Debate Over Space Data Centers: SoftBank's Masayoshi Son Expresses Skepticism

Elon Musk’s SpaceX has proposed the concept of orbital data centers to address the increasing demand for computational resources, but skepticism from industry leaders persists. SoftBank Group founder and CEO Masayoshi Son recently expressed doubts during a shareholders’ meeting, stating that “building data centers in space will contribute little to cost reduction and will take far too long to materialize.” Son emphasized that the next few years in the AI race are far more critical than potential developments a decade down the line. His comments have sparked renewed debate on the feasibility of space infrastructure for AI computing needs.

In the latest episode of TechCrunch’s podcast “Equity,” reporters Kirsten Korosec, Sean O’Kane, and Anthony Ha delved into the issue. They discussed Son’s skepticism alongside topics like OpenAI’s custom chip initiative and Groq’s $650 million funding round.

Feasibility of Orbital Data Centers

SpaceX’s vision for orbital data centers aims to address challenges faced by terrestrial data center construction, such as land scarcity, energy limitations, and cooling issues, by utilizing the vastness of space as a new frontier.

The rapid growth in AI demand has outpaced the supply of terrestrial data center facilities. Obstacles like securing land, limited energy infrastructure, and environmental concerns hinder new construction. SpaceX has proposed leveraging its launch capabilities and satellite constellation technology to develop space-based data centers as a solution to these constraints.

However, Son’s perspective presents one of the clearest counterarguments to this vision. Speaking at the shareholders’ meeting, he argued that the time and costs required to build data centers in space render them an impractical solution for the immediate needs of the AI race. “The battle for AI in the next few years is far more important than what might happen a decade from now,” Son remarked, emphasizing the priority of securing short-term computational resources.

SpaceX’s Computational Resource Leasing Business

As a step toward realizing its long-term vision of orbital data centers, SpaceX has already launched a business of leasing its computational resources to external clients. According to reporter Sean O’Kane, SpaceX has signed contracts with major players like Google and Anthropic, as well as smaller companies, marking its first post-IPO computational resource leasing agreements.

This two-pronged strategy allows SpaceX to utilize its existing high-performance computational capabilities as a revenue source, while laying the groundwork for its long-term vision of orbital data centers. However, O’Kane has raised concerns about the sustainability of this business model in the long run.

The Irony of Skepticism and Strategic

Contradictions

Kirsten Korosec highlighted the “ironic” nature of Son’s skepticism toward space data centers, given SoftBank’s history of bold, seemingly risky investments, such as in WeWork and Arm. Son’s cautious approach to this concept appears to contrast with SoftBank’s traditional risk-taking strategy.

This stance may, however, be rooted in SoftBank’s own AI strategy. The company has been aggressively investing in AI infrastructure, as evidenced by its recent sale of shares in Boston Dynamics to Hyundai, which has become the robotics company’s sole owner. This move underscores SoftBank’s prioritization of short-term advantages in the AI sector over longer-term ventures like space data centers.

Space: A Strategy for Expanding SpaceX’s

Business?

When discussing Musk’s plans for orbital data centers, Sean O’Kane emphasized the aspect of building constellations of satellites that require regular replacements. This approach could ensure sustainable business for SpaceX through recurring launch contracts.

“O’Kane pointed out that Musk’s vision of orbital data centers is essentially a guarantee of more business for SpaceX,” suggesting that the project might serve more as a strategy for expanding SpaceX’s launch operations than a purely technological solution.

Structural Challenges of Computational

Resource Scarcity

The backdrop to this debate is the severe shortage of computational resources faced by the AI industry. Son’s central argument is that “many players are constrained by computational resources.” In the face of demand that far outstrips supply, players with computational capacity are increasingly entering the business of resource leasing.

From chipmakers like Groq, which is competing with NVIDIA, to companies like Allbirds, which pivoted from selling shoes to computational resource leasing, this sector is gaining traction as a new source of revenue. O’Kane described this trend with a hint of irony, calling it “the new oil.”

However, questions about the long-term sustainability of this business remain. The rapid evolution of AI chip technology and the emergence of alternative architectures might eventually alleviate the current resource constraints. Whether these business strategies are a short-term response or a viable long-term model is as yet uncertain.

Technical Challenges of Space Data Centers

Operating data centers in space presents numerous technical challenges that are difficult to tackle with current technology. These include the high costs of deploying large-scale infrastructure into orbit, the degradation of electronics due to space radiation, heat management issues, and the challenge of regular maintenance.

Particularly problematic is the need for high-speed, low-latency data transfer between space and Earth, which is critical for real-time AI applications. Current technology struggles to meet these requirements, and latency issues could prove to be a significant bottleneck.

As Korosec noted, “The scarcity of computational resources necessitates building as many data centers on Earth as possible.” Space data centers are unlikely to provide a short-term solution to this urgent problem, a view that is gaining consensus in the industry.

Competing AI Infrastructure Strategies

While the idea of space data centers is being debated, other approaches to addressing the shortage of AI computational resources are gaining momentum on Earth. OpenAI, for example, has announced plans to develop custom chips tailored for AI workloads, aiming to reduce its reliance on NVIDIA.

Similarly, Groq plans to use its $650 million funding round to expand its inference-focused computational services using proprietary chip architecture. The company aims to differentiate itself from competitors by offering low latency and high energy efficiency.

These strategies aim to address the immediate shortage of computational resources with incremental advancements in existing technology, contrasting with the long-term vision of orbital data centers. Son’s comments appear to underscore the importance of these pragmatic solutions.

Editorial Opinion

Masayoshi Son’s assertion that orbital data centers are not a practical solution to the urgent problem of AI computational resource scarcity is a valid one. Over the next three to six months, companies will likely compete to secure resources through incremental improvements to existing technologies. It remains to be seen whether chipmakers seeking alternatives to NVIDIA, or companies like OpenAI developing in-house chips, will achieve tangible breakthroughs within this timeframe.

In the long term, the feasibility of orbital data centers will depend on advancements in technology and changes in cost structures. If SpaceX can continue to reduce launch costs and develop mature data processing technologies for space, the concept could become viable within the next 10 to 20 years. However, by that time, the computational resource bottlenecks facing the AI industry today may already have been resolved.

The question we must ask is whether orbital data centers are truly a necessary technological solution or merely a marketing strategy to expand SpaceX’s business.

References

• SoftBank’s CEO isn’t the only one with questions about Elon Musk’s orbital data center hype - TechCrunch — Published on 2026-06-27