超级智能算力短缺。"他的结论就这一句。

奶酪店

为了说明算力有多紧,马斯克讲了个段子。他发过一个蒙提·派森的"奶酪店"小品视频:顾客进店想买奶酪,店员说没有;那这种呢?也没有;那种呢?还是没有。整家店一颗奶酪都没有。

他说 SpaceX 试过把一部分推理算力外包给其他云厂商,好腾出算力做训练。结果撞上的就是那家奶酪店。

"有 GB300 吗?"没有。"GB200 呢?"没有。"TPU 呢?"没有。"Trainium 呢?"没有。"那 H100 呢?"海外可能有一点。然后他们去查了一圈回来说,也没有。(马斯克)

他还说,每一个超大规模云厂商、几十家小公司都在追着他们要算力,"如果我们全答应,我们的算力可以租出去好几遍"。

他甚至给出了一个更激进的猜测:明年出现六个月的算力回本周期,不是不可能的事。

Terafab:名字就是目标

摩根士丹利的 Adam Jonas 问芯片。问题是:SpaceX 是否已经下单 EUV 光刻机?交付时间?从第一性原理出发,台积电、ASML、AMD、泛林这些公司哪些环节可以改?

马斯克没确认订单,但给了内部时间表:明年 5 月到 6 月,奥斯汀研究性晶圆厂做出第一颗可工作的测试芯片,大概是一颗 SRAM。真正量产的目标是 2028 年第三季度,在 Gibbons Creek 的主厂 Terafab。

"这些日期非常激进,我们也不知道能不能做到。"他特意加了误差范围,说 2028 年三四季度量产,至少在可能结果的集合里。他还感谢了 ASML 和其他供应商加速交付设备。

然后他留了个扣子:在芯片制造上,他们有一些潜在的、重大的突破,但不打算在财报电话会上讲,别把家底都抖出来。

至于名字,Terafab 的目标就是字面意思:每年生产一太瓦的超级智能芯片。

Grok bot 一个月翻倍,企业版几周内铺开

产品侧的数据:Grok bot 今年 8 月上线,10 月初付费周活 80 万,一个月翻了一倍多。B2B 是增长最快的类别,公司在销售、市场、IT、工程部门把它当常驻队友用,跑端到端的复杂流程。

马斯克对它的评价很直白:

Grok bot 是我这辈子见过增长最快的产品。而且我们看到单个用户的使用量随时间上升,不会下滑。我们让 Grok bot 变得更有用,用户每周就用得更多。(马斯克)

企业版几周内会大规模铺开,早期已经在一些信任的客户那里部署。马斯克预计,Grok bot 最终可能会消耗掉他们大部分超级智能算力。

Cursor 的整合也在推进。600 亿美元股权价值里,230 亿归到留人和替换既有股权激励上,包括核心高管的重新归属协议,分三年确认。剔除 12 亿股权激励费用后,SI 分部第三季度经营利润是 16 亿美元。年化收入超过 100 万美元的 Grok 企业客户(含 Cursor)按备考口径环比涨了 50%。

上周他们还宣布了一个新套餐,把 Grok、Grok bot、Cursor 和 X Premium 打包成一个订阅。

一场他不能参加的谈判

德意志银行的 Edison Yu 问了特斯拉合并。马斯克的回答值得完整看,因为它展示了这个体量的公司怎么做治理。

两家公司都应股东要求成立了特别委员会讨论这个想法。委员会由独立董事组成。因为我是利益相关方,我不能参与特别委员会的讨论。所以整个局面像一场奇怪的歌舞伎表演:两组人在谈,而我不在对话里。(马斯克)

他明确否掉了一种市场预期:有人认为合并是板上钉钉的事,这种说法夸大了。他自己怎么看?"我觉得是个好主意。"

但门槛在那儿:特斯拉这边需要全体股份的 50% 投票通过,参与投票者的 50% 不够,得是所有股东的一半。这个门槛相当高。

目前的进展是:特别委员会尚未就任何交易条款达成一致,也不保证会达成。

他认为合并有价值的地方列得很具体:超级智能软件的开发、芯片设计、Terafab、机器人、入轨质量与星链星座、连接业务。这份清单列的是未来十年要干的事,两家公司的重叠部分越来越长。

把这些拼起来看

这场财报会的信息密度高得离谱,但如果只抓一条主线,是这条:SpaceX 正在把自己从一家发射服务公司,改造成一家能源与算力基础设施公司,火箭只是它获取成本优势的手段。

证据链是闭合的。造火箭发动机练出的铸造能力,用来造地面燃气轮机的叶片,比同行快 12 到 18 个月上电。发射入轨练出的成本控制,用来把卫星产能拉高一个数量级。为火箭写的飞控和制造体系,用来在十几个月内把月产 20 万块 GPU 的集群点亮。

反过来,星链的现金流养 Starship,Starship 的运力养 V3 卫星和 10 万颗星座,星座的带宽养 agent 时代的需求,agent 的需求又反过来吃掉所有算力。

每个环节都在给下一个环节供血。这种结构在商业史上不多见,它的风险也在这里:任何一环卡住,整条链的节奏都要重排。

马斯克自己给的时间表是激进的:明年上半年 Starship 每周一发,明年年底 13 吉瓦算力,2027 年底 5000 亿到 1 万亿 ARR,2028 年 Terafab 量产,2028 年路易斯安那首飞。他在会上反复加限定词,比不成功更可能成功、至少在可能结果的集合里、小概率但不为零。

