Jeff Bezos told Italian Tech Week in Turin last October that gigawatt-scale data centers will be operational in space within 10 to 20 years. Speaking in conversation with Ferrari and Stellantis Chairman John Elkann, the Amazon founder made a specific argument: the energy and thermal environment of low Earth orbit is so much better than Earth that orbital data centers will eventually be cheaper than terrestrial ones.
His exact words: "We have solar power there, 24/7. There are no clouds and no rain, no weather." And: "We will be able to beat the cost of terrestrial data centers in space in the next couple of decades."
Bezos is not a fringe voice on this. He founded Blue Origin. He owns Amazon, which operates one of the largest data center fleets on Earth. When the person who built the world's largest cloud tells you he thinks space is the better location for future compute, the thermal management industry should take the statement seriously.
"It's hard to know exactly when, it's 10+ years, and I bet it's not more than 20 years — we're going to start building these giant gigawatt data centers in space." Data centers currently consume approximately 5% of US electricity, projected to reach 9–17% by 2030 (EPRI). The energy constraint is the primary driver of the space thesis.
Terrestrial data centers have three resource constraints that compound at gigawatt scale: power, water, and land. Power is the binding constraint in most markets — the grid cannot deliver enough. Water is the binding constraint in arid markets — cooling towers consume tens of millions of gallons annually per facility. Land is a secondary constraint but increasingly subject to community opposition and regulatory friction.
In orbit, all three constraints disappear. Solar power is continuous and abundant. There is no water requirement — heat rejection is passive, through radiation to deep space at 2.7 Kelvin. There is no land constraint. The infinite heat sink of space solves the cooling problem by eliminating it as a design consideration.
What replaces those constraints are different ones: launch costs, maintenance access, hardware replacement logistics, and the upfront capital required to get hardware into orbit. Bezos's argument is that launch cost trajectories — driven by SpaceX and Blue Origin reusability programs — will cross a threshold where the eliminated operational costs in space outweigh the higher deployment costs. His timeline is 10 to 20 years.
Bezos specifically noted that large AI training clusters are a natural fit for space-based infrastructure. The reasoning: training runs are not latency-sensitive in the way that inference serving is. A training job that takes 30 days does not require sub-10-millisecond response times. Data can be transmitted to and from orbit at reasonable latency for batch workloads. Inference — where a user is waiting for a response — stays on Earth, close to the user.
This workload segmentation matters for the cooling industry. AI training is the most thermally intensive workload in commercial computing. A single training cluster for a frontier model can draw 100 megawatts or more continuously for months. That is the workload Bezos is suggesting moves to orbit. If he is right, the terrestrial cooling industry retains inference — which is growing but thermally less extreme — and loses the peak thermal demand that drives the largest CDU and cold plate orders.
The cooling industry has a 20-year window before Bezos's scenario becomes operational at meaningful scale — if it becomes operational at all. Launch costs need to fall by another order of magnitude. On-orbit assembly and maintenance need to become routine. Power transmission from orbital arrays to the satellites themselves needs to scale. None of these are insurmountable. None of them are solved.
The cooling industry should file Bezos's statement not as a threat but as a directional signal: the people with the capital and the technical capacity to build next-generation compute infrastructure view terrestrial cooling as a transitional constraint, not a permanent feature. That is the context in which every CDU procurement, every cold plate contract, and every facility design decision gets made for the next 20 years.