马斯克被问要不要让 SpaceX 上市,被问了快十年。他的答案一直是"不"。这家公司从 2014、2015 年前后就实现了正现金流,一直自己养活自己。那些看起来像融资的私募轮,本质上是给员工和投资人做流动性,SpaceX 甚至在多数轮次里回购股票。公司上下人人持股,没想过去公开市场找钱。
所以当他在摩根大通的活动上,和杰米·戴蒙(Jamie Dimon)对谈时说出"我们现在要上市",理由跟缺钱没关系。接下来要花的钱,多到私人市场接不住。
这场对话信息密度很高。马斯克从上市讲到 10 万颗卫星,从太空数据中心讲到在月球上造电磁炮,从芯片产能讲到美国一座高产量内存厂都没有。中间还夹着他对太阳能量、火星改造、星舰成本和团队稳定性的看法。
把这些论点拆开看,会发现它们指向同一件事:人类要把算力和能源的主战场,从地面搬到太空。
上市这件事,是为扩张期筹钱
马斯克先解释了 SpaceX 为什么长期不上市。
公司从 2014 到 2015 年前后就实现了正现金流,一直靠自己养活自己。私募轮做的是流动性,让员工和早期投资人能套现,SpaceX 在大多数融资事件里还回购了股票。换句话说,这家公司过去十年不缺资本,缺的是花钱的理由。
现在理由来了。他用了"capital growth phase"这个词,资本增长阶段。具体动作是:往轨道上放超过 10 万颗卫星,只用于通信。这些是第三代及以后的卫星,比现在在轨的第一代、第二代强得多。
强多少?按他的说法,V3 卫星的能力是 V2 的 10 到 20 倍,取决于怎么算。SpaceX 芯片团队为它流片了三款专用芯片,性能"far beyond state of the art",远超现有最先进水平。结果是整个星链系统的带宽提升 100 倍,延迟降低一半,因为轨道高度也差不多降到一半。
他顺带给了个判断:这会是世界上带宽最高、延迟最低的通信方式。
为什么需要这么高的带宽?因为 AI 和机器人对带宽的胃口,和人类完全不在一个量级。
人类的峰值带宽大概是每秒几百比特,但一台计算机可以做到每秒一万亿比特。AI 和机器人对带宽、对算力的需求会非常巨大。(马斯克)
这句话是整个访谈的底层逻辑。过去互联网服务的是人,几十亿人,每人几百比特。未来互联网服务的是 AI 和机器人,每个节点的吞吐是万亿比特起步。现有的通信基础设施,从设计之初就没考虑过这种负载。
所以上市只是表象。真正的动作是:SpaceX 要从一家发射公司,变成一家太空基础设施公司。卫星通信是一半,太空 AI 数据中心是另一半。
地面建电厂越来越难,那就把数据中心搬到天上
马斯克说,太空 AI 数据中心是另一个大规模资本项目。他认为这会成为 AI 扩张的主要方式。
理由很朴素:地面建电厂越来越难。
美国平均用电量大约 500 吉瓦。想把用电量翻一倍,就得建差不多两倍的发电厂。问题是有多少人愿意让电厂建在自己家后院?他的原话是,很少有社区对此特别兴奋。
太空没有这个约束。他给了一组让人有点发懵的数字。
当前人类文明使用的能量,不到太阳输出能量的万亿分之一。太阳占太阳系总质量的 99.8%,剩下 0.2% 里绝大部分是木星。就算你把太阳系里所有非太阳的物质都烧掉,太阳的能量占比四舍五入还是 100%。就算你从别的星系再传送两个木星过来一起烧,还是 100%。