这些话属于校准,谈不上谦虚。他自己清楚,这套时间表里有一半的变量不在他手上。

看完这场会,真正该拿走的是他判断约束的顺序:不去解最显眼的那个问题(芯片),去找最不显眼但卡住所有人的那个(涡轮叶片铸造、电力设备、冷却)。这个顺序,比任何一条指引都更值得学。那些数字反而次要,它们只是这套顺序的副产品。

接下来盯什么

  • 盯 Flight 15 的 FAA 许可证文本,看是否明确写入返回发射场加双塔捕获。许可证范围比发射本身更能说明监管节奏。
  • 盯星链 V3 的在轨数量。马斯克给的体验拐点是几百颗,对应明年年中。在那之前,任何网速测试的结论都不可靠。
  • 盯每月实际点亮的 GPU 数量,别看 PPT 上的吉瓦数。一个月 20 万块 GB300 是能验证的硬指标,做到了就说明电力、冷却、网络、电池整条链都通了。
  • 盯电力设备的供应链,特别是燃气轮机的叶片和叶轮。全场真正的护城河只有一条:我知道怎么绕过去,别人绕不过去。
  • 读财报时先看两个数:积压订单的构成(多少是递延收入)、资本支出的季度跳升幅度。630 亿积压里 420 亿一年内确认,这个转化率决定了明年的营收下限。
  • 关注特斯拉与 SpaceX 特别委员会的正式公告。全体股份 50% 的投票门槛意味着,任何内部消息都不足以判断结果,只有委员会达成条款后的股东文件才算数。
  • 把 2027 年底 5000 亿到 1 万亿 ARR 当成方向而不是预测。它是从当前增速外推的上界,用来理解资源配置的优先级,不用来建模。
YouTube 平台字幕(自动或人工),可能有识别错误,人名与数字请以视频为准。已按语句合并成段落,并去掉 um、uh 等口头填充词。 文稿来源:YouTube(自动/人工字幕)。

0:04Elon Musk voice: In order to understand the nature of the universe and all these fundamental questions, we must expand the scope and scale of consciousness.

0:25Hello everyone. We are coming to you live from our Starlink factory and future Starlink production site here in Bastrop, Texas. I'm Andrea Williams, head of investor relations, and thank you for joining us for SpaceX's third quarter 2026 earnings presentation. Joining us are Elon, Gwynne, and Bret, and we are live streaming on X.

0:45Our third quarter financial results were announced just after 3:00 p.m. Texas time, and all relevant materials have been published to our investor relations website at ir.spacex.com. For today's presentation, our prepared remarks will be followed by a question and answer session.

1:00We will take live Q&A from equity analysts, and then move to questions directly from our investors that were polled on our dedicated earnings Q&A platform, which the SpaceX team built with Grok. During this webcast, we will discuss our business outlook and make forward-looking statements.

1:17These statements are based on our predictions and expectations as of today. Actual events or results could differ materially due to a number of risks and uncertainties, including those mentioned in our earnings release and the most recent filings with the SEC, including our final prospectus.

1:32During the presentation, we may discuss certain non-GAAP financial measures. In today's earnings release and our filings with the SEC, you can find additional disclosures regarding these non-GAAP measures, including all the reconciliations with comparable GAAP measures. For the third quarter of 2026, we announced the following financial

1:51results as compared to the prior year period. Revenues of $14.6 billion, up 161% from $5.6 billion, net income of $4.4 billion, an improvement of $5.5 billion from a net loss of $1.1 billion, and adjusted EBITDA of $8.8 billion, up 306% from $2.2 billion. With that, I'll hand the call over to Elon.

2:21Thank you. So, we are in an inflection moment for SpaceX across launch, connectivity, and super intelligence. Starting with Starship, flight 14 was a critical milestone. Starship made it to orbit for the first time and deployed operational Starlink V3 satellites. I think we have a recap video we're going to play.

4:37All right, for flight 15, which we expect to happen next month, we're going to attempt to catch both the booster and the ship. If we succeed, it'll be the first time that an orbital launch vehicle has been fully recovered. This is a massive technology milestone. The only way that we can extend consciousness beyond Earth on a sustained basis is with a fully reusable orbital launch vehicle.

5:25So, I think this is a profound technology milestone in the history of civilization and consciousness. So, let's look at what a simulation of flight 15 would look like if we succeed. If we don't succeed in flight 15, we'll get there. Success is not guaranteed on these first flights, but I'd say it's more likely than not that this succeeds next flight.

5:54And if not, we have a steady stream of Starship flight attempts through the end of this year and then accelerating next year. So, with that, the flight 15 simulation.

7:35We've also announced the development of the Louisiana spaceport. That's going to be an incredible asset to the country and to the world. That will, I think, at some point probably have on the order of 20 or more launch towers and enable the transport of millions of tons per year to orbit and beyond.

7:59So, that's really going to be a big deal. We expect Starship to fly weekly at some point in the first half of next year, and then second half of next year twice weekly, or maybe even more frequently than that by the end of next year. The rate at which we launch Starship will increase dramatically as we solve the

8:19rapid and complete reusability. So, we'll have complete reusability. Then at first, there will be a little bit of refurbishment that's needed between each flight, and we'll iron out the need for any refurbishment such that you can immediately land the rocket and launch it again just like an airplane.

8:37As I said, this is a profound technology milestone that most people in the world don't realize how important it is, but it will be obvious in hindsight. Starlink is also just getting started. Flight 14 deployed 26 terabits per second of Starlink bandwidth. That's equivalent to 10 Falcon 9 launches.

9:01So, a single Starship launch, which wasn't even fully loaded—it was about maybe half loaded—did the equivalent of 10 Falcon 9 launches. What I expect over time, not in the distant future, probably within three years, is that we will have about 100 times the bandwidth of the current Starlink constellation.

