The economics of orbital AI are a complex and controversial topic, with many factors to consider. In this article, we'll explore the challenges and opportunities presented by the idea of building solar-powered orbital data centers, distributed across a million satellites. We'll delve into the costs of designing and launching satellites, the production costs, and the potential for AI satellites to operate in space. We'll also discuss the challenges of thermal management, cosmic radiation, and solar panel efficiency. Finally, we'll examine the potential for orbital data centers to be used for inference tasks, rather than training new models. But here's where it gets controversial: will the economics of orbital AI ever make sense, and what will it take to get data centers into space?
The Economics of Orbital AI
Elon Musk and his company SpaceX have been talking about AI in space for years, and now they see an opportunity to realize a version of this vision. SpaceX has requested regulatory permission to build solar-powered orbital data centers, distributed across as many as a million satellites, that could shift as much as 100 GW of compute power off the planet. Musk has suggested that some of his AI satellites will be built on the Moon.
"By far the cheapest place to put AI will be space in 36 months or less," Musk said on a podcast. But is this really the case? In a first analysis, today's terrestrial data centers remain cheaper than those in orbit. A space engineer has built a helpful calculator comparing the two models, showing that a 1 GW orbital data center might cost $42.4B—almost three times its ground-bound equivalent.
To change this equation, experts say that technology development across several fields, massive capital expenditure, and a lot of work on the supply chain for space-grade components will be required. It also depends on costs on the ground rising as resources and supply chains are strained by growing demand.
Designing and Launching Satellites
The key driver for any space business model is how much it costs to get anything up there. SpaceX is already pushing down on the cost of getting to orbit, but analysts need even lower prices to close their business case. In other words, while AI data centers may seem to be a story about a new business line ahead of the SpaceX IPO, the plan depends on completing the company's longest-running unfinished project—Starship.
The reusable Falcon 9 delivers a cost to orbit of roughly $3,600/kg. Making space data centers doable will require prices closer to $200/kg, an 18-fold improvement expected in the 2030s. However, even if Starship is completely successful, assumptions that it will immediately deliver lower prices to customers may not pass the smell test.
Production Costs
Even if launch is the bane of all space businesses, the second challenge is production cost. "We always take for granted that Starship's cost is going to be hundreds of dollars per kilo," said a space engineer. "People are not taking into account that satellites are almost $1,000 a kilo right now."
Satellite manufacturing costs are the largest chunk of that price tag, but if high-powered satellites can be made at about half the cost of current Starlink satellites, the numbers start to make sense. SpaceX has made great advances in satellite economics while building Starlink, and the company hopes to achieve more through scale.
Thermal Management and Cosmic Radiation
The space environment is not fooling around. Orbital data center proponents often say that thermal management is 'free' in space, but that's an oversimplification. Without an atmosphere, it's actually more difficult to disperse heat. AI satellites will also need to deal with cosmic radiation, which can degrade chips over time and cause 'bit flip' errors that can corrupt data.
Solar Panel Efficiency
Another challenge comes from the solar panels themselves. The logic of the project is energy arbitrage: putting solar panels in space makes them anywhere from five to eight times more efficient than on Earth, and if they're in the right orbit, they can be in sight of the sun for 90% of the day or more, increasing their efficiency. But even solar panels are more complicated in space.
Space-rated solar panels made of rare earth elements are hardy but too expensive. Solar panels made from silicon are cheap and increasingly prevalent in space, but they degrade much faster due to space radiation. That will limit the lifetime of AI satellites to around five years, which means they will have to generate return on investment faster.
Inference Tasks in Space
One outstanding question about these data centers: what will we do with them? Are they general purpose, or for inference, or for training? Based on existing use cases, they may not be entirely interchangeable with data centers on the ground. Inference tasks don't have the same need for thousands of GPUs working in unison, making them a more viable starting point for orbital data center business.
"Training is not the ideal thing to do in space," said the CEO of Starcloud. "I think almost all inference workloads will be done in space," imagining everything from customer service voice agents to ChatGPT queries being computed in orbit. He says his company's first AI satellite is already earning revenue performing inference in orbit.
The Future of Orbital AI
For SpaceX, the company's recent acquisition of xAI will let them stake out positions in both terrestrial and orbital data centers, seeing which supply chain adapts faster. The economics of orbital AI are a complex and controversial topic, and it remains to be seen whether they will ever make sense. But with the right technology development and capital expenditure, the future of orbital AI could be bright.