所以他的结论是:在太空里,你可以把可利用能量放大到地球经济的一百万倍,仍然不到太阳能量的一百万分之一。
把可利用能量当作经济产出的代理指标,这句话的意思是,太空的经济上限比地球高六个数量级。他说这事儿想起来让人谦卑,地球在太阳面前只是一粒悬浮在黑暗里的微尘。
戴蒙大概是提醒他跑题了,因为马斯克自己补了一句:你问我为什么现在上市,我却在讲太阳的功率输出,这个答案有点长。
但他坚持这个跑题是重要的。总有人问未来能源从哪来,他的答案是太阳能,或者更准确地说,star power,恒星能。
月球比火星更先动工,因为可以拿电磁炮往外打
访谈里有个容易被忽略的判断:去火星不一定非要经过月球,但月球本身值得先干。
马斯克说,在月球上建一座能自我增长的城市,会比在火星上更快。原因有两个。
第一,月球没有大气,重力只有地球的六分之一。这两个条件加起来,意味着你可以用电磁加速器,也就是轨道炮或者质量驱动器,把东西直接甩出去。往深空部署 AI 数据中心,不需要火箭,像轨道炮一样打出去就行。
第二,月球上有材料。太阳能板、散热器,都可以用月球本地材料制造。
这套组合能带来什么规模?他说,从月球出发,理论上可以把 AI 太空算力的部署速度推到每年 1000 太瓦以上。而从地球出发,大概能做到每年 1 太瓦左右。
三个数量级的差距。
他还补了一句很马斯克的话:如果能去月球度假,那会是最史诗级的假期。并非所有人都想去,但很多人想。前提是能安全地去、安全地回来。
火星是下一步。他的描述是,火星是一整颗行星,重力比月球更接近地球,还有大气。如果把火星暖起来,有朝一日可以改造成类似地球的样子,有液态海洋,有生命,出门不用穿宇航服。
他给火星起了个绰号:一颗需要翻修的房子(a fixer-upper of a planet),但潜力很大。
戴蒙问月球上该不该有酒店,马斯克说当然应该有。
SpaceX 更像联合太平洋铁路,而不是一家科技公司
戴蒙大概问了个尖锐的问题:你们做的这些事,听起来像在赌一个还不存在的市场。
马斯克的回答用了历史类比。他说,外界看 SpaceX,更像当年的联合太平洋铁路公司(Union Pacific)。
当年修这条横贯大陆的铁路时,很多人觉得这帮人疯了。为什么要把这么多货和人运到加州去?那里根本没人。但现在加州是美国人口最大的州。
人们当时觉得他们疯了。可今天加州是美国最大的州。(马斯克)
这个类比说到了根子上。基础设施的价值,跟当下有没有需求没太大关系,它能不能把需求创造出来才是重点。铁路修通之前,没有人知道自己需要去加州。星舰和太空数据中心也一样,在运力便宜到某个临界点之前,没有人知道自己需要在天上跑 AI。
星舰的真正突破:把火箭变成飞机
聊到星舰,马斯克先推荐了 SpaceX 官网的直播,说想了解星舰的人看直播最好。
然后他讲了那个被讲了无数遍、但很多人还是没真正理解的点。
星舰将是第一枚完全可重复使用的轨道火箭。
听起来像句废话。但飞机、汽车、自行车、马、船,所有交通工具都是可重复使用的,我们对此习以为常。如果每飞一次跨洋航班就要把飞机扔掉,机票会贵到没人买得起。火箭过去几十年就是这么干的。
技术难度在于,火箭要完全复用非常难。猎鹰 9 号只走了一半路,星舰要走完全程。
走完全程之后,会发生什么?
