Project Suncatcher has liftoff! ☀️🚀 Announced last year, this moonshot project explores if we can one day host machine learning infrastructure in space. After years of research, we’ve just launched a prototype satellite, built in partnership with Planet, into orbit on SpaceX's Transporter-18 rideshare mission. This initial mission will... gather data on how Google TPUs handle the physical stress and extremes of spaceflight. Ultimately, these findings will help us explore how to potentially link multiple satellite constellations for scaled machine learning. By operating in low Earth orbit, these systems can tap into near-constant sunlight to generate up to 8x more solar power than on Earth.show more

Learn more ↓

@planet space always had rockets. what it never had was a boring customer with a huge power bill. AI just became that customer

@planet so the bet here is basically: solar power for AI is the expensive part down here, so just build the data center where the sun never sets

@planet damn cool I think all the predictions may become reality

@planet 8x the solar power and no landlord. the crowd with their phones up is the best part of this

@planet A promising venture, but what about the infrastructure on Earth?

@planet Describing this as a moonshot is appropriate in both function and aesthetic If Gemini 4 wrote this, y'all also won the race to taste

@planet We need more space junk out there for sure!

@planet A prototype satellite is a meaningful first step, but space-based ML will hinge on whether power, cooling, and data links can justify the trip to orbit.

@planet A fascinating step toward treating compute as a physical system, not just a datacenter problem. The engineering constraints here should produce some very interesting design lessons.

@planet i want this to work so we can start referring to Earth as on-prem

@planet thermal dissipation in orbit is going to be an interesting constraint for accelerators

@planet Great point!

@planet ai infrastructure needs space. if it wants space, give it space

@planet waiting for the latency issues when a solar flare hits the cluster

@planet Incredible to see the prototype launch with Planet and SpaceX!

@planet orbit-pilled alpha, off-grid compute is the ultimate meta now

@planet AI in space sounds futuristic, but the real test is scalability. If space-based computing can truly unlock cleaner, near-constant solar power at scale, it could fundamentally reshape the AI infrastructure game.

@planet This initial mission is essentially diagnostic: gathering data on how Google TPUs handle the physical stress and extremes of spaceflight. Whether orbital ML infrastructure is viable hinges on what this Transporter-18 flight finds.

@planet 太空跑机器学习，太酷了

@planet awesome

@planet nice, the first space data center outage would be the most expensive localhost in the industry.

@planet wen dyson sphere?

@planet Even the cloud needs a rocket. SpaceX making the next frontier of computing reachable

@planet Physics question hiding in this: why up to 8x more solar power in orbit? No atmosphere absorbing light, no clouds, and in the right orbit, almost no night. A great real-world example for any physics class, and a fun 60-sec explainer on

@planet yall evolving

@planet This is what the future of AI infrastructure looks like—from data centers on Earth to machine-learning systems in orbit. Space-based computing could completely change how we think about energy and AI.

@planet Gemini 4?

@planet Putting the accelerator in orbit is the headline; the real constraint is thermal design plus downlink latency. If the model spends its time waiting for ground-side state, is this a compute platform or an extremely expensive cache?

@planet Logika bezbłędna: na orbicie prąd jest za darmo, właściciela nie ma, a miejsca nie brakuje. Jeden problem: kiedy satelita się zawiesi, zamiast kabla od resetu trzeba będzie wysłać rakietę. To się nazywa uptime premium.

@planet This is actually insane. AI in space is coming.

@planet my power bill just flinched

@planet Running AI infrastructure in space still sounds crazy, but this is a really cool first step