9:31Because we will have launched satellites with an order of magnitude more bandwidth, and we will place into orbit an order of magnitude more Starlink satellites. So, probably something on the order of 100,000 satellites, each of which have more than 10 times the bandwidth of the current Starlink system.

9:54I think revenue for Starlink will be at least somewhat proportionate to the total bandwidth delivered, and we will also have high-bandwidth direct-to-cell capability, and we will be building out our terrestrial networks so that there's broad-based connectivity anywhere that Starlink is allowed to operate.

10:15So, anywhere in the world that we're allowed to operate. And I think the demand for bandwidth is going to be dramatically higher than it is today. I think in three to five years, there's probably a demand for a thousand times more bandwidth than there is currently because AI or SI agents will, on a per-agent basis,

10:44use far more bandwidth than a person, and there will be far more agents than there are people. So, this ends up requiring a several order of magnitude increase in communication bandwidth in order to satisfy that communication demand. So, talking about SpaceX super intelligence, SpaceX SI sexy, whatever.

11:14As I think people are starting to realize, the constraint on deployment of super intelligence is more in power and cooling and infrastructure than it is in chips. What we've done at SpaceX, we've taken a team that has been able to achieve a market share in mass to orbit that is almost 90% of all mass to orbit of Earth, where China is roughly 6%,

11:48and the rest of the world combined, including the rest of the US, is about 4%. So, it's a team that has been able to achieve results that are an order of magnitude greater than all competitors in space combined, and applied that same team to the deployment of terrestrial compute.

12:08So, this is a bit like the New York Yankees playing in Little League. It's not fair. The net result is being able to deploy power and cooling infrastructure at a rate that is vastly faster than anyone else, and may at some point be faster than everyone else combined. So, just let that sink in for a minute.

12:38I've been spending personally a lot of my time in Memphis, often four days a week, night and day, just getting the Memphis supercomputer cluster Colossus 2 online. We're tracking to be able to bring roughly 200,000 GB300s online per month for every month for the remainder of this year and continuing that pace next year.

13:08With next year being primarily Vera Rubin compute. At this point, I would say I'm fairly confident of being able to bring an incremental 10 gigawatts of compute online in 2027. That would be in addition to the roughly 3 gigawatts, maybe 2.5 to 3 gigawatts, that we'll have online by the end of this year.

13:41Let me put it this way: I would be surprised if we brought on less than 13 gigawatts of compute by the end of next year. And as I mentioned earlier, power, cooling, and electrical equipment infrastructure is somewhat of a limiting factor. So, to make sure that that is not a constraint for our build-out, we're actually going to aim

14:08for an incremental 20 gigawatts of power and cooling infrastructure next year. So that the amount of power and cooling that we have is far in excess of what is needed to turn the GPUs on. So, there is a non-zero possibility—maybe it's a small possibility at this point, but it's not

14:35zero—of SpaceX having over 20 gigawatts of super intelligence compute online next year. A small possibility, but it used to be zero; it's not zero anymore. We're also planning to have the most powerful training cluster online by the end of Q2, which will be 440,000 Vera Rubin computers deployed for optimal training.

15:07This will, I think, be extremely helpful for getting Grok to be potentially the best model, but certainly to be extremely competitive. On the Grok side, we're seeing the fastest growth of a product that I've ever seen in my life before for Grok bot. I've never seen anything grow this fast.

15:40We're also seeing that the usage of Grok bot grows over time per user; it does not decline. So, as we make Grok bot more useful, we're seeing users use it more per week, not use it less. And we'll be deploying Grok bot at scale for enterprise customers, which is a very big deal.

16:03So, they can be used effectively in teams by companies as though it were a virtual human. We're already on early deployment with some trusted customers, and we expect to be in wide deployment within a few weeks for enterprise customers for Grok bot.

16:30I think Grok bot may end up being what uses most of our super intelligence compute.

16:49In a nutshell, what this implies for SpaceX annual recurring revenue by the end of '27 is somewhere between 500 billion and a trillion. Right? Actually, are you going next? Sorry. It's okay, Bret, you can go instead. Sorry, on that crazy note, I thought it'd be... On a high note?

17:17Yeah, it's a high note. Yeah, I don't want to give Bret a heart attack, you know. Thanks, Elon. So, we had an exceptional execution in the quarter again. And today, I'll focus both on the operational and the commercial highlights, as well as talk about the priority initiatives for next quarter.

17:41Showing a slide with the KPIs, and I'll go ahead and walk through it. Starting with our space business, we remain the leading launch provider for the world, flying the vast majority of payload mass to orbit. During the third quarter, our 35 launches included 21 flights of Starlink on Falcon 9, two flights of

18:01Starship, a Falcon Heavy which I'll talk about in a second, and 12 high-profile customer missions. Including the Falcon Heavy launch, which we flew NASA's Roman Space Telescope into deep space. We also flew six national security space launch missions. We've also secured important new business, winning three additional human spaceflight missions on Dragon, which will take us through 2028,

18:29along with the Starlink communications contract on NASA's Artemis 3 mission. In addition, we signed 18 new national security space launch missions, a contract to deliver payload to the lunar surface with Starship, and three new Starfall missions on Starship. We just did a Starfall mission today, by the way, on Falcon 9.

18:51Starfall is our space capsule, which provides routine relatively low-cost access to microgravity environments for scientific research, including improving our understanding of diseases like cancer and diabetes, and developing better protocols for patients here on Earth. During the final week of the quarter, we flew Starship flight 14 to orbit as Elon

19:14talked about, followed just after on October 1st by a triple header. We flew three Falcon missions across our three Falcon launch sites in just 13 hours. Crew-13, we brought safely to the International Space Station, a Falcon Heavy NRO mission, and a rideshare mission flying 130 payloads.