入轨成本就等于推进剂成本。因为飞行器的所有部件都能重复使用,摊到每次发射上的只有燃料。
星舰用的推进剂是液氧和液态甲烷。马斯克说这是你能找到的最便宜的推进剂。氧气直接从空气里取,甲烷从天然气来。
结果就是:星舰的推进剂成本,比航空煤油还低。这意味着往太空送货的成本,可以低于用飞机跨洋运货的成本。
这句话如果成立,整个太空经济的成本结构会被重写。现在送一公斤东西上天的价格,是航空货运的几百上千倍。降到比航空货运还便宜,等于把"上天"这件事从特殊任务变成日常物流。
V3 卫星大到只能用星舰发射
戴蒙追问 V3 卫星到底多大。
马斯克的回答是:很大,网上能查到。能力是 V2 的 10 到 20 倍。大到什么程度?只能用星舰发射,地球上其他任何火箭都装不下。
星舰的货舱直径 30 英尺,大约 9 米。V3 卫星宽约 7 米,也就是 22 到 23 英尺,体积接近一辆小巴。
技术细节他也讲了几个:更大的相控阵天线,更多的地面链路,更多的激光器。卫星之间用激光通信,这些激光器是 SpaceX 自己研发自己制造的。V3 上激光器数量更多、性能更先进,还加了 W 波段和 E 波段。
他的形容是:像一座疯狂的在轨无线电基站。
戴蒙问单颗卫星能做到多少,马斯克说"应该能做到 50",文字稿里没带单位。紧接着他给了一个更清楚的目标:星舰 V3 的目标是完全可重复使用状态下每次送 100 吨入轨,V4 的目标是每次 200 吨以上,并且能每小时发射一次。
每小时一次。这个频率已经接近航空公司的航班密度了。
太空数据中心比通信卫星简单,这一点很多人没想到
这里有个反直觉的判断。
一般人会觉得,在天上跑 AI 数据中心,比在天上放通信卫星难得多。马斯克的说法正相反。
他说,这件事不比造通信卫星难,实际上简单不少。星链 V3 通信卫星是一台极其复杂的机器。相比之下,AI 数据中心反而简单,它本质上就是太阳能板加散热器,加一些运行卫星的基本设备,再加激光链路。
激光链路负责把数据中心接入星链通信星座,数据再从星链传回地面。
这个架构解决了一个关键问题:天气。
激光在太空里传输不受天气影响。星链对地面通信用的是能穿云的频段,甚至能在一定程度上穿屋顶。所以从太空数据中心到星链、再到地面,这条链路在任何天气下都能闭合。
换句话说,太空数据中心不用自己解决对地通信。它只要把数据交给星链,剩下的星链已经解决了。这是垂直整合带来的好处,两块业务互为基础设施。
真正的瓶颈在芯片,不在火箭
整场访谈里,最有信息量的一段是关于芯片的。
戴蒙问限制因素是什么。马斯克的回答很直接:芯片制造,包括逻辑芯片、内存芯片和封装。
然后他抛出一个事实:美国现在没有一座高产量计算机内存晶圆厂,一座都没有。
美光(Micron)在爱达荷州建了一座,但量产要等到 2028 年。纽约州还有一座在规划,时间点在 2029 到 2030 年。这些产能加起来,只是所需内存的一小部分。
他进一步说,就算你按内存厂商和逻辑芯片厂商最乐观的假设去算,产能也不够满足预期的需求。
这就是为什么美光的股价会涨到那么高。他随口说了个数字,1.2 万亿美元。这个数字是否准确不重要,重要的是他背后的判断:市场在给芯片产能极度稀缺定价。
AI 需要的逻辑、内存、封装,也就是一整套 AI 计算机的供给,远远超出所有现有晶圆厂最乐观假设能提供的水平。
所以 SpaceX 要自己下场做 TeraFab。马斯克说,这几乎是必须的,否则芯片就是不够用。
这段话要单独拎出来说。很多人以为 AI 的瓶颈是电力、是数据、是模型能力。马斯克给的是另一条链:电力可以去太空解决,数据可以去太空处理,但造芯片的产能,实打实卡在地面上。
太空算力要做成开放平台,不搞封闭生态
戴蒙问这些太空 AI 卫星会用什么芯片。
马斯克的回答很明确:开放。
英伟达的 GPU 可以装,谷歌的 TPU 可以装,亚马逊的 Trainium 可以装,任何客户想装的芯片都可以装。SpaceX 未来也会提供自己的芯片,也会提供自己的 AI 软件。平台本身不锁定,客户的 AI 硬件和 AI 软件都能在上面跑。
这个定位和星链不同。星链是 SpaceX 自己运营的封闭网络。太空算力则被设计成一个别人可以接入的基础设施,像云服务商,只不过机房在天上。
至于为什么这么做,逻辑也通。太空数据中心的资本开支巨大,靠一家公司自己消化芯片需求不现实。做成开放平台,才能把整个 AI 产业的算力需求吸进来。
亲美立场、军事业务和那间不能说的部门
