Moving The Cloud to orbit

[Ike17055]

Lots of media discussion these days about the dramatic growth in data center construction and the impacts, particularly the demand on the power grid and land consumption. Sporadically, references appear about power generation in space and low latency communications to/from LEO. Where does this leave the idea of mitigating terrestrial impacts of data centers by moving data processing and storage -- "the cloud" -- above the clouds so to speak...

How far off are we from the types of technical achievements that would allow moving the cloud to LEO?  Obviously some big challenges: power generation (nuclear, solar) as well as data security, connectivity, etc.  Are there resources worth tracking or companies worth knowing about in this area?

[Paul451]

Power is only one issue. Heat dissipation is probably the bigger killer.

There's no real advantage to putting data centres in space. It doesn't help with the problems that data centres actually have/cause and adds a crap ton of new problems.

[VSECOTSPE]

http://server-sky.com/

[sdsds]

Once the internet backbone network with the highest bandwidth, greatest redundancy and least likelihood of backhoe-induced failure is in LEO it makes sense to think about servers in LEO as well. Until then, not so much.

[Coastal Ron]

Not a new idea. I knew a guy about 10 years ago that had started a company to do this, but of course launch costs, technology, etc. has changed since then.

But I really don't see the problem that this solves, unless the need for the data and processing is in space...

[edzieba]

Once the internet backbone network with the highest bandwidth, greatest redundancy and least likelihood of backhoe-induced failure is in LEO it makes sense to think about servers in LEO as well. Until then, not so much.

That's going to be a while. Starlink laser ISL links are ~ 200gbps, which is chump-change for the multi-terabit fibre links readily available for groundside datacentres, and without having to deal with the constantly-changing-routing headache of going via a satellite constellation. And on top of that, all your end users are groundside, so no matter how many ISL laser heads you attach to your datacentre the bandwidth bottleneck will be the space-to-ground RF links - and if you aggregate enough of those to not be a bottleneck, you can use that same aggregation with groundside datacentres too.

In short: unless the consumers for your datacentre's output are all also in orbit, then putting your datacentres in orbit is a waste of money and extra complexity for no benefit.

[Eric Hedman]

Not a new idea. I knew a guy about 10 years ago that had started a company to do this, but of course launch costs, technology, etc. has changed since then.

But I really don't see the problem that this solves, unless the need for the data and processing is in space...

The problem it would solve is the expected massive power demand for growth in data centers if it would be practical.  With the massive growth in AI expected all power sources that are ground based would not have the capacity for the expected need.    Last year in Ohio, the power companies said they had inquiries for connections to new data expected to be built in the next few years that would require an extra 50+ Gigawatts of generating capacity.  Math comes into the picture, There is no way that anywhere close to  50 Gigawatts of generating capacity will be built for Ohio in the next few years.  Add in that no one knows how many of those data centers will actually get funding.

Dell just released their first computers that have a Neural Processing Unit (NPU) in them that can take over some of the AI tasks that currently need to run in the cloud on AI data centers.  It is just currently guesswork if that will significantly reduce the expected demand for AI data centers.

There is interest in data centers in LEO because there is currently no practical way to power all the expected data centers with wind, solar, coal, natural gas, nuclear, etc if the high end projections for demand come true.  You couldn't build the generation capacity quick enough.  Whether LEO data centers make any sense depends on all the engineering numbers making sense in bandwidth, launch costs, etc.  Right now it is technically possible to build a small data center in LEO and and connect it to the internet.  Just like with Space Based Solar Power, the devil is in the details to see if either make any sense.  I don't think anyone has closed the business case for either just yet.

[sdsds]

That's going to be a while. Starlink laser ISL links are ~ 200gbps, which is chump-change for the multi-terabit fibre links readily available for groundside datacentres,

Well yes, you're right. A local internet exchange here in Seattle is handling 3 terabits per second. (That's with 4 member-facing 400GbE ports,
199 member-facing 100GbE ports and a smattering of others.) So 200Gbps links are as you say chump-change. What is limiting the datarates on the Starlink ISL links?

And on top of that, all your end users are groundside

This is less convincing. Fiber-connected users at the edge often still have a tough time getting more than ~10mbps, yet they can easily consume many multiples of that between their cloud compute instance and their data storage service. Even with Ka band RF Starlink is getting something like 20 Gbps to each of their groundstation gateways? That's not exactly chump-change.