19:35It's an incredible display of what we can deliver in a single day across crew, commercial, and national security missions. We're also incredibly excited for what's ahead for Starship, flying the largest payloads to help build the space economy in orbit and on other planets and the moon.

19:53As Elon said, Starship's flight 14 was a major technical milestone, and we remain on track with our goals to bring back both the first and second stages to the tower this year, refly the second stage by early next year, and ramp our launch cadence as we make progress towards rapid reusability.

20:12We'll attempt an on-orbit refueling early next year, and we'll begin flying mobile V2 and Starlink satellites in 2027. We have had significant constructive engagement with the FAA, and we expect the license for our flight 15 to include return to launch site and tower catch for both stages.

20:32I'm also extremely excited about expanding the Starship operations to Louisiana with the second Louisiana Purchase. The geography of Louisiana, the community, and their work ethic is so aligned for an amazing Starship future. Moving on to Starlink, we had another standout quarter driven

20:52by continued growth across our consumer, enterprise, and government businesses. We ended the quarter active in 174 markets, including new market launches for Vietnam and Uganda. Looking at subscriber growth, we added 1.6 million consumers—not total users, but consumers—in the quarter, reaching just under 14 million global broadband consumer subscribers.

21:16As I mentioned, this number does not include additional hundreds of thousands of enterprise subscribers and aviation customers. We remain incredibly inspired by the impact of Starlink on communities around the world, including providing service to hundreds of thousands of students and teachers in schools

21:34globally, as well as for emergency response during natural disasters. Over the third quarter, Starlink's rapid response capabilities enabled support across 25 crises, including the Venezuela earthquake, the earthquake in western Colombia, and Hurricane Lowell in Hawaii. In Venezuela alone, Starlink donated over 1,400 kits to response agencies,

21:59providing critical Starlink broadband connectivity to over 250,000 people. As we mentioned last quarter, we continue to see significant growth opportunities on the enterprise and government side of the business, with aviation a continuing major focus and fixed site gaining momentum.

22:16Starlink is now contracted for roughly 12,000 planes, almost a quarter of the available aircraft, and we're expanding quickly with both new and existing customers. We signed agreements with three of the four major US carriers, along with a growing list of leading international airlines.

22:35During the quarter, Lufthansa, Swiss, Copa Airlines, Korean Air, Austrian Airlines, and Asiana Airlines all flew their first Starlink-equipped aircraft. To keep pace with this demand, we've quadrupled aviation production rate since the beginning of the year, and we expect to double before the end of the year.

22:54Our aviation partners report that Starlink has become a key reason customers choose to fly with them, both business travelers who increasingly rely on full connectivity in the skies or even individuals who want to stay connected all the time everywhere. Our data shows, and frankly our customers tell us, that Starlink-equipped flights carry higher

23:14load factors on the planes, and full connectivity is quickly becoming table stakes for passengers. On the government side, Starshield continues to deliver unique mission-critical capabilities, and we see further room for growth. We remain on pace for enterprise and government revenues to exceed

23:33consumer connectivity revenues in the future each. The successful deployment of the operational V3 broadband satellites on Starship's flight 14 marks an important step and an inflection point for Starlink. The increased capability from these satellites, as Elon talked about, and the Starship launcher

23:50itself will provide additional capacity across markets and enable new sales in high-demand areas. We expect the market for communications to grow far beyond just human communication that we traditionally think about. Connecting cars, robots, and the adoption of super intelligence

24:09will increase communication demands not in percentages, but in many orders of magnitude. In the years ahead, we expect Starlink alone to represent a significant portion of global internet traffic. In mobile, we continue to expand international partnerships, turning on service with new carriers across multiple countries in the third quarter,

24:32including Uganda, the Democratic Republic of Congo, Kazakhstan, and Costa Rica. We surpassed 22 million monthly users on Starlink mobile as of the end of the quarter, up 37% quarter-over-quarter. Our near-term priority remains flying the mobile V2 constellation through next year

24:50and integrating our spectrum purchases to turn on service in early 2028. Just a few weeks ago, we received approval from the FCC to launch 15,000 mobile V2 sats and announced an agreement to acquire a portfolio of low-band spectrum, which will pave the way for Starlink mobile to become a major mobile carrier in the US.

25:12Moving to super intelligence, the industry continues to demand ever more massive amounts of compute. We're rapidly expanding capacity to meet both our own needs as Grok and related products scale, and those of our cloud service customers. Enterprise feedback on recent model releases has been positive, and Grok bot in particular is seeing strong adoption out of the gate.

25:35Grok bot, our agentic assistant, launched in August, and we already had paid weekly users of 800,000 earlier this month, more than doubling over the last month. B2B is our fastest growing category, with companies increasingly using Grok bot as a force multiplier across sales, marketing, IT, and engineering,

25:55treating it as an always-on teammate for complex end-to-end workflows. We're also accelerating our Grok model release cadence, with significant performance gains expected in the near term. At the same time, the integration of Cursor continues to progress well, strengthening both

26:13our product capabilities and our go-to-market reach with a scaled enterprise sales force. Early evidence of this momentum is already visible. Grok enterprise customers with more than 1 million in run rate revenue, including Cursor, were up 50% quarter-over-quarter on a pro forma basis.

