DaPeng (@DaPengX) header
DaPeng (@DaPengX) profile picture

All-in Tesla since 2016. SpaceX shareholder since 2026. Robotaxi · Optimus · 商业太空 · 用 Grok Bot 做小生意 · 在杭州。

Hangzhou.China
Joined August 2014
Photo by DaPeng (@DaPengX)Photo by DaPeng (@DaPengX)Photo by DaPeng (@DaPengX)Photo by DaPeng (@DaPengX)Photo by DaPeng (@DaPengX)Photo by DaPeng (@DaPengX)Photo by DaPeng (@DaPengX)Photo by DaPeng (@DaPengX)Photo by DaPeng (@DaPengX)Photo by DaPeng (@DaPengX)
Grok Bot 真是没有平替。 最近一个月用下来,发现把每个 Bot 当作一个部门而不是一个员工,效果会更好:尽可能把更多的活塞进一个 Bot 里,它会自己去协调每个工作的流程和先后顺序;和其他 Bot 协作的时候,也会根据自己的定时任务去协调。 有了这些 Bot,真的可以把一整个公司都搬到 Grok Bot 上面来
1
1
145
Digital Optimus is coming!
Tesla’s AI, running on our little AI4 computer, can now reach Torment II in Diablo
2
131
巨大的市场,加上严重的老龄化和地缘风险,感觉日本会是 SpaceX 和特斯拉越来越重要的一个市场了
Japan is preparing to massively expand where Starlink can be used 🇯🇵 A Japanese government communications panel has concluded that broader satellite connectivity is technically feasible for: • Low-flying drones operating over mountains and at sea • Disaster response and infrastructure inspections • Cars and other vehicles while moving • Remote monitoring and control of autonomous agricultural tractors And the biggest breakthrough here is MOBILITY Starlink's use on moving vehicles has faced regulatory restrictions in Japan, particularly for its lower-orbit satellite system Now Japan is working toward allowing satellite internet on vehicles and machines while they're actually moving For drones, that means reliable connectivity far beyond the reach of traditional cellular networks For agriculture, it means remotely monitoring and operating machinery across rural farmland For disaster response, it means keeping equipment connected where cell towers and fiber infrastructure are damaged or unavailable Japan is targeting final regulatory conclusions during fiscal 2026 And this represents a major shift
76
must read
Verizon CEO says SpaceX would need $300B to build a competing phone network. Elon has a different plan: turn millions of Starlink rooftops into cell sites, avoiding much of the infrastructure bill. Here is how: Starlink already reaches regular phones from space, and next year's Direct-to-Cell V2 satellites bring about 40x more bandwidth. But thick walls and metal roofs block the signal. Solving that is why carriers spend fortunes on towers, land, power, and fiber. Verizon's Dan Schulman estimates replicating its network would cost SpaceX $300B over 4–6 years, roughly $50–75B annually. So SpaceX has been developing another approach. In August, Shotwell said it could put cellular base stations on the hardware supporting Starlink dishes, avoiding billions in upfront low-band spectrum costs. Then, in October, SpaceX agreed to buy Grain's 800 MHz licenses for a reported $8B. Low-band spectrum travels farther and penetrates buildings better, which SpaceX's FCC filing calls "superior building penetration." But the licenses came with a clock: cover 33% of the population in each of 173 areas by July 2029 and 66% by 2034, or lose them. At Verizon's average build costs, the first deadline could mean $100B+ in under two years, making a cheaper approach essential. At DealBook, Elon revealed the solution: rooftop micro cell towers. Think of them as miniature cell towers mounted beside Starlink dishes. Rather than relying on satellite signals to penetrate buildings, they broadcast cellular signals from rooftops directly to nearby phones, bringing coverage indoors. SpaceX plans to begin production next year, potentially bundling them with Starlink terminals. For wider coverage and capacity, it will also deploy tens of thousands of conventional cell sites using Qualcomm radio equipment, starting in 2027. Together, the two layers extend coverage indoors and outdoors without replicating the traditional network site by site. The cost advantage is that SpaceX doesn't start from scratch. Every Starlink home already provides a rooftop, power, and a satellite connection. Adding a micro cell turns that installation into cellular infrastructure, while conventional sites (also built up to 40% cheaper) fill coverage gaps and handle heavier traffic. That changes the economics of expansion. Verizon must spend billions building coverage before attracting customers. SpaceX can expand its cellular footprint through a product it already sells, with every new Starlink installation potentially adding another cell site. Verizon has to build its network one site at a time. While SpaceX can make every new customer part of the network it is building. Exciting times ahead for SpaceX!
