A US startup, Lumen Orbit, proposes an unexpected solution to the growing energy demands of artificial intelligence development: building data centers in space. This bold idea is laid out in the company’s first white paper, released earlier this week, which promotes the benefits of space-based data centers to support the future of AI.
What’s Happening & Why This Matters
Lumen Orbit suggests that putting data centers in space could address the increasing energy consumption required for training advanced AI models, such as ChatGPT. Traditional data centers on Earth consume massive amounts of electricity, not only to operate but also to cool the densely packed GPUs needed for AI tasks. As these data centers grow larger to keep up with the demand for AI development, the energy required becomes a major concern, both in terms of cost and environmental impact.
Philip Johnston, CEO of Lumen Orbit, argues that training AI models in space offers distinct advantages, including access to abundant solar energy and the ability to scale up operations without being limited by terrestrial constraints. He claims, “We should train future large AI models in space to make use of abundant solar energy, cooling, and the ability to freely scale up.”
Lumen Orbit’s white paper provides a cost analysis that strengthens its case. The company estimates that running a space-based data center for ten years would cost around $8.2 million, including an initial $5 million launch cost. In comparison, operating a similar ground-based data center could cost up to $167 million over the same period. The plan includes using a massive solar array—about 4 kilometers by 4 kilometers—to generate power for these orbital centers, which could reduce the energy costs to just $2 million over a decade, a far cry from the $140 million required for a ground-based equivalent.
The company plans to build these space-based data centers using new spacecraft like SpaceX’s Starship, designed to transport humans to Mars. Lumen Orbit believes that decreasing launch costs will make it economically viable to build and maintain data centers in orbit, promoting this approach as necessary for realizing AI’s full potential sustainably.
However, the project does face challenges, particularly in protecting equipment from the intense solar radiation encountered in space. The company also highlights the need for specialized radiators to dissipate heat effectively, describing this as the most significant technical hurdle to achieving hyperscale space data centers.
TF Summary: What’s Next
Lumen Orbit plans to test its ambitious concept by launching a prototype satellite next May, with the goal of training the first AI model in space. By 2026, the company plans to deploy a “micro data center” in orbit, setting the stage for more extensive space-based AI operations. If successful, this “innovative “out-of-this-world” approach could reimagine data center management, offering a sustainable solution to the escalating energy needs of AI development. These next steps will focus on ironing out technical challenges and demonstrating the economic viability of space-based data centers to stakeholders, investors, and potential partners, and inevitable competitors.