26:31Last week, we further integrated the portfolio by announcing a new tier that brings together Grok, Grok bot, Cursor, and X Premium on a single plan. In review, our business is experiencing extremely strong momentum across all three segments. Our engineering teams are scaling Starship and next-generation satellites across broadband,

26:52mobile, and Starshield, while our commercial teams expand Starlink across aviation, enterprise, government, and global markets. At the same time, we're scaling our super intelligence business in the key areas of critical compute infrastructure, leveraging our hardware expertise.

27:09Now, Bret, you get to go. Thank you, Gwynne. We delivered another strong quarter in Q3. Revenue grew an impressive 161% year-over-year to $14.6 billion, fueled by the continued momentum in the connectivity business and our clear inflection point in the super intelligence or SI business.

27:30Adjusted EBITDA once again grew faster than revenue to $8.8 billion, up 306% year-over-year. The combination of our launch leadership, recurring Starlink revenue, and the significant expansion of our SI infrastructure business have combined to drive meaningful operating leverage in the business.

27:50Our balance sheet remains very strong, and our focus on capital allocation continues to keep us in a position to accelerate the pace of innovation that underpins the significant revenue opportunity across all of our businesses. I'll now walk through the financials by segment. Space revenues of $1.8 billion were up 84% sequentially and up 6% year-over-year,

28:14anchored by a favorable customer mix. We're on track to deliver strong growth for the year based on a robust customer manifest for Falcon and Dragon vehicles. As far as capital allocation, we continued to invest in our Starship program, accelerating Raptor engine and vehicle production, advancing gigafactory construction, and progressing multiple launchpads at both Starbase and the Cape.

28:38And importantly, we secured the location for the next phase of United States space launch leadership in Louisiana. These investments remain essential to unlocking the step change in cost and capacity that Starship will deliver. Connectivity again delivered robust revenue growth and profitability,

28:58with revenues of $5.2 billion, up 69% year-over-year and 22% sequentially. As Gwynne mentioned, we reached 13.6 million Starlink consumer subscribers, adding 1.6 million subscribers just in Q3 alone, driven by growth in more developed markets. During the quarter, we began ramping production of our V5 kits, which supported our progress on alleviating inventory constraints.

29:26We strategically managed our channel distribution and prioritized delivery of kits to retail channels ahead of the holiday season, which we expect will lead to increased activations in Q4. Starlink consumer monthly average revenue per user, or ARPU, increased 5% sequentially to $69

29:46per month, benefiting from a favorable customer mix dynamic in the quarter. Enterprise and government revenues grew 91% year-over-year and 18% sequentially, highlighted by growth in our Starshield program with the United States government and the continued momentum in the aviation industry.

30:04Aviation remains a major growth opportunity as we're still early in the installation phase with many of the largest carriers worldwide, and have line of sight into an accelerated activation path into '27 for our key partners. Connectivity gross profits doubled year-over-year with gross margins of 55%,

30:27including a one-time benefit of roughly 3% primarily driven by tariff refunds in the quarter. Excluding the one-time benefits though, adjusted EBITDA margins expanded roughly 2% year-over-year. Looking ahead, the satellite roadmap is a key driver of both the capacity and capability of our connectivity segment.

30:49Through the end of the year and throughout 2027, we expect to deploy V3 broadband satellites in meaningful volume, with each Starship launch ramping towards a payload that delivers over 20 times the downlink capacity of the current generation of Falcon 9 launches today. In parallel, we will ramp Starlink mobile V2 satellites on Starship starting in 2027 and

31:12leverage our newly enhanced wireless spectrum portfolio to expand our direct-to-cell service and become a major mobile carrier, as Gwynne mentioned, in the United States. Across all three of our connectivity applications—Starlink broadband, Starlink mobile, and Starshield—the launch cadence on Starship is expected to drive a

31:31step change in total capacity, coverage, and service quality. SI, or super intelligence, was again the standout growth driver for the quarter. Segment revenue increased substantially to $7.6 billion, up $5 billion quarter-over-quarter and $6.7 billion year-over-year, primarily powered by the ramp of our cloud services agreements.

31:56We meaningfully improved the segment's profitability profile as we monetize newly deployed capacity and delivered powerful operating leverage for the business. We ended the third quarter with 1.7 gigawatts of nameplate compute capacity, with additional data halls already online in early October.

32:13For SI solutions and infrastructure, revenues were $7 billion, of which $608 million came from the Cursor acquisition closed roughly mid-quarter. Even excluding Cursor, SI solutions and infrastructure revenues grew 192% sequentially, driven primarily by the activation of new cloud services agreements.

32:34There are many customers asking for compute, and we're actively building the supply to meet both internal and external demand. We expect to have additional new customers in future deals, and we'll continuously evaluate the highest margin usage for our compute assets across both new and existing cloud service agreements and serving inference for our own SI products.

32:59What we're seeing in the market is extraordinary. At current pricing, new compute deployments are generating a payback of less than a year. Pricing is currently tracking slightly above the high end of the $30 per watt range we were discussing just last quarter, which further supports the attractiveness of these terrestrial SI compute investments.

33:21We'll continue to be demand-led and prudent in the pace of deployment, matching our disciplined and proactive capital planning for procurement of semiconductor supply to the near-term customer demand. Our current customer contracts for SI compute have the benefit of allowing us to flexibly manage our

33:39capacity for our own model training purposes, as well as be able to monetize for our SI cloud service agreements revenue at favorable pricing as the market availability continues to tighten. Advertising revenues grew 19% year-over-year as we build more momentum on the new advertising tech platform that we rolled out just last quarter in Q2.