I don't think anyone has closed the business case for either just yet.

That's very likely true. Still as Bill Gates once said it's easy to overestimate what might happen in one year and at the same time underestimate what might happen in ten years.

[Coastal Ron]

I don't think anyone has closed the business case for either just yet.

That's very likely true. Still as Bill Gates once said it's easy to overestimate what might happen in one year and at the same time underestimate what might happen in ten years.

There are five phases to a bubble:

1. Displacement, where investors get enamored by a new paradigm.

2. Boom, where prices start gaining momentum as more participants enter the market.

3. Euphoria, where caution is thrown to the wind, as asset prices skyrocket.

4. Profit-Taking, where the "smart money" sees a bubble coming, and starts selling and taking profits.

5. Panic, where a bubble is recognized, and people start fleeing the market.

I think we are in #3, euphoria, and while I expect this 3rd phase of AI to stick around better than phase #1 or #2, I think we are definitely in a bubble. Too many people are doing things for reasons that they don't understand, but they think is OK.

So from that standpoint, I don't trust the predictions of power growth.

In addition, I would imagine most of the power projections are coming from Large Language Model (LLM) projects, but LLMs are truly horrible for specific category solutions. For that reason I think we'll see a migration to Small Language Model (SLM) solutions that can be run on local machines (i.e. on the edge), with "local" being smaller data centers that already exist.

Lastly, engineers are trying to solve the power problem at the source, on the computing hardware, and I think solutions will be found that reduce the projected power needs.

Taken together, I'm not thinking that a new category of data centers is needed, and certainly not in space.

My $0.02

[Twark_Main]

See also: Scott Manley's video on precisely this question.

Lots of media discussion these days about the dramatic growth in data center construction and the impacts, particularly the demand on the power grid and land consumption

...

 mitigating terrestrial impacts of data centers by moving data processing and storage [to orbit]

How much do the data centers pay for the offending land, power, and also water usage AKA cooling?

How much would they pay in orbit?

Subtracting those two, what are the potential cost savings for the business of moving to orbit? Is it positive or negative?



I expect that 1) it's negative, and 2) companies usually externalize these costs onto the rest of society. If were made to pay a fair price then we wouldn't see perverse outcomes like 75% of new AI data centers being built in water-scarce regions (regions which also tend to be desperate to subsidize any type of local industry).

The real solution is to stop incentivizing companies to build data centers in bad (terrestrial) locations where they cause problems, and instead to build them in better (terrestrial) locations.

[JulesVerneATV]

In a far future it could be put on the Moons, Asteroids, planets of the Solar system -173 °C which is 100 K or -279.4 F. If you have humans and robots running a self sufficient colony and access to 'cold' means that no extra energy or chemistry would need to be expended to cool the data center., the electrical components will perform more efficiently, of course it could be better if MTO had launched but NASA has experimented on laser communication and the Moon is already challenging with 1.4-second latency. Data and info could be stored at transmitted at Lagrange points of the solar system, gravitational stable positions, its hot near the Sun but Mercury’s north pole offers one the coldest places in the solar system, the moon’sof planets which have lava tubes could be useful. Maybe the worlds of the solar system could be a back up a type of Noah's Ark or seed bank as well as keeping historical art blueprints and information recovery and storage in case any region on Earth is hit by a volcano or tsunami or war or whatever.

[laszlo]

If cooling is really the big issue needing to be solved, putting the data centers at the end of a cable in the sea would be more cost-effective than orbit. We have a very robust underwater construction and maintenance infrastructure already developed for oil extraction and cable laying down where the water is permanently 4 degrees C. Off-shore wind power would be right above. And if it doesn't turn out to be economically viable, it'd be just another sunk cost.

[Ike17055]

Not a new idea. I knew a guy about 10 years ago that had started a company to do this, but of course launch costs, technology, etc. has changed since then.

But I really don't see the problem that this solves, unless the need for the data and processing is in space...

Areas in line for data center development, including mine, are struggling with land (especially agricultural land) being gobbled up, strains on the grid to power these centers, worries over the reactivation of old nuclear plants such as TMI to power data centers and more.  Our land is already in demand for housing and prices in the Northeast are skyrocketing, pricing many folks out of market and forcing businesses to operate under high cost structure for land-intensive operations, or even just lack of workforce housing.  These are the issues that could be alleviated -- as the old saying is "invest in land; they aren't making more of it."  Demand vs supply with real world impacts.