戴蒙问到地缘政治。
马斯克的回答不含糊:他非常亲美,一直如此。SpaceX 为美国国防部做了很多工作,他特意提了一句,现在叫战争部(Department of War)。公司有个部门叫 Starshield,提供军事通信。
他还说,还有一些机密项目,不能谈。但 SpaceX 确实在帮助战争部和情报部门,是政府体系里的关键一环。
这段回答没有太多新信息,但把 SpaceX 的业务结构说清楚了:商业发射、星链通信、Starshield 军事通信、未来的太空算力,四条腿。前三条已经在跑,第四条是这次上市要融钱去干的。
高管一干就是二十年,因为信这件事
戴蒙提到了格温·肖特维尔(Gwynne Shotwell),SpaceX 的总裁兼 COO。
马斯克说,她是公司第 7 位左右的员工,2002 年加入,到现在 24 年了。公司高管的任期普遍很长,CFO 布莱特·约翰逊(Bret Johnsen)已经干了 15 年。
为什么留得住?他的解释是,人们真的相信这个使命,想留下来继续把它建出来。
使命是什么?让人类成为太空文明。把人类带到火星、月球,最终去更远的地方。让科幻变成现实,去那些从未被探索过的地方。
他说了一句话,带着点孩子气:让《星际迷航》成真,那是我最喜欢的剧之一。
这段对话透露的信息是,SpaceX 的组织稳定性不靠薪酬结构,靠叙事。一个能让人干 24 年的故事,比任何期权池都管用。这在招聘极度内卷的 AI 时代,是个不太被讨论的竞争力。
他说自己变了,招人开始看心
戴蒙问他自己这些年有什么变化。
马斯克的回答是,比以前放松了。他的原话是,比以前随和多了,虽然还是不算特别随和,但肯定比过去好。
他还讲了一个招人标准的变化。过去他更看个人能力和智力水平。现在他觉得,能力当然重要,一个人有没有好心肠同样重要。
IQ 高低并非重点,是不是一个好人,这件事很重要。
这话从他嘴里说出来有点意外。外界对他在 X 上的言行、对推特的裁员方式、对超强度工作文化的印象,和"看人心肠"这个标准不太搭。但他确实这么说了,而且说这是他这些年学到的东西之一。
他也承认自己还在学,还会犯很多错。
至于未来,他给了一句挺有意思的话:也许将来 AI 会评价说,"对人类来说还不错"(Not bad for a human)。
这句话把整场访谈的底色点出来了。一个在谈把文明扩展到太阳系的人,同时也在想 AI 会怎么给人类打分。这两件事在他脑子里是并存的。
最后那句"距离 7 月 4 日正好一个月"
访谈结束时,马斯克说了句:我们距离 7 月 4 日正好还有一个月。
这场对话发生在 2026 年 6 月。2026 年 7 月 4 日是美国建国 250 周年。这句话不是随口一提,他在给一个时间坐标。
把整场对话串起来看,马斯克在讲一个分阶段的时间表:
- 短期:星舰完全复用跑通,V3 卫星开始部署,带宽提升 100 倍
- 中期:太空 AI 数据中心上线,算力以开放平台方式对外提供服务
- 中长期:月球城市起步,用电磁炮把算力设施打进深空,规模上一个数量级
- 长期:火星改造,让人类成为多行星物种
四个阶段里,前两个已经进入工程实施,第三个有了技术路径,第四个还是愿景。
而上市这件事,本质上是为第一和第二阶段筹钱。他自己说得很直白:我们在进入一个大规模增长阶段,需要资本。
这场对话真正说了什么
如果只看标题,这是一场关于 SpaceX 上市的访谈。把内容拆开,马斯克实际上给了一套完整的论证链。
第一环,AI 对带宽和算力的需求会涨到地面基础设施撑不住。人类带宽是几百比特每秒,机器是万亿比特每秒,中间隔着六个数量级。
第二环,地面的两个约束解不开:电厂没人愿意建在自家后院,芯片产能短期上不来。前者可以用太空太阳能绕开,后者只能靠新建晶圆厂硬扛。
第三环,太空提供了成本结构上的解法。星舰完全复用后,入轨成本等于推进剂成本,比航空煤油还便宜。月球的重力和真空条件,又让电磁发射成为可能,把部署规模再推高三个数量级。
第四环,谁来买单。开放平台,任何人的 GPU、TPU、Trainium 都能接进来,SpaceX 提供运力、电力、散热和对地链路。
这四环里,最不确定的是第二环。马斯克自己承认,芯片是限制因素,美国的先进内存产能要等到 2028 年之后才有规模。太空算力的野心再大,也得等地面把芯片造出来。
最确定的是第一环。AI 和机器人对带宽的需求会暴涨,这个判断几乎不需要论证。
