Jeff Bezos' rocket company Blue Origin applies to launch 51,000 datacenter satellites

Jeff Bezos’ space company Blue Origin has applied to launch up to 51,600 datacenter satellites. A Thursday filing argues that the US Federal Communications Commission should approve Blue Origin’s plans because “insatiable demand for AI workloads” means orbiting servers represent “a complement to terrestrial infrastructure by …

[Lusty]

Got to admire his optimism. Maybe get Leo (aka Kuiper) working first Jeff, then maybe start a second project.

[TheMaskedMan]

"Maybe get Leo (aka Kuiper) working first Jeff"

Not to mention a couple of lunar landers and a space station - guy definitely needs his Weetabix. But then, this likely isn't about building the things so much as his eternal willy waving contest with SpaceX / Elon Boys will be boys!

[Oneman2Many]

There is New Glenn 9x4 as well which he will need for lunar lander.

[Charlie Clark]

He's a financier at heart and is going to throw announcements where he thinks investors are looking.

[MachDiamond]

"He's a financier at heart and is going to throw announcements where he thinks investors are looking."

Not likely. He's funding BO all by himself to avoid investor difficulties.

[DS999]

That's not where he's looking to generate investor interest.

Who is going to be selling the services these "datacenter satellites" offer? AWS, that's who. This is all about increasing investor interest in AMZN shares.

[Filippo]

Cooling?

See title. Would write more, but sat would overheat.

[TheMaskedMan]

Re: Cooling?

Scott Manley did an interesting video on this very subject the other day. I can't be bothered hunting down a link, but his YouTube channel isn't hard to find.

[Chloe Cresswell]

Re: Cooling?

https://www.youtube.com/watch?v=FlQYU3m1e80 for those who want it

("It will be taking that in from the solar panels, and converting it into heat... and cat memes" great line :)

[frankvw]

Re: Cooling?

Skylab had a heat radiator fin of about 50m² which could dump about 4.2kW of heat into space. Granted, technology has progressed since then, but physics in the real world are much the same.

So with that as a starting benchmark, orbital data centers are a pipe dream.

[Chloe Cresswell]

Re: Cooling?

Skylab, like the ISS, has (well, had) to have a heat radiating system to maintain a human comfort level temp.

Computers can run hotter and the heat radiating system can therefore be hotter and work better.

Scott Manley's video went into it.

Even in Skylab's day, it could have dumped more heat if it wasn't for those pesky humans!

Not saying it's a pipedream or not, just that it's complicated.

And I'm not a spacecraft designer!

[Anonymous Coward]

Re: Cooling?

> Computers can run hotter and the heat radiating system can therefore be hotter and work better.

But the chips themselves work slower and less efficiently at higher temperatures. Just because they can survive protracted temperatures which would be lethal to humans doesn't make that a good use of launch-to-orbit resources. The cost-optimal cooling solution is water, preferably from an already-chilly source. At higher chip efficiencies, more work gets done per unit electrical input.

Political dysfunction has made it very difficult to make any meaningful grid upgrades or expansions in a reasonable amount of time. If the engineers were in charge, the design challenge would be bringing energy from diverse sources to where abundant cool water is, which in most places, will be a coast. In the US Midwest, where DCs have been popular lately, that would be the Great Lakes.

[Chloe Cresswell]

Re: Cooling?

"But the chips themselves work slower and less efficiently at higher temperatures. Just because they can survive protracted temperatures which would be lethal to humans doesn't make that a good use of launch-to-orbit resources."

60C is pretty lethal to humans, so I guess you make sure nothing you use lets it's chips get that hot then?

[Anonymous Coward]

Re: Cooling?

I keep my equipment as cool as reasonably possible, yes. I get more performance out of it and that extends its life.

For enterprise, the driving factor is cost efficiency. Orbital datacenters optimize for solar photovoltaic efficiency, where it's abundant, despite massive costs-to-orbit and reduced cooling efficiency. That reduced cooling efficiency comes at the expense of also requiring more chips running less efficiently.

