AI infrastructure in space is still far, far away.

getty

I recently covered Nvidia CEO Jensen Huang’s remarks on putting AI servers in space. He describes it as a vision for the future.

The notion of AI data centers in space is tantalizing. It suggests a bold, transformative leap—a future echoing the tech industry at its most innovative. Yet, considering the practicalities, it is clear that space-based AI data centers are not imminent—and likely not as near as many people might hope.

To start with, space is an unforgiving place—hostile and indifferent to the fragile electronics that power today’s advanced AI systems. You can’t simply deploy Nvidia GPUs into space and expect them to operate reliably. You'll need radiation-hardened chips – chips designed more for durability than performance. However, as soon as you make a compromise like that, you've already lost – you've already conceded that your AI system in space is going to be inferior to one sitting in a datacenter in Virginia or Arizona.

And then there is the cost.

The economics are simply brutal. Launching payloads into orbit is extremely expensive, despite reduced costs from companies like SpaceX. When you’re looking at the scale required for serious AI workloads, consider the physical size of a system like Nvidia’s GB200 NVL72 rack. Scaling such hardware for space, hardening it for space environment, handling maintenance logistics, addressing satellite lifespan issues and addressing the specialized needs of this equipment all make a business model fundamentally difficult.

Power is another barrier that, when you get to space, you simply cannot get past. AI systems are, by any measure, incredibly power-intensive. Even on Earth, hyperscale data centers are struggling with power consumption, which is straining local power grids. Solar power is the only viable energy source in space, but building systems that can supply the immense power demands of advanced AI adds further complexity, mass and expense to an already ambitious project.

And then there’s latency. One of the key benefits of building AI infrastructure is responsiveness, the capacity to deliver results and get them back to a user for real-time utility. Physics is non-negotiable; while space-based systems have many benefits, latency becomes a significant obstacle when communicating with Earth. The speed of light is a hard limit, and for many of the most valuable AI workloads that are being developed today, this simple fact of physics is another hurdle to conquer.

To be fair, there are some specific cases where the story gets more interesting. One example is on-board satellite processing, where edge AI enables spacecraft to operate independently and make real-time decisions without waiting for ground-based commands - a legitimate and growing field. This is, of course, a far cry from the notion of data centers orbiting Earth being used for enterprise AI workloads for businesses on Earth.

The vision of AI data centers in space is indeed compelling and it is this moonshot-thinking that has driven technology forward throughout history. Compelling visions, however, still need to pass the test of contact with reality. Orbiting AI data centers are not yet ready for the task, it is not cost-effective, its power requirements are staggering and the latency requirements defeat the purpose of the most valuable applications.

The smarter investment strategy is clear: to improve the quality, efficiency, and sustainability of our current AI infrastructure right here on Earth. As our AI models become increasingly capable and efficient at the same time and as our demand for computation continues to grow exponentially, the industry's focus is well-placed on tackling the challenges we face.

Space will indeed play a role in the future of AI infrastructure. But the chapter is still far, far away. For the time being, the data centers driving the AI revolution will remain firmly rooted on the ground.

Disclosure: Nvidia subscribes to the research reports from the company I founded, Creative Strategies, along with many other high-tech companies around the world.