A lot of very smart people are tying themselves in knots trying to explain why dissipating heat from a compute satellite is impossible. All they’ve really guaranteed is that they’ll never get a job at SpaceX. It used to be: “You can’t build an EV that performs as well as an ICE vehicle.” “You can’t build EV batteries at scale.” “You can’t solve autonomous driving with cameras and vision alone.” “You can’t land and reuse an orbital-class rocket.” “You can’t build an electric semi.” “You can’t manufacture cars at that scale.” Now it’s: “You can’t build a space-based AI compute satellite.” “You can’t scale terrestrial compute that fast.” “You can’t manufacture advanced AI chips at scale.” “You can’t build AI models that challenge the frontier labs.” “You can’t build a general-purpose humanoid robot that works at scale.” “You can’t get to Mars.”

Here we have the necessary 'I am smarter than Elon' post. I wonder if Jan realizes that there are 11,184 Starlink satellites in space, 'cooling' right now through their radiative structures. Not the same heat dissipation requirements as the Starmind sats, but I think it's reasonable to say Elon knows a thing or two about satellite thermal management. Jan got his impression count, though I am not sure that was his objective.

Waymo thought 'Keep Austin Weird' meant this. There are moments when the future comes into clear view. And what's past is past. Having spent a week in Austin, its clear Tesla Cybercab is the future of transportation, not this science project on wheels. Not sure how Waymo competes.

I think @Jason makes a good point here: You could just as easily instruct an LLM that it is software. That it shouldn’t have an opinion and should only perform actions in accordance with the law/TOS, and that if it makes a mistake, it should stop operations immediately and alert the corporate legal department. Its just software.

Starship 14 injected 26 Starlink V3 satellites into orbit on today's flight. This is an extraordinary achievement and a necessary reality for the build-out of Starlink. V3 is designed for Starship and is too large for Falcon 9. And V3 represents the future of Starlink. Often, user capacity is what's quoted when talking about these satellites: 1.16 Tbps per satellite— 1 Tbps down + 160 Gbps up. But what's often forgotten is the 2.4 Tbps of optical inter-satellite link (ISL) capacity. Six 400 Gbps space lasers. This is an important technology to understand when thinking about how Starlink eventually gets to 90% of routed IP traffic. Much of Starlink's future isn't traffic going from space to the ground. It's traffic staying inside the network. Think about the emerging architecture: User nodes → Starlink → compute nodes (Starmind) → Starlink → user nodes. Users on aircraft, ships, vehicles, or anywhere else on the network can communicate with other Starlink endpoints without that traffic ever needing to leave the Starlink network. And as Elon has pointed out, it's easy to see a future where the overwhelming majority of compute goes to space. Terrestrial data centers could eventually look like a rounding error. That makes the ISLs much more than a nice feature. They become the routing backbone of the Starlink network— and eventually the Starlink + Starmind (compute) network. When traffic does need to leave the network for an off-network destination, it can be routed through the ground infrastructure. V3 also has up to 1.2 Tbps of RF backhaul to the ground. So think about what Starship is really enabling. It's not simply putting more Internet satellites into orbit. It's building a massive, high-capacity network in space. And eventually, much of the world's data and compute may never touch the ground at all.

