Softbank is aiming to dominate the American AI cloud market with a scale of 10GW! Both Son Masayoshi and Mark Zuckerberg want to turn unused GPUs into cash flow. Is the surplus AI computing power finally being put to practical use?

The news that Meta's parent company, Meta, is preparing to lease/sell idle AI computing power infrastructure has triggered concerns of AI computing power oversupply. This has heavily impacted AI computing power investment themes and led to a global tech stock crash. SoftBank Group Corp., led by legendary investor Masayoshi Son, and its telecommunications business department will begin leasing out the AI computing power infrastructure resources they have deployed on a large scale in recent years to American companies starting in the next fiscal year. They are trying to leverage the company's growing AI data center project reserves to achieve a more robust profit path. However, SoftBank's move has further raised concerns in the market that AI computing power oversupply may become a pessimistic reality. Meta founder and leader Mark Zuckerberg exploring selling excess AI computing power, combined with SoftBank's plans to launch AI cloud computing infrastructure services in the US market, has severe short-term implications for AI semiconductor stocks, computing power industry leaders, and new cloud computing (Neocloud) manufacturers closely associated with AI computing power infrastructure construction. However, according to Wall Street analysts, a more accurate understanding is not that "Meta, SoftBank, and other large-scale computing power purchasers are admitting defeat" or "global computing power oversupply," but rather transforming the huge AI capital expenditure from a simple cost center into a monetizable computing power asset balance sheet. SoftBank has established SB Neo and plans to lease AI chips and cloud services in the US, expanding the AI data center capacity to approximately 10GW in the future. This largely indicates that tech giants are still betting on the long-term surge in demand for training and reasoning computing power. The US Department of Energy also revealed that SoftBank's SB Energy subsidiary plans to build 10GW of additional power generation in Ohio to support a 10GW data center project. This essentially integrates "computing power infrastructure + power resources," rather than a complete collapse of training/reasoning computing power demand related to AI. Regarding Meta's latest dynamics of selling/leasing computing power resources, Wall Street analysts unanimously expect that Meta's plan to sell excess AI computing power is an effort to transform the massive AI capital expenditure into external income elasticity. This not only reflects the tech giant's desire to increase the return on AI investments but also intensify competition with CoreWeave, Nebius, and other new cloud manufacturers. As for the sharp drop in US and South Korean stock markets, analysts believe that the market is more concerned about the return on investment in AI infrastructure, repeated construction, increased competition, and record leverage and crowded position trading, rather than the proven global oversupply of computing power. From telecommunications operators to AI computing power suppliers: SoftBank's US Neocloud plan, Son's bet on a new "power + chip + cloud" flywheel Two companies (SoftBank Group and its telecommunications subsidiary) announced in a statement on Thursday that they will establish SB Neo Inc. this month to provide comprehensive AI cloud computing services, including AI chips and high-performance network infrastructure access to large companies including mega-scale cloud service providers and leading AI computing leaders such as OpenAI. Junichi Miyakawa, head of SoftBank's telecommunications division, stated that this large joint venture focusing on the "new cloud" domain of artificial intelligence plans to gradually expand its cloud computing resources. The ultimate goal is to supply a massive amount of AI data center capacity of 10GW around 2030 for large-scale AI model training and the enormous cloud-based reasoning computing power demand brought by AI intelligent agents. According to sources, launching new cloud services in the US could easily double or even triple the annual operating profit of SoftBank, the mobile operator under the SoftBank group, to a scale of 3-4 trillion yen (185-250 billion USD). In an interview, Miyakawa stated that this massive new cloud business, with SoftBank holding a 51% stake in the mobile operator SoftBank and its parent company holding 49%, has the potential to generate profits of a "different magnitude." He said, "We view the full launch of our cloud computing business in the US market as the second entrepreneurship of our company." This third largest mobile operator in Japan has played a crucial foundational role for its parent company SoftBank Group, helping Masayoshi Son successfully secure huge financing for his early venture capital investments. In recent years, this billionaire's business focus has been fully concentrated on AI computing power infrastructure, utilizing the endless demand for AI computing resources from global users and part of his ambitious goal to spread the most cutting-edge AI technology across society. SoftBank is already a major shareholder of Arm Holdings Plc, the owner of the ARM instruction set architecture (with a stake close to 90%). Arm's instruction set architecture technology is widely used in the entire consumer electronics industry and is increasingly used as the foundation architecture for server CPU products at ultra-large-scale data centers. SoftBank's significant investments in recent years in companies like Arm, Graphcore, Ampere Computing, their recent massive investments in projects like OpenAI and the US "Stargate" AI infrastructure project reflect their comprehensive layout from the bottom AI hardware architecture to AI computing power infrastructure clusters to the AI application layer. SoftBank's AI platform has already evolved into a complete "AI stack": with Arm architecture and Arm's in-house data center CPU at the bottom; acquired UK AI chip design company Graphcore Ltd. and recently acquired Ampere in the middle; and OpenAI, Stargate, and a cloud-based AI computing platform in partnership with Oracle at the top level. Masayoshi Son has openly expressed his goal for SoftBank to become the largest provider of AI computing power and application layers in the era of "super artificial intelligence" (ASI) over the next decade. He has taken an almost "all in" aggressive investment approach towards OpenAI, the developer of ChatGPT and a global leader in AI large models. The rising tide of