The global competition for AI computing power has intensified, with the industry diverging into two distinct paths. Elon Musk's SpaceX is making advanced moves in space-based computing, aiming to transcend Earth's resource limitations. In contrast, Japan's Softbank Group Corp is firmly betting on terrestrial, physical computing infrastructure. On June 23, local time, Softbank founder Masayoshi Son publicly challenged Musk's concept of "space data centers" during the annual shareholder meeting of its telecom subsidiary. He stated the model has limited commercial value, asserting that the winner in the artificial intelligence race will ultimately be determined by large-scale, ground-based computing power.
During the shareholder Q&A, Son systematically deconstructed the cost and timeline logic of space data centers. He noted that the sole advantage of space-based computing is its potential to reduce electricity costs by leveraging the vacuum, low-temperature environment of space and solar energy. However, this benefit is far from sufficient to offset its comprehensive drawbacks. From an industry cost perspective, electricity accounts for only about 7% of the total lifecycle operating costs of an AI data center. The core expenses lie in procuring key hardware like GPUs and servers, as well as investments in facility construction, land, and network bandwidth. Space-based projects, on the other hand, must bear the astronomical costs of rocket launches and in-orbit equipment maintenance, while also facing unavoidable communication latency issues, resulting in extremely poor overall cost-effectiveness.
Regarding the timeline, Son's judgment was even more decisive. He acknowledged Musk as a "brilliant innovator" but stated that the technology and commercialization of space data centers are immensely difficult, requiring at least over a decade to materialize. In contrast, the global AI industry's competitive landscape is expected to be determined within the next three to five years. The lengthy development cycle for space-based infrastructure would completely miss this critical competitive window.
Softbank Group President Junichi Miyakawa also commented during the same shareholder meeting, noting that space-based computing operations still face numerous unresolved technical challenges. Consequently, Softbank will not participate in related projects in the short term. He further pointed out that Japan's domestic data center construction pace is lagging, emphasizing the corporate responsibility to address this AI infrastructure gap and support industry development.
Based on the core strategy that "AI victory will be decided on the ground," Softbank is comprehensively ramping up its global physical computing infrastructure, adhering to a "preemptive" competitive strategy. For its overseas expansion, Softbank has launched two major supercomputing projects in Europe and the US. On May 31 this year, it officially announced a cumulative investment of up to 75 billion euros in France to build Europe's largest AI computing cluster. This project will be implemented in phases, with an initial investment of 45 billion euros. It plans to establish 3.1 gigawatts of computing capacity in northern France by 2031, with future expansion to 5 gigawatts. The project will leverage France's nuclear power resources for stable energy supply and collaborate with Schneider Electric to create an industrial hub in Dunkerque that integrates AI infrastructure with robotics manufacturing, serving computing demand across Europe.
Additionally, Softbank plans to establish a super AI data center in Ohio, USA, with a proposed investment of approximately 500 billion dollars, aiming to create the largest single-site AI computing hub in history and continuously solidify its global ground-based computing foundation.
On the capital and technology front, Softbank is already deeply entrenched in the core AI arena, having invested over 64 billion dollars in OpenAI and holding about a 13% stake, making it the second-largest external investor after Microsoft. Simultaneously, Softbank is addressing chip supply chain gaps by acquiring UK AI chip unicorn Graphcore and US chip design company Ampere Computing, thereby enhancing its computing hardware industry chain and further amplifying the scale advantage of its ground-based computing power. The company also plans to enter the US new cloud market and the data center energy storage battery sector, aiming to comprehensively improve the supporting AI ecosystem.
It is noteworthy that this precise bet on the AI sector has allowed Son to completely emerge from the shadow of past investment setbacks like WeWork. He recently regained the title of Asia's richest person after more than a decade, returning to the center of capital market attention through solid ground-based computing infrastructure investments.
In contrast, Musk's SpaceX remains committed to the ultimate vision of space-based computing, outlining a disruptive industrial blueprint in its prospectus and official documents. The company plans to utilize its Starship reusable launch technology to send megaton-scale computing equipment into low Earth orbit. By exploiting space's unlimited solar energy and natural low-temperature cooling advantages, it aims to build low-cost, high-power orbital data centers, breaking the physical constraints Earth's power supply imposes on AI development.
According to the plan, SpaceX will deploy millions of orbital computing satellites, achieving high-speed transmission through inter-satellite laser links and connecting with the existing Starlink network. Formal deployment is projected to begin as early as 2028. Musk has boldly predicted that achieving megaton-scale orbital payload delivery annually could enable the construction of a hundred-gigawatt-scale solar-powered AI satellite cluster, with the long-term potential to reach one terawatt of supercomputing scale.
In reality, the concept of space data centers has long been controversial within the industry, and Son is not the first to question it. OpenAI CEO Sam Altman stated as early as February this year that the orbital data center model is impractical in the short term. The high launch costs, stringent technical barriers, and difficult in-orbit maintenance make large-scale deployment unlikely within a decade, possessing only long-term exploratory value. Coupled with recent significant stock price volatility and continuous market capitalization decline for SpaceX, market sentiment toward the commercial viability of capital-intensive space-based computing models has grown increasingly cautious.
The current industry consensus is that for the foreseeable future, ground-based data centers—with their mature technology, controllable costs, and stable latency—will remain the absolute mainstay of AI computing power supply. Even if space-based computing achieves technological feasibility, it would only be suitable for low-latency-demand scenarios like offline training and cannot replace the core position of ground-based computing power.
On one side lies the超前 exploration and long-term布局 of space-based computing; on the other, the pragmatic implementation and immediate抢占 of ground-based computing. The博弈 between these two paths is profoundly reshaping the competitive landscape of the global AI industry.
Comments