Ten gigawatts represents an extraordinary concentration of computing infrastructure
From Tokyo to the American cloud, SoftBank is staking its next chapter on the proposition that the world's hunger for artificial intelligence will outlast any single technology cycle. This month, the Japanese conglomerate launches SB Neo, a venture designed to rent AI computing power to U.S. firms at a scale — ten gigawatts of data center capacity by 2030 — that transforms infrastructure from a supporting role into a primary ambition. It is the latest expression of a recurring human impulse: to position oneself not merely as a user of a transformative technology, but as the indispensable ground beneath it.
- A genuine bottleneck in AI development — too few chips, too little data center space — has created a market where compute capacity commands premium prices and attracts serious capital.
- SoftBank enters a field already contested by fast-moving rivals like CoreWeave and Nebius Group, raising the stakes for a company whose bold bets have produced both legendary wins and painful losses.
- The 10-gigawatt target by 2030 demands sustained chip sourcing, massive real estate acquisition, and an unwavering conviction that AI demand will remain intense for years — none of which are guaranteed.
- SB Neo's integrated offering — combining SoftBank's existing telecom networks with purpose-built AI infrastructure — is the company's answer to the question of why hyperscalers should choose a newcomer over established players.
- The venture is set to begin serving customers next fiscal year, marking the moment SoftBank's strategic pivot from investor to infrastructure provider becomes operational reality.
SoftBank Group is launching SB Neo, a new company built to rent AI computing power to American firms that need specialized hardware for training and running large language models. The venture, announced this week alongside SoftBank's telecom subsidiary, positions the Japanese conglomerate as a direct competitor to CoreWeave and Nebius Group — companies that have already built profitable businesses supplying compute capacity to a market starved for it.
The scale of SoftBank's ambition sets SB Neo apart. Junichi Miyakawa, who leads the telecom unit, has outlined plans to reach 10 gigawatts of data center capacity by around 2030 — a figure equivalent to powering roughly 7.5 million homes, and a signal of deep confidence that demand for AI infrastructure will remain intense for years to come. The service is expected to launch next fiscal year.
What SoftBank brings that pure-play startups cannot easily replicate is an existing foundation: significant telecom networks, data center footprints across multiple continents, and the capital of one of Japan's largest corporations. SB Neo aims to offer hyperscalers an integrated package — connectivity, compute, and storage — rather than forcing customers to assemble solutions from multiple vendors.
Reaching 10 gigawatts will require sustained investment, real estate at scale, and reliable access to chips from Nvidia and others. It also rests on the assumption that AI's appetite for compute will not cool. For SoftBank, a company whose history runs from early Alibaba bets to the turbulent Vision Fund, SB Neo is a familiar kind of wager: large, infrastructure-driven, and timed to a moment when demand is real but the competitive landscape is still being written.
SoftBank Group is making a decisive move into the artificial intelligence infrastructure business. This month, the Japanese conglomerate and its telecom subsidiary are launching a new company called SB Neo, designed to rent computing power to American firms hungry for the specialized hardware that trains and runs large language models. The venture represents a significant bet that the global shortage of AI compute capacity will only deepen—and that there's substantial profit in meeting that demand.
The two companies announced the plan on Thursday, positioning SB Neo as a direct competitor to established players like CoreWeave and Nebius Group, which have already carved out market share in this rapidly expanding sector. SB Neo will offer AI chips and cloud services tailored specifically for hyperscalers—the massive technology companies that need enormous amounts of computing power to develop and deploy their own AI systems. The service launches next fiscal year, giving SoftBank time to prepare its infrastructure.
What distinguishes this venture is its scale ambition. Junichi Miyakawa, who leads SoftBank's telecom operations, outlined plans to build data center capacity reaching 10 gigawatts by approximately 2030. That figure is not casual. A gigawatt is enough electricity to power roughly 750,000 homes. Ten gigawatts represents an extraordinary concentration of computing infrastructure, the kind of investment that signals deep confidence in sustained demand for AI resources over the next several years.
The timing reflects a broader industry reality. Companies developing advanced AI models face a genuine bottleneck: there simply aren't enough specialized chips and data center slots available to meet current demand. CoreWeave, a relative newcomer to the space, has already attracted billions in investment and commands premium pricing for access to its infrastructure. Nebius Group, similarly positioned, has grown rapidly by offering compute capacity when the market is starved for it. SoftBank's entry, backed by the financial resources of one of Japan's largest corporations, signals that this market is moving from niche opportunity to mainstream infrastructure play.
SoftBank's existing telecom network and data center footprint give the company advantages that pure-play startups lack. The group operates significant infrastructure across multiple continents and has the capital to build new facilities at scale. By combining its telecom assets with purpose-built AI infrastructure, SB Neo can offer customers integrated services—connectivity, compute, and storage—rather than forcing them to stitch together solutions from multiple vendors.
The 10-gigawatt target by 2030 is ambitious but not unprecedented. It will require sustained capital investment, real estate acquisition, and the ability to source enough specialized chips—primarily from Nvidia and other manufacturers—to populate those data centers. It also assumes that demand for AI compute will remain robust for the next four years, a reasonable but not guaranteed assumption given how quickly the technology landscape shifts.
For SoftBank, this venture represents a strategic pivot. The company has long been known for bold, sometimes controversial technology bets—from its early investments in Alibaba to its massive Vision Fund. SB Neo fits that pattern: it's a large-scale infrastructure play in a sector where demand is real and immediate, but where competition is intensifying. The company is essentially betting that it can build data center capacity faster and more efficiently than competitors, and that hyperscalers will value the integrated services SoftBank can provide. Whether that bet pays off will become clear over the next several years as SB Neo ramps up operations and the broader AI infrastructure market matures.
Citas Notables
Plans to ramp up computing resources to supply data center capacity at a scale of 10 gigawatts by around 2030 for large-scale AI model training and inference— Junichi Miyakawa, head of SoftBank's telecom unit
La Conversación del Hearth Otra perspectiva de la historia
Why does SoftBank think it can compete with CoreWeave and Nebius Group? Those companies are already established in this space.
SoftBank has advantages those startups don't—existing data center infrastructure, a global telecom network, and access to massive capital. They're not starting from zero.
But 10 gigawatts by 2030 is a huge commitment. What if demand for AI compute doesn't grow as fast as they're betting?
That's the real risk. If the market cools or if chip supply becomes less constrained, they've built capacity nobody needs. But right now, the shortage is acute enough that they're probably confident.
Who exactly are their customers going to be?
Hyperscalers—the big tech companies building their own AI models. Companies like Meta, Microsoft, Google. They need enormous amounts of compute and are willing to pay premium prices for reliable access.
Is this just about renting out computing power, or is there something more strategic here?
It's both. Yes, they're renting compute. But by building integrated infrastructure—chips, cloud services, connectivity—they're trying to become essential to how AI companies operate, not just a commodity supplier.
What happens if they succeed?
SoftBank becomes a major player in AI infrastructure globally, not just a telecom company. That's a significant shift in what the company is.