第三环和第四环,是工程和商业问题,需要时间和资本去验证。这也是他要去公开市场的原因,私人市场吃不下这个量级的投入。
戴蒙在这场对话里扮演的角色,是不断把马斯克从太阳系拉回财务模型。问他收入可预测吗,问他卫星多大,问限制因素是什么。马斯克自己都承认回答跑题了。这种跑题反而有价值,你能看到他脑子里的地图长什么样,以及这张地图上哪些地方已经修好了路,哪些地方还只是画了个圈。
给读者的几条可操作建议
- 盯住星舰的完全复用节点。这是整个太空经济成本曲线的拐点。完全复用跑通之前,所有太空算力的商业模型都是纸面推演;跑通之后,推进剂成本成为入轨成本的上限,整个成本结构重写。SpaceX 官网的星舰直播是最快的信息源,比二手报道早几天。
- 把芯片产能当作太空算力的先行指标。马斯克说限制因素是逻辑、内存和封装,美国现在零座高产量内存厂。美光爱达荷厂 2028 年量产,纽约厂 2029 到 2030 年。这些时间点决定了太空数据中心能多快铺开,比任何发布会都靠谱。
- 关注 TeraFab 的进展。如果 SpaceX 真的自己下场做晶圆厂,这会是一个跨界的信号,说明它不满足于做集成商,要做垂直整合。这条线一旦启动,影响的不只是 SpaceX,还有整个半导体设备链。
- 理解"开放平台"这个定位的含义。太空算力不做封闭生态,意味着英伟达、谷歌、亚马逊的芯片都能上。这对做 AI 基础设施的公司是个机会窗口,运力和电力可以被外包,模型和芯片仍然自己做。
- 对月球时间表保持谨慎,对火星时间表保持更谨慎。月球城市、电磁发射、1000 太瓦年产能,这些是技术路径清晰的愿景,不是工程排期。马斯克给的是方向,不是日期。判断进度时看硬件迭代节奏,别听时间承诺。
- 留意 SpaceX 上市后的财务披露。这家公司长期不透明,上市意味着要公开收入结构、发射成本、星链用户数、资本开支。这些数字第一次可查的时候,整个航天和通信行业的估值体系都要重算一遍。
I had no idea she was there, actually.
I’ve been asked for many years about taking SpaceX public, so it’s probably been almost 10 years that people have been suggesting to me that I should take SpaceX public. We’ve been positive cash flow for quite a long time, I think since around 2014, 2015, and we’ve been self-funding. In fact, in our private equity rounds, they actually have not been fundraising rounds, they’ve been liquidity rounds for investors and employees because we give everyone at the company stock. And SpaceX has actually bought back stock in most of our sort of funding events.
So what’s different about now is that it’s a number of things, but we are embarking on a significant growth phase, like a capital growth phase, where we are going to put in orbit probably 100,000 satellites, or probably over 100,000 satellites just for communications. And these will be the Version 3 and beyond versus Version 2 and Version 1 that are currently in orbit. Version 3 is, depending on how you count it, 10 to 20 times more capable than the Version 2 satellite.