1
73
Google 的 orbital data center和Star Cloud 以及StarMind 类似,都是大型卫星。看来这个方式应该是比较主流的吧。 Starcloud 签了 Rocket Lab,Google 签了 Blue Origin,相信未来还会和 Relativity Space 签约。 Orbital data center 挤占运力的趋势越来越明显了。现在已经上市能买到的、提供运力的公司只有 Rocket Lab 和 Firefly。
Blue Origin and Google have entered into a launch services agreement for Project Suncatcher, Google’s research moonshot exploring how an interconnected network of solar-powered satellites equipped with its AI chips, Tensor Processing Units (TPUs), could scale machine learning compute in space. bit.ly/4CdBOk1
109
Zenith development: COMPLETE ✅ As flight engines clear acceptance testing, we’re shipping them down to Cape Canaveral for integration with Nova’s first stage—every week.
58
220
50
2,069
84,158
$GOOGL SELECTS BLUE ORIGIN TO LAUNCH SOLAR-POWERED AI COMPUTING SATELLITES IN 2029 Google has signed a launch agreement with Blue Origin to use its New Glenn rocket for Project Suncatcher, an initiative to explore running AI computing infrastructure in space. The launch is targeted for Q1 2029 and will deploy satellites equipped with Google's TPUs into sun-synchronous orbit. Each satellite is expected to carry computing capacity roughly equivalent to one data center rack. This will be Google's first dedicated heavy-lift launch for the project. Google has already launched an initial prototype satellite and demonstrated AI inference in orbit. Financial terms of the agreement were not disclosed.
84
95
37
962
110,296
所以还是要投资德州那帮糙汉们:SpaceX 和特斯拉可不觉得这个屎难吃
关于现在AI行业循环融资举债模式,看到一句非常精辟的话: AI就是吃屎游戏,大家都讨厌吃,但是谁都不敢停,停下来就饿死了。😅
1
309
星舰已经达成了突破。 这个突破,是在 SpaceX 零几年那个年代创造了奇迹的基础上,现在才有可能做到的。 感觉绝大多数人还没有认识到,它一旦突破之后的意义。 Starship 加 Starlink,再加上 Orbital Data Center 这两个大飞轮,在它们转起来之后,有机会让人类把轨道工业化。 轨道工业化之后就是月球工业化,最后就是火星殖民地
SpaceX is currently building a Starship upper stage—a vehicle larger than the Space Shuttle orbiter—every three weeks. The goal is to bring that down to every three days in the existing factory.
12
1,945
多数人提到太空数据中心,要么觉得太过遥远,要么觉得技术上有很多困难,比如散热问题。 SpaceX 的 CFO 明确提到,这些问题对他们来说,其实都已经在 Starlink 上面用了。无非是在 Starlink 的基础上,把通讯系统拿掉,换上 GPU,再加上辐射散热板,基本上就可以了。 其实这个和特斯拉当年做自动驾驶的感觉很像:总有一群自称专家的人会先跳出来,说这个不可能、那个不可能,然后舆论就会被这种声音主导。就像当年自动驾驶天梯图把特斯拉排在最后一名一样可笑,说纯视觉不可行。太空数据中心和纯视觉自动驾驶在这方面有点相似。 实际上,这对 SpaceX 应该是比较容易做到的事情。因为这话不是老马说的,是他们的 CFO 说的,至少在时间线上的可信度还是更高一些。太空数据中心在 2028 年,成本就可能会比地面数据中心更低。 再加上 Starship 在那个时候发射频率应该也很高了,而且 Starlink 的太空数据中心并不会遇到太多的技术难题。 这些因素加在一起,我不知道你们有没有买到足够的 SpaceX,反正我的 SpaceX 还是太少了,要再多买一点
