@ActuallyFinance But a very low, less than 1%, chance of this ever happening and only losing 5 years at most if it does doesn’t necessarily mean this kills the idea I would think? Not trying to downplay the risk but I think the company would survive?
@ActuallyFinance I assume to at least some extent it’s less of an issue due to the 5-year lifespan. If literally every single sat came down and had just been launched you would lose 5 years of economic life and have to launch them all again. Huge problem dont get me wrong
@ActuallyFinance Not fully sure. AFAIK they generally turn themselves to the side as to not be directly facing the sun during solar storms but I’m not sure if there’s some cutoff where that is not enough to protect them.
@ActuallyFinance@JerryCap Makes sense. Annualized cost to deploy a MW is equal with earth is what I mean. Revenue if they rent it out may take longer to equalize so doesn’t necessarily mean returns will be equal or better.
@HeyDanitsme@ActuallyFinance But could it truly provide 24/7/365 power at a total lower cost? 90% uptime would be like 50% less than true 24/7/365. But true full coverage is insanely expensive bc there will be one rainy week where you get barely any sun
@HeyDanitsme@ActuallyFinance I did the math on how much it would cost to power a 1 GW load in TX and found you would need something like 20 GW of solar and like 15 GWh of batteries bc of intermittence of solar. Costs have to fall a huge amount to make it cost effective
@HeyDanitsme@ActuallyFinance@JerryCap China’s been trying to do this & people have talked about it for a bit. But economics only really improve enough when you have significant enough demand to spread development costs over enough sats. I think it’s too small of a market w/o serious demand (like AI)
@ActuallyFinance@JerryCap W/o starship tho costs are impossible. And you have to assume SpaceX can continue dropping satellite costs (which they have done at ~20% per year on an adjusted per MW basis. But SpaceX is extremely good at this and I think starship is more of a when not if.
@ActuallyFinance@JerryCap I’ve been doing work on them for the past 3 months and started super bearish but the deeper I go and the deeper I get into costs the more I think SpaceX could def pull it off w/in the next 10 years. You def couldn’t use for training, only inference. But all other issues are less
@ActuallyFinance@JerryCap Serious than I expected. But yeah a GW of compute will fundamentally be less valuable than a GW on the ground, but as bugling on earth gets harder and starship brings down launch costs (assuming it works lol) I think costs can reach parity much quicker than expected
@Jamesmanny21@JohnTinsman@wholemars Yes numbers I gave were on a per GW basis. And yeah they may have gotten a good deal but still the number is for everything other than compute and it is still very low and impressive
@Jamesmanny21@JohnTinsman@wholemars Obv that is not what I am saying retard. That includes all the cooling, the networking, generators, EVERY OTHER expensive component. It is not just a warehouse, clearly u have never been to a DC. And no the rental rev is not just for colossus 1, it is for some of 2 too. Check S1
@ShanuMathew93 It’s excluding chips right? Like that is the value of the shell (and everything else needed other than chips) right? For most companies that’s like 12-15B per GW
@Jamesmanny21@JohnTinsman@wholemars I’m sorry but do you think they bought ~30-40B chips for $3B and then got all the other stuff for free? Why do you think their capex for XAi was so much higher than that number would imply???
@Jamesmanny21@JohnTinsman@wholemars You clearly know nothing about DCs bc this is how people in the industry talk about them. You talk about “shell” capex (everything other than compute) and compute capex separately bc you can’t compare costs across compute as it depends on type of chip.
@Jamesmanny21@JohnTinsman@wholemars Tf are you talking about? No it still includes all the electrical, cooling, networking, backup generators, and everything else a DC needs, it just excludes the racks themselves. To be clear other operators generally have to pay ~12-15B per GW for everything other than chips
@mxschumacher@when_to_fold_em But none of this matters as much as cost. I think SpaceX has a somewhat straight shot at getting costs lower than terrestrial even after accounting for shorter lifespan, launch, etc.