Comment by ggreer
1 day ago
If you read the Starcloud whitepaper[1], it claims that massive batteries aren't needed because the satellites would be placed in a dawn-dusk sun-synchronous orbit. Except for occasional lunar eclipses, the solar panels would be in constant sunlight.
The whitepaper also says that they're targeting use cases that don't require low latency or high availability. In short: AI model training and other big offline tasks.
For maintenance, they plan to have a modular architecture that allows upgrading and/or replacing failed/obsolete servers. If launch costs are low enough to allow for launching a datacenter into space, they'll be low enough to allow for launching replacement modules.
All satellites launched from the US are required to have a decommissioning plan and a debris assessment report. In other words: the government must be satisfied that they won't create orbital debris or create a hazard on the ground. Since these satellites would be very large, they'll almost certainly need thrusters that allow them to avoid potential collisions and deorbit in a controlled manner.
Whether or not their business is viable depends on the future cost of launches and the future cost of batteries. If batteries get really cheap, it will be economically feasible to have an off-the-grid datacenter on the ground. There's not much point in launching a datacenter into space if you can power it on the ground 24/7 with solar + batteries. If cost to orbit per kg plummets and the price of batteries remains high, they'll have a chance. If not, they're sunk.
I think they'll most likely fail, but their business could be very lucrative if they succeed. I wouldn't invest, but I can see why some people would.
You can also drink from a shoe. It's absolutely possible.
And it enjoyed some popularity. [1]
[1] https://en.wikipedia.org/wiki/Beer_boot
And there you've cut to the chase.
I was implying an unspoken obvious "but why would you?"
But of course the answer I missed was you don't, you make money from people who, for whatever reason, want to drink from shoes.
> For maintenance, they plan to have a modular architecture that allows upgrading and/or replacing failed/obsolete servers. If launch costs are low enough to allow for launching a datacenter into space, they'll be low enough to allow for launching replacement modules.
This is hiding so, so much complexity behind a simple hand wavy “modular”. I have trained large models on thousands of GPUs, hardware failure happen all the time. Last example in date: an infiniband interface flapping which ultimately had to be physically replaced. What do you do if your DC is in space? Do you just jettison the entire multi million $ DGX pod that contains the faulty 300$ interface before sending a new one? Do you have an army of astronauts + Dragons to do this manually? Do we hope we have achieve super intelligence by then and have robots that can do this for us ?
Waving the “Modular” magic key word doesn’t really cut it for me.
> Whether or not their business is viable depends on the future cost of launches and the future cost of batteries. If batteries get really cheap, it will be economically feasible to have an off-the-grid datacenter on the ground. There's not much point in launching a datacenter into space if you can power it on the ground 24/7 with solar + batteries.
Something tells me that the price of batteries is already cheap enough for terrestrial data centers to make more economic sense than launching a datacenter - which will also need batteries - into space.
The cheapest batteries today are around $100/kWh. Optimistically assuming 12 hours of sunlight per day, a 40MW datacenter would need 480MWh of batteries to cover the dark period, costing $50 million. Realistically you'd need at least 16 hours worth of batteries to cover winter months when the night is longer, raising your battery costs to $65 million. You'd also need ≈5x more solar panels than in space, and these panels would be more expensive due to shielding from weather. 120MW of ground-based solar panels would cost around $100 million.
Assuming the $165M of panels and batteries last for a decade, and there are no maintenance costs, they'll provide 3,504,000MWh over that time for an energy cost of 4.7 cents per kWh. This is competitive with grid power in some places. It also has the advantage of not needing backup generators. But maintenance costs do exist, and it makes more financial sense to buy power as you use it rather than pay upfront.
3 replies →
Same with hydrogen fuel cell vehicles, inventing a detour because it sounds cool and ultimately don't work out because Occam's Razor.
Hydrogen fuel cell vehicles are a little different because there are safety, economic, and physical constraints that mean they will always be worse than battery electric or gas cars. I and many others correctly predicted this.[1] For hydrogen to succeed, batteries would have to get more expensive and/or have worse energy density than they did in 2015.
Satellite data centers seem unlikely to me, but at least their feasibility doesn't require that existing stuff get more expensive/worse. Starcloud is a bet that three things will happen in the next decade:
- SpaceX Starship will succeed and drastically reduce launch costs.
- Batteries will not get 10x cheaper.
- There will be valuable applications for high latency, high performance compute (eg: AI training).
If any one of these things does not happen, Starcloud is doomed (or will have to pivot). If they all happen, Starcloud has a chance at success.
1. https://news.ycombinator.com/item?id=25875749
Let me alert all the NIMBY folks, let them know that data centers will be blocking their view of the moon and casting shadows on their backyards.
Just give all those astronomers mockup telescopes with little screens creating the fancy images they want inside them.
They will calm down.
Of course, they're soft targets in space war too, they could generate lots of debris.
Man I read “AI training in a high latency self sufficient satellite orbiting earth” as the start of a Sci-Fi novel…
Attach that satellite to all the imaging satellites, what's it going to be learning about?
Just another good proof of paper being an ideal medium for fiction
Any purported advantages have to contend with the fact that sending the modules costs millions of dollars. Tens to hundred millions