Next-gen Earthbound data centers can optimize for cooling efficiency where the cooling is abundant. That gets more out of the chips, which are power hungry and currently in short supply.

Moving datacenters to LEO or L2 just for the solar is a hideously expensive fantasy. These builders would be better off figuring out how to get abundant power from the US Great Plains to the abundant cooling around the US Great Lakes. The biggest problem there will be getting skilled tech workers to move to that part of the country.

[Flocke Kroes]

Re: Cooling?

The key take aways from Scott's video were:

• The original artist's concept with a Giga Watt scale solar array was not possible.
• A swarm of Starlink V3 sized satellites could be powered and cooled.
• A swarm would be more spaced out than an ordinary data centre increasing latency and restricting the problems it could work on.

The increased distance between 'racks' also increases the power required for them to talk to each other, eating into the power available for GPUs.

Satellites in LEO must have a de-orbit plan. Starlinks have two: use the last of the propellant to de-orbit in days or if propulsion fails they are in such low orbits that they will de-orbit by themselves in months. AI satellites are higher up and without propulsion would be around for centuries - slowly fragmenting and creating a ring of debris that trashes other LEO satellites. To bring latency close to the realm of sanity the swarms must be densely packed. One failure in the swarm is an instant danger to the rest of the swarm. Right now each Starlink satellite average 2½ collision avoidance maneuvers per month. This proposal requires whole swarms to dodge while staying in formation.

Orbital data centres make sense for SpaceX. They have an IPO expected within months so some bat shit crazy scheme is required to bump up the stock price. Blue Origin is a private company with no plans for an IPO any time soon. They should be focused on profitable projects instead.

[Oneman2Many]

Re: Cooling?

Bottom line I think was that a Starlink V3 sized satellite was about the same volume as a 42U rack and could handle around 20Kw of cooling.

Elon wanted to get into the chip business so I am guessing he is looking at some custom accelerators closer to tensor units which have lower power consumption for certain workloads then your NVidia monsters.

[STOP_FORTH]

Re: Cooling?

Fill space with air and add a cooling fan, problem solved!

Caution: if you fill the solar system with air at one (Earth) atmospheric pressure out to the Oort Cloud you will create a black hole. This is not advised.

[LybsterRoy]

Re: Cooling?

Would the black hole be needed to make sure the air pressure stayed up? I'm also wondering how you stop the air escaping past the Oort Cloud since we have yet to invent the force field.

[David Hicklin]

Re: Cooling?

> Fill space with air and add a cooling fan, problem solved!

You now have a box with 200 degrees C air blowing around - you still have the problem of dumping it out of the box....

What needs inventing is the Thermal Regeneration drive that produces electricity but gets very cold (heading for absolute zero) so *any* heat source could be used to run it.

[MachDiamond]

Re: Cooling?

"Elon wanted to get into the chip business so I am guessing he is looking at some custom accelerators closer to tensor units which have lower power consumption for certain workloads then your NVidia monsters."

He'd need to find the talent to design such things and hire them away from other chip companies which might get those people put on a "list".

There's a big reason companies are buying lorry loads of those Nvidia monsters. While one can create custom chips for narrow applications, that means they likely won't work as well for other things and can go out of usefulness fairly quick.

[David Hicklin]

Re: Cooling?

""Elon wanted to get into the chip business""

About time, it is really hard to buy chipboard these days....

[LybsterRoy]

Re: Cooling?

-- slowly fragmenting and creating a ring of debris t --

Wouldn't this help with climate change by blocking sunlight?

[MachDiamond]

Re: Cooling?

Scott's analysis was close to a best case scenario. That said, he does do a good job of explaining.

There's still the issue of serviceability and scaling even IF the cooling issues are sorted. They'd have to be disposable satellites as there's no way at the moment to send a crew up to update any hardware. Dream Chaser?

[Ken G]

Re: Cooling?

Vacuum tubes?

They need to get up to over 1000K to work and they need vacuum...