And there were 3 chips that the SpaceX chip design team taped out that are specific to this, that are far beyond state of the art, which means it’s 100 times more bandwidth than the SpaceX Starlink system currently offers, and also half the latency because the altitude will be about half altitude. I think it will actually be the highest bandwidth, lowest latency means of communicating, and the future with AI and robots is actually going to require a lot more bandwidth than we currently use. Because you can imagine what’s the bandwidth of a human? The peak bandwidth of a human is a few hundred bits per second, but bandwidth of a computer can be a trillion bits a second.
So the appetite for bandwidth of AI and compute, AI and robots is going to be enormous. And then we’re also doing the AI data centers in space, which is another massive capital endeavor. But I think it’ll be the primary means by which AI can be expanded. It’s increasingly difficult to build power plants on the ground. There are very few people who want a power plant in their backyard. So if you want to, say, double the electricity usage of the United States, which is on average about 500 gigawatts, we would have to build about twice as many power plants, which I don’t think most communities are super excited about.
But actually, if we go to space, we can go far beyond the electricity generation of Earth. In fact, this is going to sound crazy, but you could actually increase harnessed energy by a factor of a million and still be using much less than a millionth of the Sun’s energy. Current human civilization uses much less than a trillionth of the Sun’s energy output. It’s just humbling to think about. We’re really a tiny — when you see the true size of Earth relative to the Sun, we’re a tiny dust mote in a vast darkness. And the Sun is enormous. The Sun is 99.8% of all mass in the solar system, and most of the remaining 0.2% is Jupiter.
Sometimes people ask me — and I’m maybe going a little wide-ranging in this answer, but because they just asked me why are you doing public now, and I’m talking about the Sun’s power output. A bit of a long-winded answer. But it is important. Some of these things are important because people sometimes wonder what’s the future of energy generation. And I can say that it is absolutely solar power, or maybe a better word for solar power is star power. And the crazy thing is that if you burnt all mass in the solar system that was not the Sun, the amount of energy produced by the Sun would still round up to 100%. Because the Sun is 99.8% of the mass of the solar system.
Even if you somehow teleported 2 more Jupiters from another solar system and burnt them too, the Sun would round up to 100%. It’s very much the Sun. And you can scale to a million times Earth’s economy in space in terms of harnessed power, which is a good proxy for economic output, and still be much less than a millionth of the Sun’s energy, which is humbling, really, to think about how tiny we are. And this is just one star among many. So the TL;DR would be we’re embarking on a massive new growth phase and we need capital for that. Another thing is the revenue. I also feel pretty good about the revenue projections.
Before, revenue was a little unstable, but now I feel like the revenue is much more predictable.
Yeah, so you don’t necessarily have to go through the Moon to get to Mars. I just think that we can build a self-growing city on the Moon faster than we could do so on Mars. And there’s also the potential, if you say you want to scale far beyond what you can do from Earth, is that because the Moon has no atmosphere and about 1/6th Earth’s gravity, you can use an electromagnetic accelerator, a railgun or mass driver. Basically, you don’t need to use rockets to do AI data centers into deep space from the Moon. You can literally just shoot them like a railgun type of thing. And you can manufacture the solar and the radiators, solar-powered radiators on the Moon from Moon materials.
That would allow scaling potentially to beyond 1,000 terawatts a year, which is a truly staggering number. I think we can do probably somewhere around 1 terawatt per year of AI space compute from Earth, but we can do 1,000 terawatts or more from the Moon. And I think it would be pretty cool if you could vacation on the Moon. That would be the most epic vacation. Not everybody wants to go to the Moon, but I think a lot of people do. I think it would be pretty amazing, obviously provided you could do so safely and come back safely. But I think that will be possible in the future. And then Mars is another step beyond that.
