The Big Picture

SpaceX is developing Kessel, a custom Vera Rubin NVL72 rack that replaces internal NVLink cable cartridges with a passive PCB midplane. It keeps Oberon’s 72 GPU packages, 18 compute trays and 9 switch trays, but rearranges them back-to-back around the board. Think mini Kyber, with a smaller scale-up domain and a design intended for both terrestrial and orbital deployment. The question is whether this smaller version can make the midplane practical. (1/4)🧵

Starship Flight 15

Liftoff of Starship Flight 15

Starship is now in a long coast phase around Earth. Teams will monitor Starship’s systems over the next ~21 hours that we intend to be in orbit for. We do have a few early deorbit options available should we need them, but all systems are looking good right now

On track for Starship’s 15th flight. The 75-minute launch window opens at 4:45p.m. CT. Live coverage of the mission begins ~42 minutes before launch

Issued by the FAA just moments ago, ALL US airports are currently ground stopped. This means NO flights may depart for US airports. Flights already in the air will be allowed to land. This is related to an FAA equipment outage.

Starhopper. Incredible how far we’ve come. Feels like just yesterday that people were making fun of this thing for being a “floating water tower” and “not a real Starship”

Just some birds waiting to fly

I recently spent two weeks in Japan, and I have to say, the trip really opened my eyes to just how much better we could be doing here in America. I’m not exaggerating. Some of the differences were shocking. Here are a few things I noticed: - In two weeks traveling through some of the most densely populated cities in the world, I saw TWO visibly homeless people. TWO! I saw just ONE person who appeared to be struggling with drug addiction. - The high-speed rail system was incredible. Clean, affordable, fast, and efficient. Meanwhile, Americans are still arguing about whether high-speed rail is even worth building. - I never once felt unsafe walking around with my kids, even in massive cities surrounded by thousands of people. - There were barely any public trash cans, yet somehow the streets were almost spotless. - The difference in income inequality was noticeable. The infrastructure, housing, and public services reflected a society that seems to take much more pride in making everyday life work for everyone. Now, I already know what some people on the right are going to say. “Japan doesn’t have mass immigration!” “Japan isn’t run by Democrats!” OK, but can we stop turning EVERYTHING into a partisan argument for five minutes? Japan isn’t perfect. It has its own serious problems, and I’m not pretending two weeks makes me an expert on the country. But what I saw made one thing painfully obvious to me. We have normalized so many problems in America that we’ve convinced ourselves they’re simply unavoidable. Homelessness. Dirty streets. Failing infrastructure. Terrible public transportation. People feeling unsafe in their own communities. Why? We are one of the wealthiest nations in human history. We have the money. We have the technology. We have the talent. What we seem to lack is the willingness to stop making excuses and demand better. I’m proud to be an American. But being proud of your country shouldn’t mean pretending other countries haven’t figured out how to do certain things better. Patriotism isn’t ignoring America’s problems. It’s wanting America to be better. And after spending two weeks in Japan, I’m more convinced than ever that we can do a whole lot better.

Video footage of the Prometheus rocket engine hot-fire test at the P5 test bench at @DLR_en in Lampoldshausen, Germany. At the P5 test bench, methane flows at 110 kg/s - that's about 260 l/s!

Ship 42 moves to the Starbase pad ahead of launch

I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.