21 thoughts on “Data Centers”

  1. The linked article mentions how much water data centers use for cooling. I’ve never seen a good answer on how to provide that much cooling on orbit.

      2. Ummmm…radiators?

        It’s a lot of radiator. On Earth, there effectively is about ten thousand times as much natural radiator as the entire heat output of humanity. We could build an insane amount of data centers on Earth before it starts having an effect on global climate.

  2. I just listened to the Outernet guy being interviewed on Triggernometry. While I can understand the arguments for his network; I don’t quite buy putting a data center on orbit considering how quickly storage advances.

  3. Yeah… No. The hassles of dealing with data centers in space are much the same as those for SSPSes. Or orbital missile platforms. It’s just easier to maintain and mod when on the ground. The comms bottleneck will also still be an issue. And the further out you go the bigger the latency issue kills you.

    1. Anent use of the data processing capability of notional orbital data centers for Earth-bound applications, I think the “hassles” are far worse than those for SPSes. The technology of the latter is not advancing at nearly the same rate as that of data centers so near-term obsolescence is not nearly as much of an issue. “Hassles” that both have in common are related to largeness of size – for the radiators of a data center and of the energy collection surfaces for an SPS. The main ongoing enemy of both would be space debris strikes.

      I am baffled as to why you would include orbital missile platforms on this problematical applications list. Such platforms will be miniscule by the standards of either orbital data centers or SPSes, will have no huge power requirement and will also not be subject to quick obsolescence based on the advance of technology.

      In the fullness of time, there will be far more of everything in space – including data centers, power generation infrastructure and even people – than there will be on Earth. But most of that will have little or nothing to do with Earth directly.

      1. I am baffled as to why you would include orbital missile platforms on this problematical applications list. Such platforms will be miniscule by the standards of either orbital data centers or SPSes, will have no huge power requirement and will also not be subject to quick obsolescence based on the advance of technology.

        Dick,
        I mention it simply as a past application that was proposed as something desirable to be placed in orbit. The mistaken twin beliefs that it would render them invulnerable to attack as well as shorten warhead delivery time. Of which it does neither. Plus back in the days of liquid fueled ICBMs it is a maintenance nightmare. Solid fueled missiles aren’t as bad and why we switched. But all still need maintenance which is hard to do if they are in orbit. I wasn’t thinking of it in terms of comparative scale.

        The concept was canned at birth by those in the know. Regardless of what 2001: A Space Odyssey would have you believe.

        I mention it because of these similarities. For Earth servicing data centers its not only a question of maintenance but also upgrading obsolete gear. Not to mention PC&C (power, cooling and Communications). Or radiation hazards.

        No… As you say, it doesn’t make sense for an Earth-facing service. Now as space infrastructure, that is another question. But not the one proposed in the OP. At least not what I think of as a ‘traditional’ data center.

  4. Yeah, data centers in space, dumb idea just like a nuclear reactor on the moon. Notice how all nuclear power plants on earth are built by cooling water. Cooling towers… hello. You can roll out many megawatts of solar panels on the moon for 1% the cost of nuclear.

    Space nuclear power advocates rationally recognize the radiator problem so they want to use a Brayton which has higher heat rejection temperature thus smaller radiators. You can’t run your data centers at 300C. Plus how do you upgrade to the newest best Nvidia chip.

    1. The driver for nuclear power on the moon is the two weeks on the night side, where you have to rely on batteries, which are an extremely heavy payload for what they store. Without nuclear at last keeping systems warm, many probes don’t even survive the night.

      For example, the later LEMs had about 75 kWh of energy for a 35 hour surface stay, with over 90% provided by fuel cells and the rest by silver-zinc batteries. Lithium ion batteries weigh six to eight kg per kWh, so they would weigh 450 to 600 kg to provide the same performance. But the night lasts two weeks, and solar cells aren’t going to contribute much to large settlement for a couple days after sunrise and prior to sunset, so a battery system would probably need to handle an 18-day stretch, which is 432 hours. Lithium ions with that capacity would need to be almost 1000 kWh to provide the LEM with power for 18 days, and would weight 6 to 8 tonnes, which is more than the entire ascent module (4.5 to 4.7 tonnes).

      So without nuclear, most of the weight of a lunar station is the batteries to get it through the night.

    2. Yeah, data centers in space, dumb idea just like a nuclear reactor on the moon.
      I don’t see this analogy at all.

      1. Plus how do you upgrade to the newest best Nvidia chip.
        This makes far more sense. I suppose if access to orbit becomes as routine as transcontinental FedEx flights, you could make a case. But there are other issues.

  5. Soft errors from cosmic rays plus micrometeorites.
    The solar panels on the Moon will be 100% reliable. They will produce nothing at night.
    The waste heat from the nuke could probably be rejected by drilling into the regolith and rejecting it there.

  6. To get even a fraction of the worlds data center capacity on orbit (like 1%) you would need to increase Starship launches to more than 10 per day for years. The mind boggling scale of data centers across the globe is bonkers. Years ago, I worked at AOL and their data center in the DC area was massive. Just getting that old relic to space would require thousands of tons to be launched. And that is one center. And by today’s sizes, a small one at that.

    Powering the servers is one thing. Powering the heat removal systems is quite another. Space is not ready for it.

  7. There are some really cold places in Canada and Alaska. They must be easier to build out than space. That they’re not building server farms at the end of the highways north suggests adjacency to workers, network and power grid are more important than reducing the cooling load.

    1. There are data centers being built in the Scandinavian north for this exact reason:
      https://www.datacenters.com/news/are-we-overlooking-the-nordics-europe-s-silent-data-center-powerhouse

      Generally better infrastructure than Northern Canada or Alaska. Also lots of hydro.

      They’re talking about data centers consuming gigawatts. Considering how much engineering goes into managing teh few hundred or so watts in satellites, this will be non trivial, not to mention the gigawatts to start with.

      The human brain consumes around 20 watts, what’s missing from that picture?

Comments are closed.