IBM is exploring the feasibility of moving data centers off-planet as demand for computing power surges and strains terrestrial infrastructure; global electricity demand from these centers is projected to double by 2030. Tech leaders including those at Google, SpaceX and Amazon are increasingly looking to space to address resource limitations and environmental concerns associated with land-based facilities, which currently consume 23 cubic kilometers of water annually, a figure predicted to increase 129% by 2050. “This planet is so beautiful, and so unusual, this is the one that we’re going to want to protect,” said Amazon Founder Jeff Bezos at a 2024 summit, reflecting a growing sentiment that off-world solutions are not merely aspirational but necessary. SpaceX, for example, has filed plans with the FCC to launch as many as 1 million solar-powered satellites, while Elon Musk believes the cost of deployed AI in space will drop below the cost of terrestrial AI much sooner than people expect. Amazon filed a petition to deny SpaceX’s application, and astronomers said the satellites would permanently scar the night sky. This March, Blue Origin filed with the FCC for permission to deploy nearly 52,000 satellites as part of Project Sunrise, its proposed orbital data center system.
AI-Driven Demand Fuels Search for Off-Planet Data Centers
The relentless surge in artificial intelligence applications is rapidly escalating the demand for computational resources, prompting a search for new infrastructure locations beyond Earth. Global data center electricity consumption is projected to double by 2030, according to Gartner, placing immense strain on existing land-based systems and igniting public concern over escalating utility costs and environmental repercussions.
The primary appeal of off-planet data centers lies in the promise of virtually limitless, continuous energy; the sun emits more power than 100 trillion times humanity’s total electricity production. Google’s Project Suncatcher exemplifies this ambition, with test launches slated as early as 2027. “We want to put these data centers in space, closer to the sun,” explained Google’s Sundar Pichai in a December interview. “We will send tiny, tiny racks of machines and have them in satellites, test them out, and then start scaling from there.” SpaceX is pursuing an even more ambitious scale, filing to launch up to 1 million solar-powered satellites. Elon Musk articulated the long-term necessity of this approach, writing on SpaceX’s blog in February that space-based AI is obviously the only way to scale, and summarizing the rationale by stating that space is called “space” for a reason. While the vision is compelling, significant technical and economic hurdles remain. Despite the enthusiasm, OpenAI’s Sam Altman dismissed the idea as ridiculous, at least in its current form, a sentiment echoed by a Gartner report that labeled the excitement a peak and a bubble. Engineers are actively exploring innovative architectures, such as tiered pipelines that efficiently manage data flow from sensors to processing centers, to reduce costs and maximize resource utilization. Martin Schmatz of IBM Research explained that the goal is to be smart about what data is forwarded, envisioning a shared infrastructure where smaller companies can rent access to orbital compute.
Google’s Project Suncatcher and SpaceX’s Satellite Proposals
The pursuit of ever-increasing computational power is rapidly driving exploration beyond terrestrial limitations, with several major players now actively proposing orbital data centers as a viable, if ambitious, solution to looming infrastructure constraints. Gartner forecasts global data center electricity demand will double by 2030, creating strain on existing resources and prompting local resistance; this pressure is fueling a surge in interest regarding off-planet computing solutions. Google and SpaceX are at the forefront, each envisioning a future where data processing leverages the virtually limitless energy available in space, though their approaches differ significantly in scale and implementation. Google’s Project Suncatcher represents a measured approach, with test launches currently targeted for 2027. This strategy acknowledges the challenges of space-based computing, particularly the need for efficient energy harnessing and thermal management, while allowing for iterative development.
Despite these hurdles, researchers like Martin Schmatz at IBM Research are exploring innovative architectures, envisioning a tiered system where data is processed incrementally in space to reduce bandwidth demands and enable a shared infrastructure. Schmatz told IBM Think that the challenge is to efficiently manage the large volume of data produced by many sensors, noting that sending all that data to Earth at once is impractical.
I think the cost of deployed AI in space will drop below the cost of terrestrial AI much sooner than people expect.
