Hacker News Reader

Comment by digitaltrees | Hacker News Reader

No. He is making it an AI company. The prospectus makes that clear. Everything is in service of training and deploying AI. Twitter is data, distribution and marketing, space x is distribution with data centers and internet in space. Cursor is training data and hostile distillation.

View on HN

That prospectus didn't make anything clear. It was pictures of rockets and rubbish about "the light of consciousness". The only real information was buried deep in the middle and it showed a company with poor finances and no clear path to success. Certainly not something worth the likes of Amazon.

I agree with you. But also think the strategic positioning of being fundamentally AI was clear.

Unfortunately computers generate heat, so while Elon can build a business running on hot air on earth, he will not be able to do so in space.

This is very far from an obstacle, space-functional radiators and thermal management systems do exist.

KW scale radiators for GW scale compute demand. It is not based in reality.

it actually is not that impossible, and in fact can be done in a form factor not that different from Starlink satellites: https://www.youtube.com/watch?v=FlQYU3m1e80

I didn't say impossible, I said not based in reality. Why would you do this? What economic advantage is gained?

The radiators on the ISS radiate maybe 200kW all up. That's _one rack_ in a normal earthbound DC.

How much mass do we need to put in orbit to radiate 10 GW of data center heat? How much rocket fuel does it take to lift the rocket fuel required to lift that mass? Which part of this is cheaper or more effective than building data centers 100 feet below the ice in Antarctica ? Other than serving workloads that are already in orbit, I don't see the point. How much addressable market do we have for AI compute that is already in orbit?

I’m not even worried about the weight and heat. How do you do a disk swap or other service on a machine in a datacenter in orbit?

The appeal is solar power efficiency gains. And yes it is a notable amount of extra mass, but not a deal-breaking amount. Latency is pretty minimal regardless of location (sub-100ms) which is not nothing but it's minor for things like serving AI inference.

Not saying I think it's the best idea, but it is theoretically feasible.

When you say it is not deal-breaking - have you looked at any math? Real world data finds we need about 6kg of radiator surface to radiate 1kW. [0] . A 100MW datacenter would need about 600,000 KG just of radiator mass. ISS was only 450,000KG and it took 13 years to assemble. Meanwhile, we're projected to launch 600X that data center capacity on earth in just the next three years.

To say that this is pure folly is an understatement. It is not folly, and it is not foolishness. It is fraud.

One of the most cited papers: https://starcloudinc.github.io/wp.pdf

Does not contain the word "nuclear". Instead its just breathless gushing about how much more efficient the solar power is.

[0] https://arc.aiaa.org/doi/epdf/10.2514/1.A35030

Radiative heat transfer exists in space. In fact, it works quite well for cooling as long as you’re in the shade.

It exists, but "works quite well" is only for a small amount of heat. It takes a big radiator to get rid of heat to ax vacuum. Much easier and cheaper to just keep the computers on earth and cool them with standard air or liquid cooling setups.

The required radiator for cooling isn’t that much larger than the required solar panel for powering the thing in the first place, and you don’t see everyone saying those are impossible. Is it easier to keep them on the ground? Obviously, but that wasn’t the claim. It just isn’t nearly as hard to cool things in space as a lot of people seem to think.

Under rated comment. May be legendary some day.

Are we to assume that nobody at the company who had launched more satellites than the rest of the planet combined is aware of this problem?

NewBird AI is an AI company which had an extremely irrational stock price bump because it's all about AI. It AI'd its AI with some more AI to AI harder with AI... and that equaled money.

If you're writing a prospectus right now and want a lot of large institutional investment you put AI between every letter. We are in a bubble, I don't think there's any disagreement about that, when the bubble will pop nobody knows - but while we're in this bubble everything is AI. I think it's unwise to read that prospectus in good faith given all the other factors (the court case against OpenAI, the race to IPO first, Google's additional stock grant, Elon's history of corporate bailouts) that are pretty plain to see.

> Cursor is training data and hostile distillation.

What is "hostile distillation" please?

So what are the rocket ships for?

To launch the data center into space and the satellites that deliver the data to users over star link. The same reason standard oil owned railways. Controlling distribution is a competitive moat no one else can replicate. Only antitrust stopped the standard oil monopoly. Not market forces.

Sounds like a convoluted and expensive way to pollute our skies for generations to come.

maybe he secretly wants the kessler effect?

Counterpoint: what if he lied in the prospectus? With Elons proven track record of misleading investors and customers I think that possibility has to be considered.

Reply on news.ycombinator.com