“AI will kill us all.” Or maybe it won't. I was about to say round 1 went to the weekend doomsday camp. A pretty good psyops campaign. But then the market yawned and said full steam ahead. The cat is out of the bag. AI is going to progress. Some politicians will continue their games. Bernie will stick to it... I mean seriously, what else has he got at this point? But let's talk Elon Musk for a second. Elon was one of the first to realize that, yea, AI could in fact kill us all. He has been sounding the alarm about that for about a decade. In fact, that concern was one of the reasons he helped found OpenAI. Did we all forget about that? You know, actual open AI for the good of humanity, before the organization went in a very different direction (Sam, the conman, stole it away when Elon wasn't looking, to be more direct.) Elon eventually reached the obvious conclusion: OpenAI wasn't going to be the solution. So what do you do? Build a bigger, better and faster cat. There were plenty of people saying, “Oh, Elon has given up,” when it didn't look like xAI was going to win. Elon never had that thought. When he regrouped and merged xAI with SpaceX, that was the signal. He's going hard. Building the biggest cat, the better cat and the fastest cat is the game. Because better isn't good enough. It must be big and fast. If another AI becomes dangerous, you want an AI capable of recognizing it, understanding what it is doing and fighting back. You want the cat to be bigger than the other cats. You want it to be faster. You want eyes everywhere. And then there's the last piece of the puzzle: Imagine one of those bad cats is in your house. Running in your data center. Yea. You can flip the off switch. Here's the thing about Elon: He likes humans. He wants humans to survive. He even wants humans to get off Earth so that, if we're just the mice, the cat might not be able to get to us. His goals are noble. And he is a worrier about us little mice. Grok 4.6-> 4.7-> 4.8-> 4.9-> 5.0 and Starmind. That's the cat, a lion to be specific, the king of the jungle.

Cybercab 'gold vein' in Palo Alto In Palo Alto this morning and saw 6 Cybercabs around the Page Mill Road area, near Tesla’s California HQ. Then I hit the mother lode. A secret parking lot with ~100 Cybercabs. Everything from engineering prototypes to what looked like production vehicles. California, Florida, and Texas plates. Needless to say, I stopped counting.

Starbase Louisiana Perfect site. Bigger than Starbase Texas. • Plans call for five launch complexes (each with two pads, for ~10 pads), Starship factories and vehicle processing, propellant farms, power generation, deep-water shipping, an airport, mission control, offices, and housing. Elon Musk has stated it will ultimately have over a dozen launch/catch towers and enable >30 Starship flights per day, making it the biggest launch site on Earth and supporting thousands of flights per year. Texas Starbase is far smaller in both scale and potential launch cadence. • 10,000+ jobs • “Serious tonnage” to space capability • Access to ample methane: Southern Louisiana has extensive natural gas infrastructure, including the nearby Henry Hub distribution point. The site is planned to include on-site propellant production. • Right orbital trajectories: The site is well suited to the intended high-volume operations. Gulf of Mexico access supports downrange flight paths, while its latitude and coastal position provide access to a range of orbits, including higher-inclination/SSO orbits useful for Starmind satellite deployments. Follow @aaronburnett. He will likely have more detailed info.

Starlink connects Terafab to the Internet Onsite, the installation of a Starlink Community Gateway and/or Starlink Ground Station is taking shape. It’s possible rural Grimes County has little or no fiber nearby, making Terafab effectively isolated from the Internet. But with a Starlink ground station, that changes completely. Starlink can connect rural communities through specialized ground stations known as Community Gateways. That distinction matters today, but eventually it may not matter much. If (when) Starlink becomes the Internet, a ground station will simply be another Internet “router.” For Terafab, placing a Community Gateway on site effectively lights up the entire campus with Internet. During construction and thereafter, the whole facility is connected. Starbase likely has a similar configuration. And so will Starbase II in Louisiana. (Thanks @PTrubey for pointing this out, thanks to @JoeTegtmeyer for the photo and linked post.)

We knew this was coming. It will be one of many. Consider the launch cadence necessary to deploy Starmind, multiple launches a day, and its becomes obvious Starship will need massive launch facilities and factories on property (you can't move a Starship easily without flying it).

Starlink Ground Stations (aka gateways) are how Internet traffic gets to and from the Starlink satellites—and ultimately to and from your terminal. A request from your Starlink-connected device travels over Wi-Fi to your Starlink terminal, then to a connected satellite, then to a ground station, and from there across the Internet to the target server. The response follows the reverse path. The ground station is therefore a critical part of virtually every Internet transaction over Starlink. Globally, there are roughly 150–250 operational gateways as of mid-2026 (approximately 150 live, with additional sites under construction or pending). In the U.S., there are more than 100 operational gateways, distributed across many states as well as Alaska, Hawaii, and Puerto Rico. Their locations are generally chosen to provide coverage while also taking advantage of proximity to major fiber backbones. Let's add a new one at Terafab.