3 ms·
It's extremely unlikely that companies will find no locations that let them build zero power zero water data centers. And bulk AI stuff isn't latency sensitive
by Dylan16807 3mo ago
It's extremely unlikely that companies will find no locations that let them build zero power zero water data centers. And bulk AI stuff isn't latency sensitive either so the acceptable build radius is huge.
- lokar 3mo agoExactly. This whole thing is not about being able to build whatever they can afford, it’s about being in a (stupid) race to build it as fast as possible.
- Dylan16807 3mo agoEven for ultimate speed I don't see how it works. Facilities to make and launch satellites add a lot of delays and need their own permits.
- rayiner 3mo agoThe queue to hook up to the Texas grid is 3-5 years long now.
- lokar 3mo agoIt’s long been 2-3 years nationwide (for redundant transmission grid connections).
- Dylan16807 3mo agoGood thing this is about off-grid datacenters.
- rayiner 3mo agoBuilding an off-grid data center is much easier in space where sunlight exposure is predictable and not affected by weather.
- Dylan16807 3mo agoThat's an advantage. "Much easier" overall is quite a claim.
- mayama 3mo agoMost locations that could provide guaranteed power for western companies with stable governments are also full of NIMBY or NGO who could easily blockade those datacenters. Then you'll have to run to offgrid locations where you have to build 24hr battery for solar with closed loop cooling. Is 24hr battery+solar on earth cheaper than LEO satellites in constant sun without batteries? That's the question that would make or break satellite datacenters, as political climate against AI datacenters is increasingly getting bad with increasing electricity prices.
- Dylan16807 3mo ago> Is 24hr battery+solar on earth cheaper than LEO satellites in constant sun without batteries? I'll be shocked if it isn't.
- rayiner 3mo agoCost of a full H100 rack is approaching the price to shoot it into space at $1,500 per kilogram.