Space data centers seem crazy. They make a lot more sense than it seems

“Space, the final frontier” became a classic pop culture phrase thanks to the series Star Trek. Now there are those who complete it with “… data centers”, because that is what Elon Musk certainly wants to achieve, and he has a plan to achieve it. At first glance it seems crazybut it turns out that the idea is not at all crazy.

Free cooling, nothing. As explained in a very deep report in Semianalysismany analysts support the idea by defending erroneous premises. The space, for example, does not offer free cooling. Since there is no atmosphere, heat is not dissipated by convection, and huge and expensive thermal radiators are necessary. Solar energy is also interrupted in low orbits (LEO), so satellites must be placed in sun-synchronous orbits, a resource that is beginning to become saturated.

The current cost does not compensate. The analysis carried out in this study for the Total Cost of Ownership (TCO) for a currently standard 30.5 kW cluster (with two servers with 16 Nvidia B300 GPUs) does not add up. Deploy this infrastructure In space it is necessary to invest 4.1 million dollars, when doing the same on Earth costs 1.4 million dollars. Space data centers are currently 260% more expensive than on the planet’s surface. Bad business.

Space transportation makes everything more expensive. He biggest problem What affects these costs is the costs of transporting the material to space. In that proposed example, of the $3.1 million total cost of space infrastructure, $1.6 million is due to launch. But there is also the problem of the useful life of this data center: on Earth these facilities pay for themselves in 15 years, but in space wear and radiation in orbit reduce the operational life of the particular satellite to only five years, which multiplies those capital expenses dedicated to the project almost by 20.

The first bottleneck is the chips. Even solving these problems, the main obstacle is simply semiconductor manufacturing capacity. The demand for TSMC’s N3 wafers and the supply of HBM memories is much higher than the supply even without this idea of ​​​​space data centers. That would add even more demand to an absolutely saturated system.

But there is also the (lack of) energy. The reason why Musk wants to promote this idea as soon as possible is that obtaining power supply for terrestrial data centers is increasingly complicated. Thus, getting a connection to the electrical grid in Virgnia (USA) already takes seven years. Companies are creating their own power generation plants to solve this problem. Even so, according to the study, it will become increasingly more expensive to access this supply: they estimate that the cost of “terrestrial energy” will be above 20 million dollars per MW when this decade ends.

That’s why Terafab. To solve this first bottleneck, Elon Musk has launched its colossal Terafab project in Austin. It is a huge chip manufacturing factory that will need 10 GW of electrical power to produce one million semiconductor wafers each month. The plan takes into account that 80% of the chips produced are destined precisely for space data centers.

Starship changes the equation. But Starship stands in front of all these problems. SpaceX hopes to be able to reduce launch costs significantly in the coming years, going from the current $1,400-1,800 per kilo for the Falcon 9 to just $250 per kg for the Starship. This, together with the improvement in radiator and solar panel technology, will reduce the cost gap with terrestrial infrastructure. Now it is 260% more expensive, but at the beginning of the next decade it will be only 30% more expensive and will achieve economic parity by 2040.

But. The accounts could therefore come out in the medium term, but it is necessary to take into account other factors as the so-called long-term computing cost. On Earth, between 3% and 6% of GPUs in data centers fail each year and require manual replacement by a technician. In space that option disappears, so it is necessary to oversize the satellites with 20% chips to provide redundancy and thus absorb potential radiation failures.

In Xataka | Aragón is quietly becoming a data center “powerhouse” – now it has taken a crucial step