Amazon founder Jeff Bezos forecasts a future where the insatiable energy demands of artificial intelligence are met not by Earth’s strained power grids, but by massive data centers located in orbit. Speaking at Italian Tech Week, he outlined a vision where gigawatt-scale facilities, powered by uninterrupted solar energy, will be constructed in space within the next one to two decades. This orbital infrastructure, he argues, will eventually outperform and become more cost-effective than any comparable terrestrial facility, marking a pivotal shift in how global data infrastructure is powered and managed.
The core of the proposal addresses a burgeoning crisis on the ground: the exponential growth of AI is driving a massive spike in electricity and water consumption at data centers worldwide. As companies train increasingly complex AI models and expand cloud computing services, the environmental and logistical constraints of Earth-based operations are becoming critical bottlenecks. By moving this heavy industry off-planet, Bezos suggests humanity can leverage the constant, powerful energy of the sun, bypassing terrestrial limitations like weather, nighttime, and the need for vast quantities of fresh water for cooling. This initiative follows a historical pattern where essential infrastructure, from weather forecasting to global communications, has moved into orbit to better serve the planet.
An Insatiable Thirst for Energy and Water
The rapid expansion of artificial intelligence is creating an unprecedented demand for computational power, which in turn requires vast amounts of electricity and water. Globally, data centers are on a trajectory to consume an ever-larger share of energy production. Projections indicate that by 2028, the AI sector in the United States alone could consume as much as 300 terawatt-hours of electricity annually, equivalent to the power usage of over 28 million American households. This surge is driven by the energy-intensive chips used to train and run large AI models, with the newest generation of data centers using five times more power than their predecessors from just a decade ago.
This energy consumption has a direct and significant impact on freshwater resources. Data centers require immense quantities of water, primarily for cooling the thousands of servers that generate enormous heat. A large data center can consume up to 5 million gallons of water per day, comparable to the daily needs of a town of up to 50,000 people. By 2027, AI’s global annual water footprint is projected to be as high as 6.6 billion cubic meters. This demand places a heavy strain on local water supplies, particularly as many data centers are located in water-stressed regions. The combined pressure on energy grids and water sources makes the current terrestrial model for data center expansion a growing environmental concern.
The Promise of Orbital Power
The primary advantage of moving data centers to space is access to continuous and abundant solar energy. A satellite in geostationary orbit, approximately 36,000 kilometers above the Earth, is exposed to uninterrupted sunlight, free from the day-night cycle, cloud cover, and atmospheric absorption that limit terrestrial solar farms. This allows space-based solar panels to generate significantly more energy per square meter, providing a consistent and reliable power source ideal for the 24/7 operational demands of AI training clusters. Bezos asserts that this constant power stream will eventually make space-based facilities more economical than their Earth-bound counterparts.
A Concept with History
The idea of space-based solar power (SBSP) is not new, having been first proposed in the 1960s. For decades, the high cost of rocket launches made it economically unviable. However, recent advancements have renewed interest and investment in the technology. Several nations, including the United States, China, Japan, and the United Kingdom, are actively developing SBSP projects. The California Institute of Technology (Caltech), backed by over $100 million in private funding, successfully launched a test array in 2023 to demonstrate wireless power transmission from orbit, validating a key component of the concept.
From Photons to Processing
The proposed system involves large satellites equipped with vast solar arrays. These arrays would collect solar energy, convert it into electricity, and use it to power the co-located data processing hardware. The results of the computations, rather than the raw power, would then be transmitted back to Earth. This transmission would likely use high-bandwidth laser communication systems, which offer data transfer speeds 10 to 100 times faster than current radio frequency methods. This model eliminates the immense challenge of beaming gigawatts of power through the atmosphere, focusing instead on high-speed data transfer.
Building Factories in the Final Frontier
Realizing a vision of gigawatt-scale orbital data centers requires a revolution in in-space servicing, assembly, and manufacturing (ISAM). It is impossible to launch such a massive structure in a single rocket fairing; it must be built in orbit. This involves the robotic assembly of modular components launched separately. This approach mirrors the construction of the International Space Station but would need to be executed on a far grander scale and with much greater autonomy.
Advancements in robotics are critical for these tasks, which are too complex and dangerous for extensive human astronaut involvement. Autonomous systems must be capable of intricate maneuvers, docking modules, and making connections with high precision, all while contending with the unique challenges of the space environment. The falling cost of launch, driven by reusable rockets from companies like SpaceX and Blue Origin, is a key enabler. As launch costs potentially drop to $100 per kilogram, the economic calculus for ambitious in-space construction projects becomes increasingly favorable.
Hurdles Beyond the Horizon
Despite the promise, the path to orbital data centers is fraught with immense technical and logistical challenges. The primary concerns highlighted by Bezos himself are maintenance and upgrades. Servicing a data center in orbit is orders of magnitude more complex than on Earth; a faulty server cannot be simply swapped out by an engineer. Any repair mission would require a costly and complex robotic or crewed launch.
The Latency Question
While laser communications can transmit massive amounts of data, the speed of light imposes a fundamental limit on latency. For a satellite in geostationary orbit, the round-trip time for a signal is significant, potentially approaching half a second. While this delay would be acceptable for the massive, non-real-time workloads associated with training foundational AI models, it would be unsuitable for applications requiring instantaneous feedback, such as video conferencing or financial trading. Data centers in lower Earth orbit could reduce this latency, but would require a much larger constellation of satellites to ensure continuous coverage.
Space Debris and Security
Operating critical infrastructure in orbit introduces unique risks. The growing problem of space debris poses a constant threat of catastrophic impact. A collision could destroy an expensive facility and create a cloud of debris that endangers other satellites. Furthermore, these data centers would be high-value targets, requiring robust physical and cybersecurity measures to protect them from potential threats. Ensuring the resilience and security of such assets will be a paramount concern for any future operator.