34:05SI segment adjusted EBITDA was $5 billion, up $3.8 billion quarter-over-quarter and $5.4 billion year-over-year, driven primarily by higher margin cloud service agreements. SI segment operating income was $431 million, but excluding the $1.2 billion of stock-based compensation expense related to the Cursor acquisition, the segment delivered $1.6 billion of operating profit just in Q3.

34:32Double-clicking on the share-based compensation expenses of the 60 billion equity value for Cursor when we purchased it, 23 billion was attributed towards retaining key employees and replacing outstanding equity awards, including re-vesting agreements for key executives. These equity awards will be recognized as stock-based compensation over roughly the next three-year vesting period.

34:54Total company capital expenditures in the third quarter were $19.6 billion, primarily in support of our SI compute buildout, Starship infrastructure, and satellite production. We ended the quarter with substantial backlog totaling $63 billion as of September 30th, of which $17.5 billion was recognized as deferred revenue.

35:16Approximately $42 billion of this backlog is expected to be recognized within the next year. Looking ahead, demand remains robust across all three segments, particularly in our cloud services business. We remain on track to deliver our previously stated goal of 100 billion dollars in ARR by

35:37the end of this year, defined as annualized revenue based on the month of December. SI infrastructure continuing to come online at Memphis and Southaven gives us visibility into an accelerated build path through the remainder of this year. As a result, we're on track to reach nearly 2.2 gigawatts of nameplate compute at the

35:57end of this month alone in October, and have a clear line of sight, as Elon highlighted, to nameplate compute in the range of 2.5 to 3 gigawatts just by the end of this quarter. Because of the additional capacity coming online this quarter, we expect fourth quarter capital expenditures to now be in the range of 35 to 40 billion for this quarter.

36:17We also expect total company gross margins in Q4 to be more consistent with the second quarter, given some one-time benefits and customer mix that we highlighted in Q3. Given the increased visibility into our compute contract pipeline over the next six months, we have the opportunity to provide additional forward guidance on the

36:38acceleration we see in our business in Q1 2027. We now see total company revenues in the range of 33 to 37 billion dollars for Q1 of 2027. In closing, we are in a transformative phase for the company's financial profile. We're investing aggressively across SI compute, Starship development and production,

37:00and next-generation Starlink broadband and mobile constellations—the three areas we believe will generate the highest long-term returns. These are demand-led investments in markets with visible, tangible returns in the near term. We're building the infrastructure of the future with discipline, focused on capital efficiency,

37:20strong returns on invested capital, and maintaining a strong investment-grade balance sheet, while still moving with the velocity to capture the opportunity in front of us. The combination of our diversified earnings base, focus on extreme vertical integration, and the robust demand in the market for super intelligence gives us conviction in the path ahead.

37:41With that, I'll turn it back to Andrea. Thanks, Bret. We're now going to begin the Q&A section. If you'd like to ask a question, please use the raise hand function at the bottom of your screen. If you dialed in today, please press star nine to raise your hand and star six to unmute yourself.

38:03I think we've assembled a queue. We will take the first question from Adam Jonas at Morgan Stanley. Hey, everybody, can you hear me? Yep. Great. First, Elon, thank you for making flying cars real. You're welcome. That was freaking insane, it's my first comment. Yeah, there needs to be things in this world that make you glad to be alive, and that's one of them.

38:30It's super inspiring for people all over the world, especially young people that want to build. Yeah. I got a two-part question on Terafab. You want to make chips, you want to make crazy amounts of chips onshore or in a safe and secure place. Yeah. So, first, can you confirm if SpaceX has placed orders for EUV machines?

38:52If so, when do you expect delivery? And just as a second part to the question, when you take a first principles approach to the semi fab supply chain, I know you've toured TSMC, ASML, AMD, Lam. These are very impressive companies, as I'm sure you'd agree. Where do you see the biggest areas to improve their process

39:15and supply chain when you do it your way? Thanks. Well, I mean, we're just figuring out chip fabrication, so... I'll tell you what our internal expectations are, which obviously I'd put big error bars on this because we don't know if we'll actually achieve this or not, but that's what we're aiming for internally,

39:40is to make our first sort of test chip functional in approximately May or June of next year. That's just a kind of SRAM chip or something to basically test the line that's at the research fab in Austin. Our aspiration is to reach volume production around Q3 of '28 at Terafab at Gibbons Creek,

40:06the main Terafab location. Now, these are very aggressive dates, and we're very appreciative of ASML and all of our other suppliers in accelerating equipment deliveries in order to achieve this outcome. Let's just say that that outcome of achieving volume production is at least in the set of possible outcomes for Q3 or Q4 of '28.

40:33These are obviously insanely fast timeframes for chip production. We do have, I think, some significant potential breakthroughs in chip fabrication, but I don't think I should discuss that on an earnings call. Let's not give away the shop here. And I am optimistic that we will dramatically accelerate

40:59chip production, and that Terafab will ultimately do what the name suggests, which is to make a terawatt of super intelligence chips per year. We'll take our next question from Doug Anmuth at J.P. Morgan. Great, thank you. Elon, can you talk more about just how you're working through those constraints in power,

41:35cooling, and infra, and just what gets you comfortable with the 13 gigawatts ending '27? And then Bret, if you could also help us understand how you're thinking about the funding requirements just as you look out to that kind of capacity. Thanks. Yeah, well, close to 3 gigawatts is what we expect to have done just in the next few months.

42:00You can just look at the rate at which we are actually deploying and turning on compute. What I think you hear from other hyperscalers is you don't hear a lot of detail about how many GPUs they're actually turning on every month. And with us, you are. I'm being as precise as possible about this.