[You aint sin me, roit]

Vacuum tubes...

I like it! Save weight on the tubes by not bothering with the glass - it's all a vacuum in any case!

[Baucent]

Re: Cooling?

"Vacuum tubes? They need to get up to over 1000K to work and they need vacuum..."

And you wouldn't need the glass tubes either. I suppose you might even use focussed sunlight to heat the cathode. :)

I imagine a typical AI capable GPU would need to be planet sized if implemented using thermionic valves.

(Old enough to have messed around with that tech when a kid.)

[Ken G]

Re: Cooling?

I'm trying to remember a the name of a technology I read about in a late 60's Analog editorial by John W. Campbell which was basically solid state vacuum tubes (microtubes in a silicon?? fabric maybe) which relied on massive temperatures to work but above a certain size could be self sustaining.

[LybsterRoy]

Re: Cooling?

The other big plus is that if things stopped working a good thump on the side (or top) would bring the picture back :)

[midgepad]

warm to the touch, but

Our vacuum tubes, thermionic valves, were warm to the touch, ranging up to hot, but I don't think low power ones had anything inside quite that hot.

[[e-mail]]

Re: warm to the touch, but

The cathode needs a high temperature (and some magic coating like Ba or Th) to emit a lot of electrons. The design avoids leaking heat from the cathode to the environment, as that would be a waste of energy. Furthermore, the cathode is the innermost one. The outermost one, the anode could get red hot when too many electrons slammed into it at high speed.

[Bitsminer]

Re: warm to the touch, but

My favorite vacuum tube:

Part number 4CX250000A

One quarter megawatt plate dissipation.

Istr the Triumf cyclotron needs a bunch of them

[frankvw]

"...a long way over the horizon..."

Over the Event Horizon, more likely.

Blue Origin (or BO for short, which is not awkward at all) is desperately struggling to remain relevant, and this application is more of a PR exercise than the first step toward anything serious that will actually materialize.

[Charlie Clark]

Re: "...a long way over the horizon..."

I hardly think Blue Origin is struggling to stay relevant. It now has proved launch and landing capability.

Doesn't mean this isn't a bit of PR or, more likely, a pitch to potential investors.

[NetMage]

Re: "...a long way over the horizon..."

Demonstrated more than proved - two launches and one landing does not prove anything.

[Persona]

Re: "...a long way over the horizon..."

Technically it's "Blue" for short.

[Ken G]

Nuclear data centres?

New Small Modular Reactor agreements are part of Amazon’s plan to transition to ...world domination?

[Pete 2]

Six impossible things before breakfast

> ‘Project Sunrise’ needs a network that doesn’t exist, a rocket that’s hardly flown, and FCC approval

To build a system that is too inaccessible to maintain, for a customer that cannot be found and at a cost no-one can imagine

[Jellied Eel]

Re: Six impossible things before breakfast

To build a system that is too inaccessible to maintain, for a customer that cannot be found and at a cost no-one can imagine.

Think positive! Meteorite hunters might get a new revenue stream salvaging slightly crispy Nvidia and RAM!

(also +1 for Scott Manley's video about power & cooling)

[Baucent]

Re: Six impossible things before breakfast

"To build a system that is too inaccessible to maintain, for a customer that cannot be found and at a cost no-one can imagine"

Pretty much defines the current AI mania I would have thought.

[Ken G]

Re: Six impossible things before breakfast

I don't know how much the first two cost, but I'm sure he can afford FCC approval.

[Naich]

The unspeakable in pursuit of the unworkable.

