SoftBank’s CEO isn’t the only one with questions about Elon Musk's orbital data center hype

The vision of data centers floating in Earth's orbit, harnessing endless solar power and the cold vacuum of space, sounds like something straight out of a science fiction novel. Yet, this is precisely the ambitious future Elon Musk and SpaceX are pushing towards, with plans for a vast network of AI-centric satellites. However, not everyone in the tech world is convinced by this grandiose vision. A notable voice of dissent comes from SoftBank Group CEO Masayoshi Son, who has openly questioned the economic sense behind placing data centers beyond our planet. This growing debate highlights the significant opportunities and formidable challenges at the cutting edge of AI infrastructure and space technology.

Elon Musk's Orbital Ambition: "Starmind" and the AI Future

Elon Musk, known for his audacious goals with Tesla and SpaceX, has set his sights on revolutionizing AI infrastructure by moving it to space. SpaceX has revealed plans for a constellation of up to one million AI satellites, reportedly named "Starmind," designed to function as orbital data centers. This initiative aims to address the escalating demand for computational power driven by artificial intelligence, which is straining terrestrial power grids and facing increasing community opposition due to environmental concerns.

Musk suggests that many of the technologies needed for this endeavor are already developed for SpaceX's Starlink constellation. The idea is to leverage Starlink V3 satellites, which feature high-speed laser links, to host significant onboard computing capabilities. SpaceX's initial design for an orbital data center satellite, dubbed "AI1," is envisioned as a 70-meter spacecraft carrying up to 150 kilowatts of peak compute capacity, roughly equivalent to a rack of NVIDIA GB300 chips. These AI satellites would operate at altitudes between 600 and 800 kilometers, aiming for a latency of about 3 milliseconds. The company plans to use its heavy-lift Starship rockets to deploy these AI satellites, with Musk projecting the ability to launch 30 to 50 AI1 satellites per flight. Volume production is targeted for late 2027 at a new facility called Gigasat, with prototypes scheduled for launch in early 2027.

The core arguments for orbital data centers, as presented by Musk and other proponents, revolve around several key advantages:

• Unlimited Solar Power: In orbit, solar panels can theoretically receive continuous sunlight, free from atmospheric interference, night cycles, or weather conditions. This offers a higher capacity factor compared to terrestrial solar and potentially lower energy costs.
• Natural Cooling: The vacuum of space is often cited as an ideal heat sink, allowing for passive radiative cooling without the need for water or air, which are major resource consumers for Earth-based data centers.
• Reduced Terrestrial Constraints: Orbital data centers would eliminate issues like land acquisition, zoning restrictions, local community opposition, and reliance on strained power grids and water supplies.
• Edge Computing for Space-Based Workloads: For satellites collecting vast amounts of data (e.g., Earth observation, defense systems), processing data in orbit rather than sending it all back to Earth could significantly reduce latency and bandwidth requirements.

Masayoshi Son's Skepticism: A Grounded Perspective

Despite the futuristic appeal, SoftBank Group founder Masayoshi Son has publicly dismissed the immediate viability of orbital data centers, particularly for the critical AI race unfolding now. Speaking at a SoftBank shareholder meeting, Son questioned the fundamental economic premise, arguing that electricity costs, while reduced in space, constitute only a small fraction (around 7%) of the total operational expenses for AI infrastructure. He emphasized that the majority of costs lie in chips and other hardware.

Son contended that any savings from lower electricity costs in orbit would be heavily outweighed by the enormous expenses associated with launching hardware into space, maintaining it in a hostile environment, and managing communication delays. He stressed the importance of speed in the AI competition, stating, "He who strikes first wins," and suggested that the next few years are crucial for deciding the AI race, favoring infrastructure that can be built and deployed on Earth now, rather than relying on rockets still under development. SoftBank, under Son's leadership, is focusing on building "formidable" data center capacity on Earth, having already committed substantial investments, including approximately $65 billion to OpenAI and hundreds of billions more for global data center infrastructure.

Son is not alone in his skepticism. OpenAI CEO Sam Altman has reportedly called orbital data centers "ridiculous," aligning with Son's more grounded view despite OpenAI's ties to SoftBank's terrestrial buildout.

The Hard Realities: Challenges of Space-Based Data Centers

The skepticism from industry leaders like Masayoshi Son is rooted in significant technical and economic challenges that orbital data centers face:

• Massive Cost and Operational Complexity: Launching and maintaining hardware in orbit remains vastly more expensive than terrestrial infrastructure. Estimates suggest a hypothetical 1-gigawatt orbital data center could cost over three times that of an equivalent ground facility, with launch and satellite expenses accounting for about 60% of the total investment. Achieving cost parity would require launch-related costs to fall by roughly 70%.
• Cooling in a Vacuum: While space is cold, the vacuum environment makes heat dissipation surprisingly difficult. On Earth, data centers use convection (air or water) to move heat away from components. In space, heat can only be radiated, which is a much slower process. Engineers face a major challenge in designing radiator systems large and efficient enough to dissipate the enormous waste heat generated by powerful AI chips. Some calculations suggest a 1-megawatt orbital data center would need about 1,600 square meters of radiators, roughly the size of a hockey rink.
• Radiation and Hardware Degradation: Space is a harsh environment with constant bombardment from cosmic rays and solar radiation. This radiation can corrupt data, cause "bit flips," and significantly degrade commercial off-the-shelf hardware not designed for space. Rad-hardened chips exist but are typically years behind commercial counterparts in performance and far more expensive.
• Maintenance and Refresh Cycles: Unlike Earth-based data centers where failed servers can be swapped in minutes, maintenance in orbit is extraordinarily difficult and expensive, if not impossible for individual components. AI hardware advances rapidly, with refresh cycles every 3-5 years on Earth. An obsolete or dead GPU in orbit would likely remain so, posing a significant economic and operational challenge.
• Latency for Earth-Bound Applications: While orbital data centers might reduce latency for space-based workloads, signals between an orbital data center and Earth still take measurable time to travel, which could impact real-time AI applications sensitive to delay.
• Space Debris and Orbital Crowding: A constellation of millions of satellites, as proposed by SpaceX, raises concerns about increasing space debris and the risk of collisions, which could interfere with other space missions and astronomical research.
• In-Space Assembly: Large-scale orbital data centers would likely need to be assembled in space, requiring new equipment and capabilities for in-space servicing, assembly, and manufacturing, which are currently underdeveloped.

The Broader Landscape: Other Players in Orbital Computing

Despite the challenges and skepticism, the allure of space-based computing is undeniable, especially with the surging demand for AI infrastructure. Several other companies are actively exploring or developing orbital data center solutions:

• Orbital: This company is building "the first space-based AI compute infrastructure," aiming to launch its first mission, a "Pathfinder," in 2027 to demonstrate AI inference compute in low Earth orbit. Their vision includes a constellation of satellites running AI servers, powered by solar and cooled by the vacuum of space.
• Starcloud: Backed by NVIDIA, Starcloud is developing data centers in space to provide GPU compute to other satellites and address AI's energy demand. They successfully launched a satellite with an NVIDIA H100 GPU in late 2025, becoming the first to train an LLM and run Google's Gemma in space.
• Axiom Space: Known for its work on commercial space stations, Axiom Space is developing orbital data center nodes and computing infrastructure for cloud computing, AI, and cybersecurity in space. They launched their first two orbital data center nodes in low-Earth orbit in January 2026, building on a prototype deployed on the ISS in late 2025.
• Google (Project Suncatcher): Google has also been testing its own satellite concept, Project Suncatcher, aiming to launch prototype satellites with TPU hardware in space by early 2027 to validate systems.
• Blue Origin: Jeff Bezos' space venture has also predicted "giant gigawatt data centers in space" within two decades.
• Sophia Space: This Pasadena-based startup is developing orbital computing and data centers using a unique passive cooling system and solar panels, with plans to begin testing technology in space by next year.

These companies are pursuing various architectural concepts, from "smart satellite" edge nodes with modest compute capabilities on individual communication satellites to "orbital server rack" clusters and even monolithic "orbital data center" platforms.

Industry Implications and the Road Ahead

The debate around orbital data centers highlights a critical juncture in the evolution of AI and space technology. The demand for AI compute is skyrocketing, with terrestrial data centers facing genuine constraints in terms of power, land, and environmental impact. This pressure is driving exploration into unconventional solutions, including space.

While the long-term potential of orbital data centers for AI is compelling – offering a path to scale compute far beyond Earth's limitations – the immediate economic and technical hurdles are substantial. Masayoshi Son's comments serve as a reminder that the "why" must be as robust as the "how." For many, the focus remains on optimizing and expanding terrestrial infrastructure, where investment is projected to reach trillions of dollars in the coming years.

The industry is in an early "Demonstration Phase" (2025-2027), focused on proving that complex computing hardware can survive and operate effectively in space. The success of companies like Starcloud in training AI models in orbit is a significant step, but scaling to the "gigawatt" levels envisioned by some will require breakthroughs in thermal management, radiation hardening, in-space assembly, and dramatic reductions in launch costs.

Ultimately, the future may not be an either/or scenario. Orbital data centers might complement terrestrial infrastructure, serving specialized "space-native" workloads, AI training jobs that can tolerate higher latency, or acting as resilient, physically isolated computing hubs. However, the timeline for widespread commercial viability at scale remains a subject of intense debate and significant engineering challenges.

Frequently Asked Questions

What are orbital data centers?

Orbital data centers are proposed computing facilities located in Earth's orbit, typically in low Earth orbit (LEO). The idea is to leverage the unique space environment—like abundant solar power and the vacuum for cooling—to host powerful servers for artificial intelligence and other high-demand computing tasks.

Why is SoftBank's CEO skeptical about them?

SoftBank CEO Masayoshi Son is skeptical because he believes the immense costs of launching and maintaining hardware in space, along with challenges like communication delays, far outweigh the potential savings from reduced electricity consumption. He argues that chips and other hardware represent the majority of data center costs, not just power.

What are the main benefits pitched for orbital data centers?

Proponents highlight benefits such as virtually unlimited solar energy, natural passive cooling in the vacuum of space, freedom from terrestrial constraints like land and power grid limitations, and the ability to provide edge computing for space-based workloads with lower latency.

What are the biggest challenges facing orbital data centers?

Key challenges include the massive cost of launching and maintaining infrastructure in space, difficulties with heat dissipation in a vacuum, the damaging effects of space radiation on electronics, the near impossibility of in-orbit maintenance and hardware upgrades, potential communication latency for Earth-based users, and concerns about increasing space debris.