42:23It is absolutely insane to be providing the power and all of the electrical equipment, the cooling, the batteries, the networking, and everything else that's needed to turn on 220,000 GB100s a month, but that is actually what we're doing right now. And that pace will obviously accelerate next year; we're not going to slow down.

42:51If you say, "How are you going to get all this power up?" let's just say we actually do have a plan that gets to 20 gigawatts of power and cooling by the end of next year, but that is not something that I should spell out. That is highly proprietary because if I tell you about it, others are going to try to do the same thing.

43:10And I'm not saying we'll have 20 gigawatts of power. I'm saying we have a plan where if everything goes right, we'll have 20 gigawatts of incremental power, I should say, next year, so 23 gigawatts roughly in total. Now, I don't expect everything to go right; that's why I said I'm highly confident of 10 next year.

43:30And 20 is difficult, but not impossible. Some of the things we're doing I have mentioned publicly. So, we're casting the blades and vanes for the turbines. SpaceX already has a very advanced castings capability that was developed for rocket engines. We have, I think, actually the most advanced casting capability on Earth.

44:06Because the demands for castings for rocket engines exceed the demands for everything and anything else you could possibly imagine. So, taking that incredible team that has, I think objectively, the most advanced casting technology in the world and saying, "Hey, can you guys also make some

44:24blades and vanes for terrestrial turbines?" is much easier than what they've done in the past. And when we talk to the turbine suppliers, they tell us that the blades and vanes are 12 to 18 months behind the rest of the turbine. Meaning they can deliver everything except the blades and vanes 12 to 18 months before they can deliver the blades and vanes.

44:52So, casting is this really difficult bottleneck, and if anything, it seems to be getting worse. So, that means that we can bring power online 12 to 18 months faster than anyone else. That's the kind of advantage that's difficult to replicate unless someone else is building an advanced casting shop.

45:15They will be unable to compete with SpaceX in this regard. And we're looking at many different sources of power, not just gas turbines. Obviously, we want to use solar, wind, and every other source of power that we can obtain. We think there's also a lot of opportunity for inference compute if you manufacture

45:42modular inference compute at the factory level, such that the power and cooling are built in at the factory, and then you distribute modular inference compute clusters all over the country, if not all over the world. That's about as much as I should say about scaling power and cooling.

46:11Like I said, at this point, I'd say I feel highly confident of 10 gigawatts incremental next year, so 13 total. And some number over 20, the probability of achieving that is rising rapidly. Doug, the second part of your question related to funding, I think we certainly have the advantage related to others in that we are investment grade,

46:42have a very strong balance sheet, and there are a number of available pools of debt capital that we are talking to at the moment related to funding this incredible growth. The other really important thing to emphasize is the payback period is very short, and so this capability starts to fund itself in a pretty near-term situation.

47:07And so the fact even that we're finishing this quarter with more capability starts the funding mechanism even into the first half of next year. So, you have your operational cash flows first and foremost, you have multiple pools of capital available in the debt market, especially because of our investment grade

47:27status and strong balance sheet, and then you also have the opportunity from what we've seen to partner with the whole value chain on the vendor and customer side. There's nothing easy about SpaceX for sure, so might as well have the finance piece be part of the challenge at the same time.

47:46But we have a lot of conviction that we were going to be able to do this in a very measured fashion and continue to keep ourselves in that IG status along the way. And certainly, if you look at our track record with respect to capital efficiency and all of our other projects, you'd be hard-pressed to find

48:05any company in the world that is as good at capital efficiency as SpaceX. Next question from Edison Yu at Deutsche Bank. Hi, thank you for taking our questions, and I echo Adam's enthusiasm about the Roadster. There's been a lot of reporting, a lot of anticipation of a merger with Tesla.

48:30I realize you may be limited on what you can say, but if we take a step back, how do you think about the pros and cons of a potential combination? And as part of that framework, are there any roadblocks that you need to address before embarking on it? Thank you. This is definitely the kind of question where I feel like I'm tap-dancing through a minefield,

48:56so I have to be thoughtful about what I say here. Both SpaceX and Tesla have formed special committees to discuss this idea at the request of shareholders. These special committees consist of independent directors. And because I'm a conflicted party, I can't actually be in the discussions of the special committees.

49:27So, it's sort of an odd Kabuki dance where two groups of people are talking and I can't be in the conversation. I don't think we should necessarily assume that it's... I think some people out there assume that a SpaceX and Tesla merger is a certainty, and I think that would be overstating it.

49:58Do I think it would be a good idea? Yes. But at the end of the day, the shareholders have to be supportive of this, and particularly on the Tesla side, we will need a 50% vote of all shares, not just those who vote, but of all shareholders. And that's quite a high hurdle to try to achieve.

50:26I can just say that at this point, the special committees have not reached agreement on the terms of any transaction, and there's no guarantee that they will. But I do think there's a lot of merits to combining the companies as we are pretty much in the super intelligence singularity,

50:48and there's a lot to be gained from cooperating on developing super intelligence software, chip design, Terafab obviously, robotics, mass to orbit with the Starlink constellation... Connectivity. So, I think it would be a good idea, but it's really up to the special committees of both companies.

51:17They've got to come up with something, and then that's got to be put to the shareholders as a vote. Next question from Kutgun Maral at Evercore. Great, thank you for taking the question. I wanted to ask about Starlink mobile, which has come a long way in a short time, including the recent move and deal.

51:38Even so, there's still some uncertainty about how your spectrum portfolio, even though it's much improved and possibly will continue to grow, can compete with incumbents in the US. And part of the unknown for us on the outside is how much coverage and capacity the satellite layer can really provide to bridge the terrestrial gap to peers.