[Anonymous Coward]

Costs

None of these grand schemes for tens of thousands of satellites mention the environmental cost of getting all those satellites into orbit. Nor do they mention what happens when a satellite reaches end of life. A recent report in the New Scientist showed that a lot of unpleasant pollutants get released. This is assuming that there's a reasonably complete burn up. With the proposed numbers, I can't believe that some won't hit land: manouvering out of their assigned position would be fraught with danger

I can't help feeling that this won't end well

[retiredFool]

Re: Costs

I was glad to see someone else point this out. I've mentioned before. Leon has said recently he expects a spacex launch cadence of every 3 hours. We have gone from a launch event to a hohum I need to take a pee cadence. Just like a few AC freon leaks were not a big deal or a few cars were a blip on the atmosphere, the current day launch plans are going to have an impact just like millions of cars and millions of AC leaks did.

[MachDiamond]

Re: Costs

"Leon has said recently he expects a spacex launch cadence of every 3 hours. "

I wonder where he's going to get all of the Methane and LOx. Beyond that would the the trucks that can haul them in large enough numbers.

[renniks]

Is this not the same nonsense idea Space Karen had?

[williamyf]

All these tycoons wanting to put (AI) datacenters into space is a strong indicator that:

Moore's law, while not dead, has slowed to a crawl.

There are no more big arquitectural gains to be had on the hardware side.

Otherwise, it makes no sense to put an (AI) DC up there, just to de-orbit/burn it after 4 years

So we can expect that cutting end hardware progress will stall in the comming years

[ComputerSays_noAbsolutelyNo]

Maybe they think, if we can get the computers to run any faster, make the computers move faster.

[that one in the corner]

Space-based datacenters ... fundamentally lower the marginal cost of compute capacity

> compared to terrestrial alternatives

The eternal problem with Excel, forgetting to include in the sum all those columns that the terrestrial alternatives don't have. Little things, like launching...

Except for things which require resources that *only* exist in space, when has a space-based thing *ever* been cheaper than terrestrial? We're even looking at terrestrial replacements for "it has to be in space" things, like optical telescopes (it being questionable whether there is a purely science-based reason* for replacing/retaining the Hubble Space Telescope, as we have the capability now to build earth-bound instruments which would be huuuge and costly to build initially than the HST was, but way cheaper to get into service - no launching - maintain and upgrade).

* as opposed to social out-reach or even political reasons

[Korev]

Re: Space-based datacenters ... fundamentally lower the marginal cost of compute capacity

When I saw HST, I thought you meant the classic British train, that's notoriously nicer to travel in than its replacement...

[NetMage]

Re: Space-based datacenters ... fundamentally lower the marginal cost of compute capacity

The point isn’t to be cheaper it’s to try and avoid NIMBYism while being feasible. How much would an AI company pay to shave years off of their construction timelines?

(Not to say the economic model closes even so.)

[cyberdemon]

Feasible???

> The point isn’t to be cheaper it’s to try and avoid NIMBYism while being feasible.

You owe me a new keyboard.

What part of "relocating gigawatts of compute capacity to sun-synchronous orbit" is remotely, never mind objectively "feasible"?

It fights fundamental physics on at least three fronts:

1. The mass of equipment required to go to orbit. How much does the equipment inside a datacentre weigh? Including coolant, 10kA copper busbars, etc. How much does it weigh when it needs a passive black-body radiator to cool it? What is the (terran) energy/pollution cost of hauling all that shit up to orbit?

2. Cooling. As alluded to above, there is no air cooling and no evaporative water cooling in spaace. Chips burn up not much past 125C/400K. That doesn't radiate much. Sure, the backsides of space solar panels don't overheat, but that's because they have a huge surface area for free. How do you transport the heat from a tiny 1kW chip out to a 10 sq.m surface, while minimising weight?

3. Power. A Gigawatt is a fucking huge number. The sun's power in space is only 30% more than it is on Earth, so even in spaace, you'd need a huge area of panels to provide that sort of power

Then there's ionising radiation, comms, cost, pollution, geopolitics (defensibility against missile attack), and other "minor" issues.

Feasible, it is not. And "NIMBYism" is the least of their worries for building on Earth. This is desperation from the tech bros; distraction from an obviously faltering bubble.

[NetMage]

Re: Feasible???

Strawman much? No one is suggesting (except the naysayers) that an orbital data center would be a single satellite- in fact, they are suggesting large numbers of satellites that work together.

