Is this SciFi enough?

We are witnessing another SciFi idea turned into reality. A startup company, Starcloud, launched and placed in orbit the first AI Data Center.

One interesting note, I couldn't find any mention of it in the major new agencies. Reuters, nothing. AP News, nothing. I tried CNN, nothing. The only mention was on NBC and it was an interview between the reporter and the CEO.

I found the interview a little bit inaccurate and also a little bit strange.

Let's start with the inaccurate part first. The reporter was obviously very excited, because how often do you get to talk about something so cool as a computer in space. In his rapid coverage, he explained that putting a data center in space is an awesome idea because there is no limit on space to put it, a never ending source of sun for energy and a vacuum for cooling. Here is the thing, a vacuum on its own doesn't cool anything. Later in the interview the Starcloud CEO clarified this detail by explaining that any heat will be dissipated using a cooling system which depends on thermal radiation. Perhaps a semantic detail but an important one. More on that later.

The interviewer later asked the question why - except the abundance of energy and space - would anyone build a data center in space?

The answer was - we are collecting so much data in orbit that it takes forever to get it to the Earth and it would be far easier to process all the data 'up there' and just download the results. An example was used that we can spot wildfires in California and that you want to get the information much quicker. Perhaps.

Before we continue with the rest of the interview, I think it is important to get some numbers to better understand the economics of this project and create a context.

Let's say that you too, would like to build a data center in space. And not just any data center, it has to be an AI Data Center.

One of the backers of this project is Nvidia and naturally there can't be any other AI chips in space than the Nvidia ones. We can send up the NVIDIA DGX H100 system which contains the H100 GPUs and based on configuration can set you back somewhere between $300,000 and $450,000. Actually the price doesn't matter. What matters is the weight which is 376 lbs (170.45 kg). Another important piece of information is the power consumption, which for the  NVIDIA DGX H100 is 10 kW.

The one area where the Starcloud team saved some weight is placing the satellite on the Sun-Synchronous Orbit (SSO) , where the satellite passes over a given point of the planet surface at the same local time. Which means that the data center will have 24/7 sunlight and therefore no need for batteries. Batteries are heavy.

We also need the solar panels to generate the electricity and cooling panels to keep the servers from heat stroke. To cool 1 kW the cooling system weights somewhere between 3.5 kg to 7 kg.

Is that everything? Of course not. You have to also pack the servers and most importantly cover them with extra protection from radiation. Up there, without the protection of the atmosphere, the electronics are exposed to high energy particles from solar flares and cosmic rays; not to mention micrometeorites. Since Nvidia is selling these chips on Earth, these chips don't have the required properties to function without any protection. Can you buy chips like that? Yes you can. An example is the RAD750 designed by IBM and manufactured by BAE. Except one chip like that is somewhere around $200,000 (per 2002 reference).

Finally, we have to ship the data center up with all the other required parts. Using Falcon 9, the cost on the ride sharing mission is $1.1 million for 200 kg (440lb). That barely covers one NVIDIA DGX H100. When adding the protection and cooling and solar panels we might add extra 50% of weight. (Yes, I am fully aware that these numbers are just estimates. The exact weight and cost really depends on many factors - the type of material where you balance cost and weight, the height of the orbit, e.g. how far the payload has to travel, etc.)

Also for people getting snuggly with trees, one should note that Falcon 9produces 336 - 440 metric tons of CO2. Compare this to 450 - 490 grams CO2 per kWh for a gas-fired power plant.

The lower end of 336 metric tons is the same amount of CO2 which a gas power plant would generate over 4 years if it is going to power and cool one NVIDIA DGX H100 system. The expected lifespan of an NVIDIA DGX H100 system is 3 - 5 years. While if we say that the unit will last 4 years and we have to pay for the electricity on Earth at the rate of U$ 0.1 we would have to pay about $700,800 which is at least $400,000 less than the $1.1 million we have to pay to get it up.

So the cheap sun energy and 'vacuum cooling' doesn't look like the real argument for taking servers to space.

Now it is time to continue with the NBC interview.

“... you've hit the nail on the head with those two points. So the first is almost unlimited low cost energy in the form of solar and then the second is we can dissipate that heat out in infrared into the vacuum of space ... so those are the two big advantages as not having to go through long permitting cycles on Earth.”

And then you go to the Starcloud website and you read that Starcloud offers “.. global data storage and premium sovereign cloud computing ..”

This project is not about cheap power and cooling and locating fires in California. This project is about taking data and moving it outside of any jurisdiction, oversight and control. I am sure that there will be lots of takers for that offer.

The recurrent pattern? This is not SciFi turned reality project. It's just a new place to hide family jewels.