[Coastal Ron]

If cooling is really the big issue needing to be solved, putting the data centers at the end of a cable in the sea would be more cost-effective than orbit.

Yep, Microsoft has already tried that, and it worked. They didn't pursue it, but changing business conditions may require a review of the concept..

Not a new idea. I knew a guy about 10 years ago that had started a company to do this, but of course launch costs, technology, etc. has changed since then.

But I really don't see the problem that this solves, unless the need for the data and processing is in space...

Areas in line for data center development, including mine, are struggling with land (especially agricultural land) being gobbled up, strains on the grid to power these centers, worries over the reactivation of old nuclear plants such as TMI to power data centers and more.  Our land is already in demand for housing and prices in the Northeast are skyrocketing, pricing many folks out of market and forcing businesses to operate under high cost structure for land-intensive operations, or even just lack of workforce housing.  These are the issues that could be alleviated -- as the old saying is "invest in land; they aren't making more of it."  Demand vs supply with real world impacts.

You probably have a lot of lakes though, right? Deep enough that a data center can be submerged, and recreational use can continue? That significantly reduces the cooling costs of a data center, though you may still need a boost in local power for the compute part of the data center.

And while I don't know if it is ultimately viable, it does show (per laszlo's post above) that we haven't run out of different ways to solve the core issues before committing to moving data centers into space.

[Ike17055]

Not a new idea. I knew a guy about 10 years ago that had started a company to do this, but of course launch costs, technology, etc. has changed since then.

But I really don't see the problem that this solves, unless the need for the data and processing is in space...

Areas in line for data center development, including mine, are struggling with land (especially agricultural land) being gobbled up, strains on the grid to power these centers, worries over the reactivation of old nuclear plants such as TMI to power data centers and more.  Our land is already in demand for housing and prices in the Northeast are skyrocketing, pricing many folks out of market and forcing businesses to operate under high cost structure for land-intensive operations, or even just lack of workforce housing.  These are the issues that could be alleviated -- as the old saying is "invest in land; they aren't making more of it."  Demand vs supply with real world impacts.

The states and counties in the Northeast has invested hundreds of millions of dollars in purchase of agricultural easements (purchase of development rights) from local farmers as a means of preventing ag land being converted to other uses. This is intended to preserve viability of this key industry, sustain food production on the most fertile areas near the bulk of the nation's population and provide for some open space needs and affiliated benefits, social and environmental.  Now, hyperscalers are coming into the area with enormous offers, undercutting these efforts and presenting challenges to communities and their decisionmakers regarding local impacts.  I attended a recent presentation by the area power supplier and their characterization of the shortfalls in power capacity were sobering. Solutions were few due to a variety of constraints. There will be a public backlash as these concerns take root, reasonable or not.

[Crispy]

If cooling is really the big issue needing to be solved, putting the data centers at the end of a cable in the sea would be more cost-effective than orbit.... And if it doesn't turn out to be economically viable, it'd be just another sunk cost.

Very good

[Ike17055]

If cooling is really the big issue needing to be solved, putting the data centers at the end of a cable in the sea would be more cost-effective than orbit.

Yep, Microsoft has already tried that, and it worked. They didn't pursue it, but changing business conditions may require a review of the concept..

Not a new idea. I knew a guy about 10 years ago that had started a company to do this, but of course launch costs, technology, etc. has changed since then.

But I really don't see the problem that this solves, unless the need for the data and processing is in space...

You probably have a lot of lakes though, right? Deep enough that a data center can be submerged, and recreational use can continue? That significantly reduces the cooling costs of a data center, though you may still need a boost in local power for the compute part of the data center.

And while I don't know if it is ultimately viable, it does show (per laszlo's post above) that we haven't run out of different ways to solve the core issues before committing to moving data centers into space.

No. no real lakes to speak of.  Data centers seem destined to replicate our area's experience with warehouses, which located here due to convergence of highways and easy access to major population centers.  There is a major pushback against warehouses now by preservation advocates and residents; this seems bound to spread to the data center craze gripping the land mania already underway.  But farmers are going to take the best deal they get offered.  Even once the craze hits shakeout phase, we likely will see the same issue we have with warehouses...an abundance of overbuilt facilities and now-obsolete smaller facilities that are empty shells, similar to having a whole bunch of bankrupt shopping malls.