Mars is a whole planet, with gravity much closer to that of Earth’s, and it has an atmosphere. And if you warm up Mars, you could one day make Mars like Earth, meaning with liquid oceans and life, and where you could walk outside without a spacesuit. So Mars — I call Mars a fixer-upper of a planet, but it’s got a lot of potential.
Yeah, absolutely. We should have Moon hotels, don’t you think?
Actually, you think of us more like we’re kind of like the Union Pacific. When they built Union Pacific back in the day, people thought they were crazy because, why are you trying to carry all this cargo and people to California? No one’s there. But now California is the biggest state in the country.
Our webcasts are very good. So I recommend anyone who wants to learn about Starship, the SpaceX website or any of the Starship livecasts are a great way to learn about it. But the fundamental breakthrough of Starship is that it will be the first orbital rocket that is fully reusable. This might sound like an obvious thing, but in every other mode of transport, whether that’s aircraft, cars, bicycles, horses, ships — these are all things we take for granted as reusable. An aircraft journey would be very expensive if you had to throw the plane away every time. And that’s how rockets have been in the past.
But it’s very difficult from a technology standpoint to achieve full reusability for a rocket. We’ve got part of the way there with Falcon 9, but we’ll get all the way there with Starship. And once you achieve full reusability, then the cost of access to orbit is simply the cost of propellant, because now you can reuse all aspects of the vehicle. And the propellant we use for Starship is liquid oxygen and liquid methane, which is the cheapest propellant you could possibly get. We can literally get oxygen from the air and methane from natural gas.
So the cost of propellant for Starship will be less than the cost of jet aviation fuel, which means that you should be able to actually send cargo to space for less than the cost of cargo on an airplane going on a transoceanic trip.
Yeah, so the V3 satellite is a giant, and you can look it up on the internet. As I said, it’s 10 to 20 times more capable than V2. It’s a very big satellite. In fact, it can only be launched on Starship. It’s too big to be launched on any other rocket on Earth. Starship has a 30-foot diameter cargo bay, and the V3 satellites are about 7 meters, so about 22, 23 feet wide. Very big, like the size of a small bus essentially. There are a bunch of technical details — we have much bigger phased array antennas, we’ve got more ground links, we’ve got more of our lasers. Our satellites communicate with each other with lasers that we developed and manufacture.
So it’s got a lot more lasers and more advanced lasers. It’s also got W-band and E-band. These are technical details, but it’s like a crazy orbiting radio station.
It should be able to do 50.
Yeah, because Starship V3 is aiming to do 100 tons to orbit with full reusability. And then Starship V4, we’re aiming for over 200 tons per mission and then being able to launch every hour.
We don’t think this is a particularly difficult thing to do. In fact, we think it’s easier than our communication satellites, quite a bit easier than our communication satellites. The Starlink V3 communication satellite is an incredibly complex machine. The AI data center would be much simpler by comparison because it’s really just solar power plus radiator, some basic equipment for operating the satellite, and then the laser links which would connect to the Starlink communications constellation, and then back, and then to the ground.
And it would — the connection would happen no matter what the weather is, because once you connect via lasers to the Starlink communication constellation, the Starlink communicates to the ground with frequencies that are cloud-penetrating. So they, in fact, even roof-penetrating to some degree. So you would always be able to close link with the data centers.
We try to — so what’s the limiting factor? And the limiting factor, what we see as limiting factor, is being able to make chips, both logic, memory, and packaging. It’s worth noting that there’s not a single high-volume computer memory fab in America right now. Zero. There’s one being built in Idaho by Micron, but that will not reach volume production until, I believe, 2028. And there’s something built in New York, but they are in, I think, ’29 and ’30. And this is a tiny fraction of the memory that’s needed.
And in fact, even if you take the best-case assumptions of the memory makers and the logic makers, it is not enough to meet the demand that is anticipated, which is why you’re seeing stocks of Micron go to, I think, $1.2 trillion or some quite high number. So there’s just clearly a need for AI logic, memory, and packaging — AI computers essentially — that is far beyond what even the best-case assumptions of the existing fabricators can do. And that’s why we need to do the TeraFab. Yeah, it seems like essential, otherwise there will not be enough chips.
