Amazon CEO, Jeff Bezos: Within 10-20 years, the giant, gigawatt-scale data centers for AI training clusters will be built in space This would leverage 24/7 solar power and make space a key part of Earth's industrial and technological infrastructure

Making literally everything harder and solving nothing. Imagine trying to cool a data center in a vacuum 🤦

So will Amazon shift to offshoring Indians to space? Asking for a friend….

No rain in space, but you have to deal with hyperspeed particles! I’m very skeptical about the feasibility of data centers in space. There is no way to service them; so they would need additional levels of redundancy that land based would not.

Skynet approves

What a dumb take. So unnecessary to send mass into space.

If you believe this, i have a space-rock to sell you for a billion dollars. Deal of a lifetime, I contains 9999 trillions of worth minerals, ripe for the picking.

The solar panels would be sun side oriented, while the data centre would be in the shade With deep space offering a 2.7 kelvin. Even in the earth's shade, the temperature would be a chilly 120 kelvin, a good temperature to stabilise the gpu heat. The trick would be to avoid additional complications of cooling infrastructure but use only deep space for heat dissipation. ISS with a 3244 square mt array generates 215 kW in direct sunlight, which is not enough for power guzzling data centres. A factor of improvement would be needed to make this feasible. Optical inter satellite links are already operating at up to 100 gbps, commercial downlinks, and uplinks speeds are creeping upwards, while NASA's TeraByte InfraRed Delivery (TBIRD) achieved 200 Gbps in April 2023. Geostationary lift costs are still the most expensive component at 20k to 35k per kg. New dense and light weight cluster development is the main blocker for a space based data centre. Redundancy and radiation protection is a major challenge. Radiation hardened chips don't come cheap and the whole point is to make it a cost effective solution. As of now, the math may not be mathing, but @elonmusk with Starlink, Grok, and SpaceX has the components in place

Bro I cant

The most idiotic thing ever. Between everything else (like cooling) how to hell you fix issues when your "datacenter" is in orbit ? You send trained monkeys by space rocket to replace broken HDDs ?

These people are fucking delusional. No one tells them no, and they are incapable of thinking completely through the problem anymore.

I doubt it will ever be built. Millions of GPUs require gigawatts of energy. 1 GW require several square kilometers of solar panels. So for 5 million GPUs you need around 20 km2 of solar panels and it would require over 5000 Starship flights to bring all that equipment to space. Fusion will sort this out. Moon is not bad option if latency is not a problem

Imagine investing in the only reliable publicly traded rocket launcher before something like this were to happen… $RKLB

Glad to see Jez taking notes.

Not gonna happen. You can only do ineffective radiative cooling in space. people have done the calculations, its not viable because of insufficient cooling.

With teleoperated Optimus and SpaceX rockets, we could probably start building them now. Why wait a decade?

Heat dissipation in space is actually harder than on Earth - no convection, only radiation. Data centers generate enormous heat. You'd need massive radiator arrays, adding weight and complexity. This sounds impressive in a keynote but falls apart under engineering scrutiny.

Timing is everything $RKLB!

Tesla's space beams, but instead of energy it is information

How do I leverage myself to be ready to build these space data centers? I want to hold my clipboard in a spacesuit

Anyone who talks like this doesn't realize how hard it is to shed heat without an atmosphere.

This is inevitable, and the proposed timeframe looks feasible since the necessary transportation and connectivity are mostly already in place — @SpaceX has the #Starship and #Starlink programs, and @amazon has @blueorigin New Glenn and Project #Kuiper. Low Earth orbit is easily accessible and offers advantages in cooling, energy production, and scalability. The major question here is the design of the next generation of data centers — space data centers. The #Skylab size looks almost perfect for this purpose, as it allows for the combination of the necessary engineering systems (liquid cooling, heat radiators, accumulators, solar panels, etc.), space for robotic and human maintenance, and the ability to hold up to ~100 25 kW racks with a PUE of ~1.10. However, it is also clear that such a data center would require a highly efficient, small-scale nuclear reactor of more than 1 MW.

whooaaaaa so the intermittency issue of solar is solved in space?? No batteries? That's actually the coolest thing I've ever heard of.

He is basically admitting that the current push towards AGI is failing because they don't have enough power to do it. Hopefully some saner people, maybe Chinese and Russian, will keep on working on reasonable base AI models that anyone can download and train with their own specific set of documents to answer questions about subjects that they are interested in. We will make more progress if 100 physics research groups attempt to train an AI that will solve physics problems. And the same in chemistry. The same in metallurgy. ETC.

The logistical capability to move all that material into orbit doesn't currently exist at any level that would enable this idea to make sense economically. Maybe that will change in the near future. Space elevators would do the trick

So this is why inflation hit so hard, people like this need more money to accomplish their own goals while everyone else starves. Let’s hope at least this data centers are used to find cures to cancer and not just store payment information.

@grok what about temperature and cost of transportation/maintenance, etc? evaluate this idea.

This is how $TSLA will be able to create generational wealth in the near future.

Jeff should probably focus on the price of his underperforming stock instead of grandstanding to an audience of clapping Seals. The ROI would be negative into the Billions of dollars. You’re talking about something that would be 10’s of millions of pounds at least in Space? DUMB!

What a clown, we have fusion, Jeff. Either you are fibbing or an idiot.

Nonsense! Space grade chips are more complex and expensive. That's not economically viable. What is actually are space solar panels that can beem down energy to the facility on the earth.

It’s a great place to mine Bitcoin too! It’s the only tech that monetizes orbital solar energy into digital value. That digital value (Bitcoin) can be sent to Earth, Mars, and Space Stations! This lays the groundwork for interplanetary money! It’s the only tech that has a unique advantage over Orbital Data Centers! @IntercosmicNRG is working on it now! Check it out!👇

They will have a hard time getting radiation to no fuck up their processors.

Data centers are like the worst thing to put in space...

no air cooling and no liquid cooling, how to get away the heat from those gpu, you tell me?