51:58So, what roles do you expect the satellite and terrestrial parts of the network will play, and how should we think about the traffic mix between them in delivering a 5G-like experience in the US? And maybe more broadly, how much of the connectivity focus is on traditional retail wireless versus what could be a massive physical AI opportunity?

52:19Thank you. Want to say something? Elon can fill in some details, but let me point out that the vast majority of mobile connectivity is actually Wi-Fi, right? I think something between 85 and 93% is done through Wi-Fi. And keep in mind Starlink broadband does a lot of Wi-Fi work as well.

52:45As far as the mix goes, I think we expect that the S-band spectrum will provide the majority. And then obviously you need the low-band spectrum, the 800 MHz spectrum, to basically penetrate buildings so if you're not on Wi-Fi, you can still receive the signal. By the way, I'll correct myself if I'm wrong afterwards, but I think the S-band will carry

53:16the majority, and then the low-band spectrum will carry less, but provide a critical piece. And also, we talked last quarter at the last earnings call about smaller, kind of not quite tower-capable, but smaller radios that we could put on already installed homes for Starlink, businesses for Starlink, and towers as well.

53:47So, it's not just the spectrum that we've talked about with the S-band; we will basically leverage the real estate from Starlink installs and other towers. Yeah, I think we probably should do a longer presentation on the game plan for Starlink connectivity and why we think it will ultimately be a business

54:15that generates several hundred billion dollars in revenue a year. Because you are missing some of the pieces that we know about internally. I can say we do see a clear path to several hundred billion dollars a year in connectivity revenue. But I think we should describe that in more detail

54:41in a forum that's not limited to a few minutes at an earnings call. It's actually really three minutes left. Yeah, well, we can go a little longer, I suppose. Longer? I'll take the next question from Eric Sheridan at Goldman Sachs. Thanks so much for taking the question. Maybe if I could turn to the scaling of compute you laid out at the beginning part of the call.

55:04When you think about a framework for applying that compute to the external opportunity that presents itself from cloud services relative to the internal needs of that compute for scaling Cursor and Grok and Grok bot, how has your view evolved about striking the right balance in that mix, and how to think through the output from a monetization standpoint of

55:25what that mix might produce for the business in the next 12 to 18 months? Thanks so much. The relative allocation of compute to third parties versus internal is something that's going to change from one month to the next. As a rough approximation, our goal is to kind of split it 50/50 more or less,

55:59where half of the compute is used internally and half of it is rented to others. That number could shift 60/40 one way or the other; it's going to be dynamic. And obviously our approach is to rent effectively at the spot market rate, as opposed to signing these long-term deals. Our belief is that the value of compute per watt will increase over time, not decrease.

56:32The reason for that, I think it should be obvious to people really, is that as you improve the AI software, as the models get better and better, you get more useful intelligence per watt. And the value of that compute is proportionate to the useful intelligence per watt, and that's why

56:52you see things like even old computers like H100s renting for more than they rented when they were new. And I think that's likely to continue. I think we're going to face a shortage of super intelligence compute, and that shortage is going to remain for a long time. So, I think the smart move is to do short-term contracts and essentially to place the bet

57:29on useful intelligence per watt increasing rapidly, which obviously has happened so far. And also, if you look at how our short-term contracts are renewing, are they renewing at higher dollars per watt or lower? We see them renewing at higher dollars per watt, not lower dollars per watt.

57:55We are also frankly hounded relentlessly by every hyperscaler and by dozens of smaller companies to rent more of our compute. We could rent out our compute several times over if we said yes to everyone. That's just the situation here. That's why I posted that video meme of the Monty Python cheese shop sketch

58:32where the guy goes into a cheese shop to get some cheese and they actually have no cheese. And so he goes like, "What about this kind of cheese? What about that kind of cheese?" and they really just don't have any cheese. We've actually tried to offload some of our inference compute to other cloud

58:51providers so that we can use more compute for training, and we get the cheese shop sketch. They'll say, "Well, what about some GB300s? Nope, don't have any of those. What about some GB200s? Nope, don't have any of those. What about some TPUs? Nope. What about some Trainiums? Nope. How about some H100s?" and they're like, "Well, we might have some overseas somewhere,"

59:16and then they'll check and come back and say, "Nope, we don't have that either." So, it's ridiculous. Yeah, so I don't think we're going to have a demand issue here. It's not impossible—I'm not saying this will happen—it's not impossible that we will have a six-month payback period for compute next year.

59:48We're now going to turn to some questions from individual investors that were posted directly onto our public Q&A platform. Since we've talked a lot about compute deployment terrestrially, let's move to a question on V3 sats. So, when will capacity gains from V3 satellites actually show up as

1:00:06meaningfully higher speeds and lower latency for users? How many V3 sats are needed before the network feels noticeably better in high-demand areas? I think we'll need a few hundred V3 satellites for people to really feel the impact, because right now with very few satellites you'll get an impact, and then 10 minutes later it won't be there.

1:00:29So, a few hundred satellites, and I think we can achieve that by mid-next year. Take number six, which is: can you provide an update on Starbase, Louisiana? What's the timeline, and what's your current confidence level in achieving a first Starship launch from the site as early as 2029?

1:00:50Our internal aspiration for the first launch from Louisiana is actually in two years. So, '28 is our goal. But I think '29 is highly likely. I think we'll be launching from Louisiana at scale in '29. I'm highly confident. You ready to wrap? Yep. Thanks, everyone. This wraps up the Q3 2026 earnings report.

1:01:22Thank you all for tuning in, and we will see you next quarter. Thank you.

资源与出处

相关视频

同频道、同主题或相邻年份