SpaceX has filed for an initial public offering that its founder Elon Musk says will finance an ambitious push to transform the rocket maker into a major artificial intelligence infrastructure provider. The plan centers on launching up to one million data-center satellites into orbit, a concept designed to bypass the physical limits of energy, land, and water that constrain computing on Earth.
The proposal immediately triggered comparisons with earlier attempts to rethink how and where data centers operate. One of the most notable precedents came from Microsoft, which explored underwater computing through its experimental initiative known as Project Natick.
Back in 2015, Microsoft deployed a shipping-container-sized data center on the seabed off Scotland. The project aimed to reduce energy consumption by using natural seawater for cooling while tapping offshore renewable energy sources such as wind and tidal power. At the time, it was presented as a potential breakthrough for the data-center industry.
Technically, the experiment succeeded. According to sources familiar with the project, the underwater system met all its engineering objectives and demonstrated strong reliability. However, despite these achievements, Microsoft ultimately abandoned the concept more than two years ago due to a lack of customer demand and unfavorable economics.
A spokesperson for Microsoft confirmed that while the company no longer operates underwater data centers, it continues to use the research as a platform for testing ideas related to reliability and sustainability. The shift highlights a critical distinction between technical feasibility and commercial viability—an issue now central to the debate around SpaceX’s plans.
Industry specialists say the parallels between the two concepts are striking. Both rely on modular, sealed units that are expensive to deploy and difficult to upgrade or repair once operational. These characteristics run counter to the needs of the rapidly evolving AI sector, where hardware improvements occur at a fast pace and flexibility is essential.
Roy Chua, founder of research firm AvidThink, noted that the challenges SpaceX faces could be even more severe than those encountered underwater. Operating in space introduces additional complications, including extreme temperatures, radiation exposure, and the absence of conventional cooling mechanisms.
Cooling remains one of the most significant unresolved issues. On Earth, data centers rely heavily on water or air-based systems to dissipate heat generated by high-performance computing. In orbit, where there is no atmosphere, managing heat efficiently becomes far more complex and energy-intensive.
Cost is another major obstacle. SpaceX’s IPO could raise as much as $75 billion, potentially making it the largest in history. The company has already expanded its AI ambitions through its acquisition of xAI, which includes assets such as the social media platform X and the chatbot Grok.
Yet analysts warn that the scale of investment required to deploy orbital data centers would far exceed even that substantial capital raise. According to estimates from MoffettNathanson, Musk’s vision of deploying one million AI satellites could ultimately cost trillions of dollars.
Launching hardware into space remains prohibitively expensive, even with recent advances in reusable rocket technology. For space-based data centers to become commercially viable, launch costs would need to fall dramatically—from thousands of dollars per kilogram today to just a few hundred dollars per kilogram.
Tim Farrar, an analyst at TMF Associates, emphasized that the key question is not whether the technology can work, but whether it makes economic sense compared with expanding conventional data centers on Earth. Land-based facilities continue to dominate because they are easier to upgrade, maintain, and scale.
This was precisely the issue that undermined Microsoft’s underwater experiment. Although the technology functioned as intended, customers showed little interest in adopting it at scale. Instead, they preferred traditional data centers, which allowed faster and more cost-effective upgrades as AI workloads grew.
The “locked-for-life” design of modular units poses a particular challenge. AI chips are improving rapidly, often on an annual cycle, while infrastructure such as satellites or submerged containers typically has a lifespan of five to seven years. This mismatch creates a risk that expensive hardware could become obsolete long before it is replaced.
Space-based systems would likely face the same limitation, with even fewer options for mid-cycle upgrades or repairs. Once deployed in orbit, accessing and servicing satellites is significantly more difficult and costly than maintaining terrestrial facilities.
Despite these concerns, Musk remains confident that technological innovation will overcome the barriers. He has pointed to advancements in reusable rockets, particularly SpaceX’s next-generation Starship, as a key factor in reducing launch costs and enabling large-scale deployment.
However, Starship itself has faced delays and technical setbacks. Since 2023, several of its suborbital test flights have ended in explosions, highlighting the challenges involved in developing a fully reusable heavy-lift rocket. Analysts estimate that achieving Musk’s vision would require around 3,000 Starship launches per year—equivalent to roughly eight launches per day.
The concept of space-based data centers is not limited to SpaceX. Blue Origin, the space company founded by Jeff Bezos, has also expressed interest in orbital computing. Its Project Sunrise concept aims to add AI processing capacity in space while leveraging solar power.
Even so, experts generally view such initiatives as complementary rather than transformative. Claude Rousseau, research director at Analysys Mason, said that space-based data centers are unlikely to replace ground infrastructure in the foreseeable future. Instead, they may serve niche applications, such as supporting satellite constellations, military systems, or space stations.
The International Space Station already hosts experimental systems that process data in orbit, reducing the need to transmit large volumes of information back to Earth. These use cases illustrate the potential benefits of localized processing in space, but they remain limited in scope.
Skepticism also extends to the broader AI industry. Jensen Huang, head of Nvidia, has questioned the economics of orbital AI infrastructure. Speaking on a podcast earlier this year, he suggested that efforts should focus on improving ground-based systems before pursuing more complex solutions in space.
Critics argue that moving data centers off Earth risks creating new challenges rather than solving existing ones. Advances in energy efficiency, renewable power, and cooling technologies continue to improve the sustainability of terrestrial data centers, potentially reducing the need for radical alternatives.
Chua highlighted ongoing innovations in AI chip design, water recycling, and modular nuclear power as examples of solutions that could address current limitations without leaving the planet. These developments may prove more practical and cost-effective than deploying infrastructure in extreme environments.
Still, Musk’s argument is rooted in a long-term vision of exponential growth in AI demand. He has suggested that future scenarios—where autonomous vehicles dominate, robots outnumber humans, and space travel becomes routine—will require vast computing capacity beyond what Earth can sustain.
Not everyone is convinced by that premise. Farrar argued that framing Earth’s constraints as insurmountable may be overly pessimistic, noting that many challenges can be addressed through incremental improvements and better resource management.
The debate ultimately reflects a broader tension within the technology sector: the balance between ambitious, forward-looking innovation and practical, economically viable solutions. SpaceX’s IPO and its orbital data center strategy represent one of the boldest attempts yet to redefine the infrastructure underpinning artificial intelligence.
Whether that vision can be realized will depend not only on technological breakthroughs but also on market demand and financial sustainability. For now, the lessons from Microsoft’s underwater experiment serve as a reminder that even successful engineering projects can fall short if they fail to align with economic realities.
As the AI industry continues to expand, the pressure to find scalable, efficient computing solutions will only intensify. Space may eventually play a role in that future, but for the foreseeable horizon, Earth-based data centers remain the backbone of the digital economy.