BREAKING: Here's the full interview of SpaceX CFO Bret Johnsen with Gavin Baker at the TomorrowX Summit today. 0:00 Launch Video Intro 1:42 See a Launch 2:38 Flight 15 Learnings 2:59 Second Stage Catch 4:15 Orbital Compute Timeline 5:22 Orbital Compute Economics 7:24 Cost Breakdown Details 8:18 Deployment Timeline 8:35 Zero Operating Costs 11:10 Terrestrial Compute Edge 13:08 Casting Own Blades 14:30 Financing & Payback 17:04 Compute Demand Strong 19:28 Anthropic Insourcing Risk 22:33 GrokBot & Software Tools 24:20 Next Grok Versions 24:33 Starlink V3 & DTC 26:56 Community Impact Stories 28:00 Cochise Benefits Stats 29:16 Efficient Network Design 30:13 AI for Ordinary Americans 31:31 Brazil Starlink Impact 32:22 AI in Three Years
2
44
6,567
Typeless 推出了 200 美金一个月的 Ultra 订阅。有钱就是好啊,可以每个环节都用最顶级的
206
特斯拉和 SpaceX 的 PE 都挺高的,在这个榜单里面显得格格不入。这也表明了一种趋势:在 AI 时代,真正的护城河就是要深度结合物理世界。 这个趋势只会越来越明显、越来越确定,因为越来越多的事实在不断摆出来: • Cybercab 的密度在德州最近已经开始提高了 • Starship 越来越成熟,带动Starlink 和 Orbital Data Center这两个飞轮转起来 这些都越来越难以忽略了。特斯拉和 SpaceX 有希望在收入提升的同时,PE 不会降太多
1
1
1
16
65,783
这是不是也意味着特斯拉开启 unsupervised 的订阅了? 估计从 trial 到真正有意义的大规模还需要 1 年甚至 3 年的时间,但从现在开始,那个让股价爆发的临界点随时可能到来。 即便有合并,合并并不意味着一定要把股价框在某一个范围内,因为 SpaceX 的股价也是在变动的
Coming soon. If things well, may start trials next month. Much better to manage a fleet of cars than have to drive just one car by yourself!
4
410
不知道施密特用了什么魔法,让 Relativity Space 的进展这么顺利 2027 年,Relativity Space、Stoke Space、Rocket Lab 哪一家的中型火箭会首飞?
October was a full month of progress for Terran R. 🚀 The historic A2 test stand – built for Apollo and later used for Space Shuttle engine testing – is back in action, now firing for Terran R. Plus: ✅ Interstage structural testing complete ✅ Thrust structure breakover + mate with stage one in progress ✅ LC-16 lightning protection towers at full height ✅ Propellant tanks cryo-shocked
3
1
6
2,727
火箭首先要好看,其次才是性能 好看的火箭一定差不了
October was a full month of progress for Terran R. 🚀 The historic A2 test stand – built for Apollo and later used for Space Shuttle engine testing – is back in action, now firing for Terran R. Plus: ✅ Interstage structural testing complete ✅ Thrust structure breakover + mate with stage one in progress ✅ LC-16 lightning protection towers at full height ✅ Propellant tanks cryo-shocked
109
大家都去 mute 那些垃圾营销号,会让 X 干净很多
对待 x 上的各种信息污染的一个处理方法是,遇到营销号散布以偏概全,夸大其辞,缺乏正确比例感的信息发布时,果断将其 mute, 这样它就不会出现在你的时间线上。 如果这个号真的有价值,它发布的信息迟早会被别的有价值的账号转发,重新回到你的视野里,再把它 unmute 不迟。你并不会损失什么。 营销号的文字特点是,动不动就说: 这太疯狂了!This is insane! 重磅!突发! 或者大量使用明显是 ai 泔水的文字,比如:这不是 xxx, 这是 xxx. 不把这类污染信息源隔断,人注定无可救药的成为一个 sb.
1
1
410
Robotaxi 可以卖给个人,未来大概率也会卖给小公司运营,Waymo 不可以。 Robotaxi 这条纯视觉路线,会直接为未来的 Optimus 打好基础,Waymo 不可以。 Robotaxi 可以在任何路况下运营,Waymo 不可以。 Robotaxi 更酷,而 Waymo 很丑。 Robotaxi 本质上是一台机器人,而 Waymo 不是。 所以在德州 Robotaxi 的数量超过 Waymo 之后,希望以后不要再去和 Waymo 对比了,因为根本就没有可比性
2
4
650