Orbital Data Center Challenges: Radiation, Cooling, and Power
Blue Origin’s filing with the Federal Communications Commission for permission to deploy nearly 52,000 satellites as part of Project Sunrise underscores the escalating ambition surrounding off-world data storage, but realizing this vision demands overcoming significant engineering hurdles. While the allure of continuous solar energy and escaping terrestrial resource constraints are powerful motivators, the practicalities of building and maintaining computational infrastructure in space present formidable challenges, particularly concerning radiation hardening, thermal management, and power delivery. These aren’t simply scaling problems; they require fundamentally rethinking data center design. One critical issue is the relentless bombardment of cosmic radiation, a phenomenon that can corrupt data or permanently damage sensitive electronics. Existing AI hardware wasn’t engineered to withstand such an environment, necessitating either costly shielding or the development of radiation-tolerant components. Beyond radiation, effective cooling presents a unique obstacle; the vacuum of space prohibits convection, the natural process that drives heat dissipation on Earth.
Current designs often rely on large radiator panels, sometimes exceeding the size of the computing hardware itself, to radiate heat away, a bulky and potentially inefficient solution. The sheer scale of power requirements also looms large. The International Space Station, a massive structure covering the area of a football field, generates enough power for a single rack in a terrestrial data center, which can house thousands. Scaling to a facility capable of handling significant computational loads demands innovative approaches to energy harvesting and distribution. However, researchers like Schmatz and his colleagues are exploring a tiered computing architecture, envisioning a network of sensors, space data centers, and terrestrial links. “It’s like a small telecom company using the network of a larger one,” Schmatz noted, suggesting a pathway toward cost-effective orbital computing. Any viable system, Schmatz stressed, must also prioritize longevity and security.
There were still many people who said you couldn’t run an H100 in space because of the thermal dissipation and radiation hardening problems.
Varda’s Cost Analysis Compares Space vs. Terrestrial Facilities
The escalating demands of artificial intelligence are forcing a reevaluation of where computing happens, and a surprising contender has emerged: space. While the promise of limitless solar energy and reduced terrestrial strain fuels this ambition, practical economic realities remain a significant hurdle, as highlighted by recent analyses comparing orbital and land-based data center costs. Varda, a space startup, has developed a web-based calculator to assess these costs, revealing a stark disparity; a 1-gigawatt orbital facility would currently cost around USD 51 billion to build and operate over five years, exceeding the USD 16 billion price tag for an equivalent terrestrial installation. “If you run the numbers honestly, the physics doesn’t immediately kill it, but the economics are savage,” explained Andrew McCalip, an engineer at Varda, emphasizing the financial challenges. However, a growing number of researchers believe that reframing how orbital infrastructure is built and shared could dramatically alter this equation.
This model envisions a network where sensors, such as rovers and low-orbit satellites, initially collect data and then relay it to more powerful space data centers for aggregation and filtering before transmission to Earth. This tiered system also opens the door to a new commercial model, potentially allowing smaller companies to access orbital computing resources without the massive upfront investment of building their own infrastructure.
The worst thing is that somebody comes to the idea, ‘Hmm, I’ll place 200 of my satellites 3,000 miles in the air,’ and some other operator says the same.
IBM Research Proposes Tiered Pipeline for Shared Space Compute
The vision of vast data centers orbiting Earth often conjures images of monolithic structures, single entities undertaking the immense task of processing and relaying information. However, a new approach proposed by IBM Research suggests a fundamentally different architecture, one built on distributed, shared infrastructure rather than centralized power. Researchers are increasingly focused on reframing the economics of space-based computing, moving beyond simply replicating terrestrial models in orbit and toward systems designed specifically for the unique challenges and opportunities of the space environment. This shift in thinking is detailed in a 2024 paper, “Designing (Not Only) Lunar Space Data Centers,” which outlines a tiered pipeline for processing data collected in space. The concept draws a parallel to modern cellular networks, where seamless handoffs between towers allow for continuous connectivity.
Instead of a single, all-encompassing orbital platform, the IBM Research team envisions a system where data moves progressively up the chain, starting with low-power sensors like rovers and satellites. These sensors, brimming with raw data, would initially offload processing to space data centers, intermediate nodes capable of aggregating signals, filtering noise, and compressing results before transmission back to Earth. The potential for a shared infrastructure is particularly compelling. Rather than requiring companies to invest in entire orbital systems, this model allows smaller entities to launch specialized sensor hardware and rent access to the upper tiers of the network.
We will send tiny, tiny racks of machines and have them in satellites, test them out, and then start scaling from there.