[Antony Shepherd]

Hey Siri...

Hey Siri, how do I say I want to induce Kessler Syndrome without actually saying "I want to induce Kessler Syndrome"?

Because with Bezos and Musk engaged in a "Who can shit out the most data centres into space" willy waving contest, that's how we'll get Kessler Syndrome.

[NetMage]

Re: Hey Siri...

Stop trying to make Kessler syndrome a thing. It was never a thing, it will never be a thing.

[midgepad]

put them in the Moon

Lots of mass for cooling.

And shielding.

Lots of Silicon there, if one builds a fab.

Vacuum available.

Latency between clusters whatever you build.

To Earth, well, a bit more.

No re-entry burnups.

No Kessler Syndrome.

Very large solar arrays, mirror fields etc stable.

[cyberdemon]

Re: put them in the Moon

Put them in Hell

Shove them up your arse

We don't need AI

It's all a fucking farce

[MachDiamond]

Re: put them in the Moon

"Lots of Silicon there, if one builds a fab."

It takes a lot more than Silicon.

It's like saying there's Oxygen on Mars, so humans could live there. Humans don't breathe Oxygen, we breathe Nitrogen with an Oxygen chaser and there doesn't seem to be much N in Mars' inventory.

[steviebuk]

Here's....

"500million for orange tango man. Can I get permission now FCC".

Yes, of course.

[OldSoCalCoder]

Re: Here's....

It does explain why I saw Amazon Prime pimping the Melania movie for a while.

[steviebuk]

Re: Here's....

We have 2 MAGA lovers so far.

[Uh, Mike]

Moving the Date Closer

Eventually, a hundred thousand satellites in low earth orbit will trigger the Kessler Syntrome.

Don't worry, be happy.

[ecofeco]

AI workload demands?

And who's fault is that?

Sounds like a personal problem.

[Anonymous Coward]

Re: AI workload demands?

> And who's fault is that?

The obsolete and expendable carbon units.

[STOP_FORTH]

Broken dreams

I was half thinking of taking up astrophotography as a hobby. Seems a bit pointless now. I can just draw lots of lines on a piece of paper, take a picture and invert it.

Rats!

[M.V. Lipvig]

Call me a pessimist

but the actual goal is a place to hide siphoned data from beyond the reach of terrestrial laws. Once it goes up there, it stays there until someone devises a way to force them down without turning them into sprays of shrapnel.

[MachDiamond]

Re: Call me a pessimist

"Once it goes up there, it stays there until someone devises a way to force them down without turning them into sprays of shrapnel."

It's easier than that. Just disable those satellites so nobody can access the data and leave the shell where it is subject to the whims of gravity, solar wind, etc.

[Anonymous Coward]

Datalink

The bandwidth needed to run to/from orbital data centres is what's worrying me. And if each satellite is basically only a server rack or 10, then how do you connect hundreds or thousands with low latency?

[Oneman2Many]

Re: Datalink

Laser links

[JamesTGrant]

Simple - it’s a scam for investors.

[trevorde]

Statement from Elon Musk, CEO SpaceX

We will be launching 120,000 datacenter satellites to process data and train our models for Tesla Full Self Driving, which will be coming in 2027

[Sanguma]

renaming suggestions

To avoid a namespace clash with Qantas, which got there first, I suggest, Project Sunset, or even more appropriately, Project Moonshine.

[Winkypop]

Re: renaming suggestions

As it’s QANTAS, just rename if “Affordable Sunrise”.

That’s something QANTAS will never be interested in.

[Anonymous Coward]

Cooling

Easy.

Only turn them on at night and leave a window ajar.

[Anonymous Coward]

What’s with the communist solar-powered satellites?

Why aren’t they powered by Trump-approved Clean Coal (TM)?

[Ken G]

Re: What’s with the communist solar-powered satellites?

At least they don't have windmills. They're so pathetic and so bad. Only stupid people put windmills in space.