[Robotbeat]

Not a new idea. I knew a guy about 10 years ago that had started a company to do this, but of course launch costs, technology, etc. has changed since then.

But I really don't see the problem that this solves, unless the need for the data and processing is in space...

Areas in line for data center development, including mine, are struggling with land (especially agricultural land) being gobbled up, strains on the grid to power these centers, worries over the reactivation of old nuclear plants such as TMI to power data centers and more.  Our land is already in demand for housing and prices in the Northeast are skyrocketing, pricing many folks out of market and forcing businesses to operate under high cost structure for land-intensive operations, or even just lack of workforce housing.  These are the issues that could be alleviated -- as the old saying is "invest in land; they aren't making more of it."  Demand vs supply with real world impacts.

The states and counties in the Northeast has invested hundreds of millions of dollars in purchase of agricultural easements (purchase of development rights) from local farmers as a means of preventing ag land being converted to other uses. This is intended to preserve viability of this key industry, sustain food production on the most fertile areas near the bulk of the nation's population and provide for some open space needs and affiliated benefits, social and environmental.  Now, hyperscalers are coming into the area with enormous offers, undercutting these efforts and presenting challenges to communities and their decisionmakers regarding local impacts.  I attended a recent presentation by the area power supplier and their characterization of the shortfalls in power capacity were sobering. Solutions were few due to a variety of constraints. There will be a public backlash as these concerns take root, reasonable or not.

Wake me up when solar uses even a third the land that is wasted for growing ethanol corn.

These aren’t reasonable concerns. It’s superstition.

[sdsds]

I'm hoping we start to see more creative ways to integrate photovoltaics into terrestrial landscapes. As an example it seems like right now there's a focus on getting solar panels that extract the maximum amount of energy out of the light that hits them. In the extreme that leaves total darkness below. Imagine instead overhead panels that block only half the light. If those corn fields in e.g. Michigan could instead be converted to shade-grown coffee I'd get really excited! ;-)

But that's looking only at the light that's already hitting Earth. The fundamental claim of all Dyson-esque futurism is that we're silly to be ignoring all the solar power that's just streaming out of the Sun into empty space. And yes, "moving the cloud to orbit" is essentially a Dysonist concept.

[Ike17055]

Not a new idea. I knew a guy about 10 years ago that had started a company to do this, but of course launch costs, technology, etc. has changed since then.

But I really don't see the problem that this solves, unless the need for the data and processing is in space...

Areas in line for data center development, including mine, are struggling with land (especially agricultural land) being gobbled up, strains on the grid to power these centers, worries over the reactivation of old nuclear plants such as TMI to power data centers and more.  Our land is already in demand for housing and prices in the Northeast are skyrocketing, pricing many folks out of market and forcing businesses to operate under high cost structure for land-intensive operations, or even just lack of workforce housing.  These are the issues that could be alleviated -- as the old saying is "invest in land; they aren't making more of it."  Demand vs supply with real world impacts.

The states and counties in the Northeast has invested hundreds of millions of dollars in purchase of agricultural easements (purchase of development rights) from local farmers as a means of preventing ag land being converted to other uses. This is intended to preserve viability of this key industry, sustain food production on the most fertile areas near the bulk of the nation's population and provide for some open space needs and affiliated benefits, social and environmental.  Now, hyperscalers are coming into the area with enormous offers, undercutting these efforts and presenting challenges to communities and their decisionmakers regarding local impacts.  I attended a recent presentation by the area power supplier and their characterization of the shortfalls in power capacity were sobering. Solutions were few due to a variety of constraints. There will be a public backlash as these concerns take root, reasonable or not.

Wake me up when solar uses even a third the land that is wasted for growing ethanol corn.

These aren’t reasonable concerns. It’s superstition.

This isn’t the issue being discussed. Scarce Land needed for homes, farms and business expansion is being squeezed by data center  builders. Also, Not enough power for homes and data centers…these issues need creative solutions or we will see this industry grind to a halt with local opposition and NIMBY obstruction AND POLITICAL INTERVENTION.