SpaceX & xAI – Space Debate

It is not as crazy as it sounds.

• The acquisition of Cursor will certainly help xAI push its business forward, but the real question is not whether we can build data centres in space, but whether it is economically viable to do so.
• Mr Musk is well known for interesting ideas, but not all of them make very much commercial sense.
• It is fair to say that he kick-started the idea that electric vehicles could be a viable option, but his statements regarding the financial business model of robotaxis are outlandish.
• The Hyperloop is another example of an idea that sounds great but is simply not grounded in reality.
• This is why, when he talks about putting data centres in space, it is worth taking the time to ask the question of whether or not it makes economic sense.
• The surprising answer is that data centres in space are not nearly as nutty as some of his other ideas.
• The merger of xAI and SpaceX is a clear sign of this intention, and adding Cursor into the mix will help xAI compete against Anthropic for the enterprise AI spend.
• However, whether xAI will end up building the compute it needs to offer these services on the ground or in space remains to be seen.
• The USA has a problem with electricity as its total generative capacity is only 1.3 Terawatts compared to China, which has three times as much, and it is already rapidly building more.
• Hence, when one looks at an energy source as abundant as the sun, using solar power in space has to be considered as an option to solve the power constraint in the Western world.
• Furthermore, solar power in orbit is far more effective than it is on the ground and would have the effect of reducing the operating cost of energy for compute from $1.2bn/GW/ year to almost zero.
• The problem is that, at the moment, the business model of compute is broken, and it is not the cost of the energy that has broken it.
• Instead, it is the realisable revenue per GW combined with the capital cost of building a data centre that has caused the business model to break.
• The industry can currently earn around $10bn/GW for a data centre that costs $40bn to build and costs $1.5bn or so to operate and lasts for 5 years giving a 5-year return on investment of 1%.
• RFM has concluded that because Vera Rubin will cost around 50% more to build, the revenue per GW must rise to at least $20bn/GW to prevent a financial correction.
• The cost of electricity is not the problem here because if I assume that the electricity is free (as it would be in space), then the 5-year return on investment rises to just 5%.
• This means that the business model is still broken and clearly indicates that the cost of electricity is not the main problem.
• Instead, it is the capital cost of the data centre and the relatively low amount of revenue that 1GW of compute capacity is able to earn in the open market.
• This is the problem that has to be solved to prevent a financial correction, and building data centre’s in space is not going to solve it.
• This is because the IT infrastructure is already 77% of the capital cost (rising to 85% with Vera Rubin), and being in space is not going to reduce any of these costs at all.
• Instead, all of this infrastructure will need to be lifted into orbit and then assembled in the vacuum of space.
• However, this idea is not as crazy as it sounds.
• SpaceX has done a marvellous job of reducing the cost of airlifting goods into orbit, but I am not convinced that it is yet close to a point where building the data centre in space will be cheaper than building it on the ground.
• SpaceX currently charges around $1,500 to airlift 1kg of cargo into space for the dedicated Falcon Heavy.
• A data cabinet weighs around 100kg, where around 8,000 Vera Rubin cabinets are needed to provide 1GW of compute.
• At current prices, this would cost $1.2bn to airlift that into orbit before any cost to assemble, test, maintain and run is taken into account.
• Given that the electricity costs $1.2bn per year in the USA and in space, this would be practically free, the cost of the airlift is paid back within 12 months.
• Hence, from an energy perspective, this is a no-brainer.
• However, I think that the other costs to build and assemble such a structure in space will be substantial, and I suspect further progress here is needed to make this a viable proposition.
• Hence, I am not yet convinced that data centres in space are a viable option, but I am willing to concede that this is one of his ideas that actually could make a great deal of financial sense in the long run, especially given the energy constraints the USA and the West are facing.
• For once, Mr Musk has positively surprised me.