This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
Space is also covered by various international treaties, and nations on Earth would be quick to update any such treaties to cover datacenters even more strictly if companies were using such datacenters to flaunt the law of countries. And at least in their current state, these treaties basically make satellite operators subject to the laws of their host nation.
Long distance oceanic LNG shipping is about 3 to 5 cents per KG. I imagine it would be cheaper when you're just shipping out to international waters from the coast. Starship's most optimistic projections for price per kg to low Earth orbit is a little under $100 per kg, but could end up closer to $1,000 per kg.
As opposed to the cost of sending people into space to fix the space datacenter? Most space DC proposals I've heard have actually proposed not having any human maintenance at all because of how expensive it would be, instead opting to add extra redundancy for essential components and just writing it off when a GPU or PSU fails.
Please, please, please show me your math for this. Even use the $100 per kg to LEO price if you want to make itas favorable to your argument as possible. I would be willing to bet you $100 donated to the charity of the winner's choice if you math shows it being cheaper than a container ship (or oil platform type structure) 370 KM off the coast (i.e. in international waters) of some LNG processing hub.
People didn't do some very basic pricing math with Solar Roadways and similar grifts. Solar Roadways would cost a metric ton more per km than just using asphalt, and a metric ton more that just sticking solar panels in a field or the desert or on a rooftop somewhere etc., while being a worse road surface and producing less power.
Space datacenters face similar economic disadvantages, and none of the proponents seem to be saying anything about the financial math here. I personally find the discussion of stuff like "How are you going to cool it?" irrelevant and a distraction, except inasmuch as they affect the cost. Cooling the datacenters is absolutely feasible, but it definitely complicates the engineering (and drastically increases the amount of material that has to be launched) far above and beyond what would be needed for a nomal terrestrial datacenter or some of the other exotic options I mentioned.
First, you have to determine what is a minimally-viable node in each context. SpaceX is proposing essentially a single-rack node with 120KW power at a mass of about 2 tons. Let's assume that the same Nvidia racks would be used in an oceanic platform, so we can disregard the silicon costs. Starlink satellites are constructed at scale at a cost of about $1M/ton, so a reasonable cost estimate for Starmind satellites is about about $2M per satellite. Add $200K in launch costs and we're about $2.5M/node up front, with ~0 ongoing costs.
If we assume that solar arrays are impractical for oceanic data processing, the minimum viable node would have to be some kind of hull with active station-keeping and enough fuel storage to fuel a diesel generator and station-keeping for extended periods between refueling (30 days?). It starts getting sketchy here, but working with requirements of about 25 tons fuel capacity, it seems like you're looking at a 30-40 meter DP1 vessel. I couldn't find costs for new construction, but listings for similar class vessels decades old are around $3M (e.g., https://maritimesales.com/DAB17.htm), so that seems like a reasonable conservative estimate. And this the up-front cost only. Assuming it's autonomous, it will still need monthly fuel deliveries, regular PMCS and overhauls on engines, gensets, and thrusters, other seaworthiness maintenance like painting, cleaning, and lubrication, and you can expect substantial wear and tear and damage from environmental forces. Fuel replenishment alone is going to be at least $30K/month. And this is all for a single Nvidia rack!
Now of course as you start scaling up, the economics shift, but my point was that one of the advantages of orbital deployment is the ability to scale node sizes down.
If you're scaling things down to a single rack, then for $2.5 million I'll gladly stick it in my basement (which has gig fiber internet, I'll upgrade to the 2.5 gig plan if they'd prefer for that kind of money) and handle all maintenance for them.
If the point of space datacenters is being able to do them at a very small scale, there are a million better and cheaper options that don't run into the same sort of political/NIMBY resistance that big datacenters have. There's plenty of vacant office buildings with good internet and electrical hookups that would be far cheaper and easier to maintain than chucking a rack into orbit.
The economics of this literally make no sense, there's no point in doing this stuff at a small scale because you lose all the benefits of economies of scale.
I'm skeptical you have 120 KW service to your house
True, I probably only get about 50. I'll take a mere $1 million to host a third of a rack instead, after all scaling these things down doesn't matter ;)
Would fit better under my basement stairs at that size anyway.
Also, I don't think your calculations for Starmind accounted for the solar panels, radiators, etc. which are far more than a normal Starlink satellite would need.
Assuming the minimum viable product can even scale that small, what you're describing is not a lot different than what cryptominers were doing at home for a while, and their activities were often extremely unpopular with their neighbors. I'm curious how you plan to dissipate 50 KW of heat from under your basement stairs without annoying your neighbors. Not to mention, the $7.5K/month power/fiber bill takes a pretty big bite out of that after a 5 year operation cycle. After 10 years, you're probably not breaking even.