new cloud forces (Neoclouds) - a new category of cloud computing infrastructure providers renting out immense computing power for AI training/reasoning fields to address the continuous surge in demand for computing capacity. New cloud companies like CoreWeave Inc. and Nebius Group NV are quickly leasing out dedicated AI chips and complete AI computing power infrastructure resources needed for customers to train and run AI models. SoftBank's new cloud operating business may have a ready-made customer in OpenAI; its parent company SoftBank Group has promised that the total investment amount in OpenAI will reach approximately $65 billion by October. However, this field has become increasingly crowded. In addition to specialized new cloud service providers, traditional cloud computing industry leaders like Amazon Web Services (AWS), Microsoft Azure, and Google Cloud are also actively selling access to AI computing power resources. More significantly, according to reports, Facebooks parent company, Meta Platforms Inc., is also formulating similar entry plans. Miyakawa stated that the significant advantage of SoftBank's upcoming US cloud computing business in the stock market lies in its unique ability to source strong power supply, mainly coming from SoftBank's vast gas power plant system. The project envisioned by SoftBank under Masayoshi Son's leadership in Ohio is a large AI infrastructure construction project focused on data centers with a scale of $50 billion. The project's capacity of 10GW will make it one of the world's largest AI data center projects. In Japan, SoftBank's telecommunications department is building data center parks in Hokkaido, in the northern island, and Sakai City, Osaka. From a scarcity of GPU to proactively laying out AI computing power infrastructure monetization, Meta and SoftBank reshaping the computing power supply system Although concerns of "oversupply of computing power" led to a crash in US and global tech stocks, Wall Street analysts believe that Meta's sale of computing power, coupled with SoftBank's establishment of SB Neo to enter the US AI cloud market, is not essentially a bet on "computing power oversupply," but a bet on the long-term expansion of demand for training and reasoning computing power into the giga-scale infrastructure era. SoftBank plans to launch its new cloud service in the US by the end of the company's new fiscal year in March 2028, targeting large enterprises in the US and mega-scale cloud manufacturers to provide the computing power needed for large model training and reasoning. This will be gradually expanded based on SoftBank's development of a 10GW-level energy and AI infrastructure; combined with the US Department of Energy's earlier disclosure, SoftBank and SB Energy plan to build 10GW of additional power generation capacity, of which at least 9.2GW will be natural gas-powered to support the development of 10GW data centers. This shift reflects that the real competition has evolved from "who has GPU resources" to "who can package electricity, land, data centers, AI chips, network infrastructure, and customer contracts into a supercomputing power platform that can be financed and actively monetized." Meta exploring the sale of excess AI computing power, which caused short-term pressure on chip stocks and new cloud manufacturers. However, the more accurate understanding is not that Meta is admitting defeat or that there is an oversupply of computing power globally, but rather that the huge AI capital expenditure is being transformed from a simple cost center into a monetizable computing power asset balance sheet. Meta is considering selling excess AI computing power and may offer a model access service similar to AWS's Bedrock. This news has boosted Meta's stock price but put pressure on new cloud companies like CoreWeave and Nebius, as Meta transitions from being a customer to a potential competitor. The core of the debate among institutions like Bank of America, UBS, and Morgan Stanley is the sale of computing power, which aims to clarify the return on investment (ROIC) of AI and act as an "EPS bridge" and cash flow buffer. For new cloud companies, this is a test of pricing power and renewal capabilities under commercial pressure from large tech customers. In the view of Wall Street analysts, the single-day crash in US and South Korean stock markets appears to be more about deleveraging triggered by crowded positions, leveraged ETFs, momentum trading, and the narrative of "computing power oversupply," rather than a real collapse in AI demand. The KOSPI composite index in South Korea plummeted nearly 10% after regulatory warnings about the risks of leveraged ETFs investing in chip stocks, where Samsung Electronics and SK Hynix dropped by over 12% and triggered temporary trading halts; however, on the same day, SK Hynix announced plans to invest 100 trillion won (approximately $64.3 billion) in building new large-scale DRAM/NAND/HBM memory chip manufacturing facilities by 2029. Therefore, analysts believe that the price behavior reflects the market's concerns about whether the AI hardware chain has risen too much in a phase, whether the leverage is too high, and whether capital expenditures can realize returns, rather than a sudden disappearance of HBM, AI server clusters, and massive cloud-based AI reasoning demand leading to computing power oversupply. For investors focusing on the AI super bull market, the real investment insight from Meta and SoftBank's layout of AI computing power infrastructure sales/leasing channels is that the AI infrastructure theme trading is transitioning from "absolute scarcity, blind purchasing of hardware" to the second stage of "scarcity still exists, but verification of utilization, rental fees, customer contracts, power costs, and free cash flow must be validated." CoreWeave had previously raised its minimum capital expenditure for 2026 due to strong demand for AI cloud, with a backlogged revenue order of up to $99.4 billion, indicating that high-quality AI computing power infrastructure is not being ignored. However, as Meta, SoftBank, SpaceX, traditional cloud providers, and new cloud manufacturers all enter the computing power rental space, the future winners will shift from merely owning GPU and other AI computing power resources to owning a comprehensive advantage in low-cost electricity, stable customers, advanced scheduling software, model/API access points, and capital cost. In other words, this is not the end of the AI super bull market, but a new stage in the AI computing power infrastructure investment market entering price discovery, ROIC judgment, and financial market deleveraging and clearing of crowded positions.