Yeah, so we do intend with our SpaceX AI satellites, to allow people to put whatever GPU or TPU they want. So NVIDIA GPUs can be put on it, Google TPUs can be put on it, Amazon Trainiums or any other chips that people want to put on can be put on. We will also offer our chips in the future, and I think we also want to offer our software, our AI software as well in the future. But it will be such that you can run anyone’s AI hardware or software on the SpaceX AI satellites.
Yeah, I’m incredibly pro-American. And frankly, always have been. SpaceX does do a lot of work for what used to be called the Department of Defense, the Department of War these days. We have a division called Starshield, which provides military communications. And there’s some other stuff that’s kind of classified — can’t talk about that, but we are helping the Department of War and Intel, part of the government, a vital element of that. So yeah, I’m just super pro-America, always have been. Thank you.
Yeah, Gwynne was, I think, around the 7th person to join the company, and that was 2002. So it’s been like 24 years. And generally the senior executives at the company have a very long tenure. I think Bret Johnsen’s been CFO for 15 years. And yeah, I think because people really believe in the mission, they want to stay and they want to keep building it. We want to make humanity a spacefaring civilization. We want to take humanity to Mars and the moon and ultimately beyond. We want to make — it’s like science fiction — make it real. Go to places that have never been explored before. Make Star Trek real. That was one of my favorite shows.
I think I’m probably more chill than I used to be. I’m way more laid back than I used to be — which I’m still not that laid back, but more than I used to be for sure. And one of the things that I found — I’ve mentioned this before — is that in terms of recruiting people to the company or having people work at the company, their individual abilities and their intellectual capabilities matter a lot, but it also matters if they have a good heart. It’s not just about whether somebody has a certain IQ or whatever, but just are they a good person? That matters a lot. I don’t know. I guess I’ve learned a lot. Although I feel like I still have a lot to learn and make a lot of mistakes.
But I think maybe the future AI will say, “Not bad for a human.”
Thank you. Thanks for having me. I will say that we are exactly one month away from the Fourth of July.
人类的峰值带宽是每秒几百比特,而计算机的带宽可以达到每秒一万亿比特。
The peak bandwidth of a human is a few hundred bits per second, but bandwidth of a computer can be a trillion bits a second.
所以火星——我称火星为一颗有待修缮的行星,但它潜力巨大。
So Mars — I call Mars a fixer-upper of a planet, but it’s got a lot of potential.
当年他们修建 Union Pacific 时,人们觉得他们疯了,因为——你为什么要把这么多货物和人运到加利福尼亚?那儿根本没人。但现在加利福尼亚是美国最大的州。
When they built Union Pacific back in the day, people thought they were crazy because, why are you trying to carry all this cargo and people to California? No one’s there. But now California is the biggest state in the country.
如果每次飞完都得把飞机扔掉,那坐飞机就会非常昂贵。而过去的火箭正是这样。
An aircraft journey would be very expensive if you had to throw the plane away every time. And that’s how rockets have been in the past.
然后 Starship V4,我们的目标是每次任务超过 200 吨,并且能够每小时发射一次。
And then Starship V4, we’re aiming for over 200 tons per mission and then being able to launch every hour.
值得注意的是,美国目前没有一座大规模量产的计算机内存晶圆厂。零。
It’s worth noting that there’s not a single high-volume computer memory fab in America right now. Zero.
我们想让人类成为太空文明。我们想把人类带到火星、月球,并最终走向更远的地方。我们想让——就像科幻小说那样——让它成真。
We want to make humanity a spacefaring civilization. We want to take humanity to Mars and the moon and ultimately beyond. We want to make — it’s like science fiction — make it real.
但我觉得,未来的 AI 也许会说:“对一个人来说,不算差。”
But I think maybe the future AI will say, “Not bad for a human.”
资源与出处
- 视频在 YouTube 收看
- 档案silicon-valley.tech 档案页本页视频与文稿数据来自 Elon Musk Archive
- 精读本篇精读 Markdown 版