Interesting concept though: assuming you want to add, say, 2GW of inference capacity, that would be something like 40,000 1/3-size racks. Could you spread that out around the country enough that the increased load could be absorbed by the already-present electrical grid and generation capacity? The national average looks to be around 500 GW, so this seems plausible. I suspect the challenge would be getting people to sign up for having the noise in their homes.
Datacenter hardware lifecycles are usually 3 to 5 years so they can just give me another million and a replacement third of a rack every 3 years.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Space datacenters don't have to be cheaper than ground datacenters, they just have to be cheaper for processing space data than ground datacenters. The latency and bandwidth of space-to-grond will make space datacenters naturally attractive as space scales.
That's maybe a trivial compared to the actual core question, which is whether space datacenters can make sense for processing ground data. I think that's an open question. But the economics will become known as space datacenters are built to process space data.
What are you smoking? Firstly, there's hardly any space data other than scientific probes by various space agencies, and communication satellites operated by both public and private entities. The latter, by their very nature, need to communicate with the ground.
Secondly, the latency difference would depend largely on where both the datacenter satellite and the source of the space data are located. In LEO, space datacenters would be less consistent in their latency than ground stations because they're orbiting the Earth roughly every 90 minutes (or you'd have to be constantly passing the data around to different space datacenters to keep latency somewhat consistent, but this data transfer would likely kill any gain you got from the latency reduction and then some).
This is not true and hasn't been true for over a decade. There are ~13,000 of satellites in space now. They are constantly collecting data and monitoring events. Even in the 2010s Planet Lab was able to scan every part of the Earth multiple times a day, and this was with a fleet of a few hundred satellites. Starlink now processes large amounts of internet traffic, set to increase enormously. Video sensors lidar sensors optical sensors infrared sensors synthetic aperture radar sensors particle detectors magnetometers gravitometers the works. All these systems are constantly collecting data. All these satellites are at risk of running into each other, or colliding with space debris, or have regular system checks that get reported back to Earth. And the number of satellites in orbit is going exponential as SpaceX and Rocket Lab make the costs to launch cheaper than ever before.
Even just the imagery needs are enormous. Planet Labs for example processes something like 30 Terabytes of imagery a day. I don't know how they do it, but it's not hard to imagine the constraints. Even the most basic Planet Labs Dove 1 and Dove 2 CubeSats from the early 2010s didn't report every image they took back to ground antenna networks. The stress on bandwidth would be enormous. What would happen even then is that each CubeSat would take a multitude of images and algorithmically select the best with which to phone home. Now consider that as a satellite moves across space, images of the same location on Earth are taken from different satellites. Deciding which images to send (which videos, which reports, which temperature data, etc.) can become a giant coordination problem. And there's no special reason why that wouldn't be done in space: it's simply a question of convenience.
A datacenter is just a warehouse of computers. There's nothing special about it. Datacenters process and manage data storage and dissemination for other computer. This is already happening in space. In a trivial sense we already have datacenters in space. We just don't call them that, because the ones in space are not as large and generalized as the ones we have on Earth. But there's not really any serious debate that datacenters in space are going to grow. The question is, at what scale? Will orbit-bound datacenters grow into a significant computer industry, or will they stay at this trivial level? Will they ever be efficient for processing data that originates from earth, or will they only be useful for data that originated in space? These are open questions.
This is basically how Starlink already works. When you connect to Starlink it switches which satellite you're connected to every few minutes.
With Starlink or a similar system I imagine it would be easy to beam a lot more of it down.
Yes, which is why putting them in space has very limited advantages that rarely outweigh the benefits of a terrestrial datacenter.
Starlink works by just handing off a connection/stream to a different satellite. It doesn't need to pass off 30 terabytes of image data that are already partially processed, it's a fundamentally different problem.
You mean data-collecting satellites could outsource their data processing to specialized satellites that coordinate intense computing problems for them?
Nope, at least not easily. But they can outsource their data transmission (which takes orders of magnitude less energy and computing power than actually processing it) to Earth.
I really don't think this was the gotcha you thought it was.
A satellite that coordinates transmission scheduling is functionally a data center. It probably even runs SQL. We’re simply debating how far it will scale.
Starlink is useful because it allows internet from anywhere in the world. It needs a certain minimum amount of computing power onboard to be able to accomplish this. A 5G cell tower, by constrast, only provides internet in a limited radius around the tower.
Building a datacenter in space offers no significant economic, legal, or technological advantages over a datacenter on the Earth.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link