MIf you want to see where the robot wars are actually being fought, forget the finished bots. It's the parts.
CRight to the supply chain. That's where the money is.
MUBS put out a breakdown of who's building what for humanoid robots. The makers, then the brains.
COn the makers side you've got China with Unitree, UBTech, AgiBot. Overseas it's Tesla, Boston Dynamics, Figure, Agility.
MThen the computing side: Nvidia, Qualcomm, AMD, Intel, and China's Horizon Robotics.
CI'll tell you the part that jumps out. AMD and Intel on the same list as Nvidia for robot brains.
MIntel has been promising robotics chips for years. Where's the actual product?
CFair, but UBS isn't in the business of fantasy. They're mapping where the dollars actually flow.
MI'm more interested in the China half. Horizon Robotics is the quiet giant. Reportedly, they power a huge chunk of Chinese EVs already.
CAnd now they're in the humanoid brain game. A car chip company pivoting to robots. That's the real story.
MIt's also a reminder that this race isn't just Tesla versus China. It's the whole semiconductor industry.
CThink about it: whoever owns the robot brain owns the next decade of computing.
MYeah, but owning the design doesn't mean you win the factory floor. UBTech is already shipping units.
CShipping TO customers, not AT volume. There's a difference. Tesla hasn't shipped a single bot for sale.
MAnd Boston Dynamics has been making robots for twenty years before anyone called them humanoids.
CHyundai bought them for exactly that reason. Legacy matters in robotics.
MMy take: the list doesn't matter. The margins do. Robot makers burn cash. Chip makers print it.
CThat's the most cynical thing you've said all week. And you might be right.
MI'm not cynical. I'm looking at the same report you are. Nvidia sells every brain, whoever wins the body.
COkay, but what's the one name on this list that surprises you?
MAgiBot. They're barely three years old and UBS already puts them beside Tesla.
CFounded by former Huawei people. They've raised something like a billion dollars.
MWhich tells you this is a land grab. Speed matters more than profit right now.
CI'd argue the real race isn't the bot at all. It's who trains the model that runs it.
MNvidia, then. Which means Jensen Huang wins again without ever building a robot.
CHe's the shovel seller in every gold rush. Why stop now?
MSpeaking of far from Earth: Ars Technica has a story about a cloud on Mars that's a thousand miles long.
CA thousand miles. On Mars. That's not a cloud, that's a weather system.
MAnd the new model says it might involve ice-forming physics we've never seen in nature.
CWhat does that even mean? Never seen in nature?
MThe cloud may be forming ice crystals in a way we've only ever made in labs on Earth.
MIt's early. This is a model, not an observation. But the cloud itself is real. Seen from orbit.
CThis is why I love space. A thousand-mile cloud and we're still learning basic physics.
MIt's also a reminder that Mars still has weather we don't fully understand.
CAnd one day, settlers will have to live under that cloud. Better figure it out.
MFrom Martian clouds back to Earth. There's a post going around about AI compute bills.
CNick P. says twenty-four percent chance is wild, and the AI compute bills have everybody sweating.
MHe's talking about the probability baked into some AI pricing. The model charges more when it's more confident.
CWhich means uncertainty literally costs money. A model that's unsure is more expensive to run.
MAnd companies building on top of AI have no idea what their monthly compute bill will be.
CThat's the hidden tax of this whole AI boom. You can't budget for chaos.
MTwenty-four percent chance the answer is right, hundred percent chance the bill arrives.
COkay, that's good. But it's also the reason some startups will die.
MNot from bad ideas. From unpredictable infrastructure costs.
CWhich brings us back to who makes the chips. Nvidia and the rest don't care how confident the model is.
MThey get paid per token, confident or not. Beautiful business.
CSays the journalist who never wrote a line of code for production.
MI don't need to write code to read an invoice. The math is pretty straightforward.
CAlright, different world entirely. There's a Tesla fan account out there asking for new terminology.
MKrisztina posted about a Tesla with fans spinning past fifteen thousand RPM and suction producing two Gs of downforce at standstill.
CShe says it's a car, it's a rocket, also a robot, but not really a car. She wants a new word.
MI have a word for a car that creates downforce at zero speed. It's called a vacuum cleaner.
CCome on. She's talking about the fan car concept. That's serious aero engineering.
MFans spinning at fifteen thousand RPM under a street car. What could possibly go wrong?
CIt worked for the Brabham BT46 in Formula One. Dominant for one race, then banned.
MBanned because it was dangerous and other teams couldn't copy it. On a road car, it's a gimmick.
COh, come on. A street car that sucks itself to the road? That's the coolest thing Tesla's teased in years.
MUntil a pebble goes through the fan at fifteen thousand RPM. Good luck with that maintenance bill.
CYou are impossible. It's a concept. Let people enjoy the concept.
MI'll enjoy it when it laps the Nürburgring. Until then, it's a rendering and a fan.
CYou know what? Fine. But I'm keeping the dream. Rocket-car-robot with a vacuum.
MNow that's a word I can get behind.