SoftBank stakes $4B on securing AI data center power and capacity — DigitalBridge purchase indicative of AI industry's increasing investments in energy supply
SoftBank has agreed to acquire DigitalBridge in a deal valuing the digital infrastructure investor at roughly $4 billion, including debt. The transaction, if it closes, would take DigitalBridge private at $16 per share in cash.
SoftBank says the acquisition is intended to accelerate its global AI infrastructure strategy by giving it direct access to data center development pipelines, financing mechanisms, and power-constrained real estate as AI compute demand collides with physical limits. SoftBank’s founder, Masayoshi Son, has been repositioning the company around AI after years of mixed results from its Vision Fund strategy.
A power and financing platform
DigitalBridge, a specialist investor and asset manager, does not operate data centers in the way hyperscalers do. It functions as an infrastructure investment and management platform that raises capital, acquires or develops assets, and places operational teams around them. According to its most recent public disclosures, DigitalBridge manages more than $100 billion in digital infrastructure assets across multiple portfolios. Within its data center holdings, the company says it has a "power bank" of around 22 gigawatts across land it owns, facilities already operating, and projects under development.
Modern AI data centers are increasingly defined by how quickly developers can secure grid connections and deliver reliable power at densities far above historical norms. DigitalBridge has been clear that its competitive advantage lies in sourcing entitled land with power access, financing long-duration infrastructure, and then pairing those assets with operators or hyperscalers. Its portfolio includes well-known data center platforms such as Switch, Vantage Data Centers, DataBank, AtlasEdge, Yondr, and AIMS, alongside Takanock, a vehicle created specifically to assemble powered land for future data center builds.
This solves a problem for SoftBank that money alone cannot fix. AI compute buildouts are increasingly bottlenecked by physical infrastructure rather than silicon supply. Even when accelerators are available, bringing a new AI facility online can take years due to grid interconnection queues, local permitting, and the need to design cooling and power delivery systems that can sustain extremely high continuous loads. DigitalBridge’s model is designed to compress those timelines by doing the slow work in advance and spreading capital risk across long-term infrastructure funds rather than a single corporate balance sheet.
Why AI data centers are different
Traditional enterprise and cloud data centers were built around rack densities measured in single-digit or low double-digit kilowatts, but AI systems have surpassed that by a significant margin. Training clusters built around modern accelerator pods routinely target rack densities that approach or exceed 100 kilowatts, forcing fundamental changes in power distribution and cooling architecture.
These requirements ripple through the entire facility, with high-density AI racks demanding liquid cooling, often with direct-to-chip cold plates and increasingly with rear-door heat exchangers or full immersion in some designs. Power distribution moves away from conventional raised-floor layouts toward busways, higher-voltage feeds, and redundant substations sized for continuous peak load. Water availability, heat rejection, and serviceability become first-order constraints rather than secondary design considerations.
This is where DigitalBridge’s emphasis on megawatts and entitled land comes in. A parcel of land capable of supporting 50-100 megawatts of reliable power, with room for substations and cooling infrastructure, is far more valuable for AI than a conventional colocation shell that must be retrofitted. DigitalBridge underscores this point by valuing capacity in megawatts rather than square footage, reflecting how AI economics increasingly scale with power rather than floor space.
While accelerators from vendors such as Nvidia dominate headlines, the surrounding infrastructure can represent an equal or greater share of capital expenditure at scale. Liquid cooling systems, power delivery equipment, switchgear, and grid upgrades all add to the bill of materials, so securing sites where those systems can be deployed quickly is becoming a priority for anyone deploying AI at scale.
SoftBank has big ambitions
SoftBank has hardly been coy about its intention to sit at the center of the next AI buildout, both through equity stakes and infrastructure. The company has backed or partnered with major AI developers, including OpenAI, and has discussed large-scale compute initiatives that would require multiple gigawatts of capacity over the coming decade. Acquiring DigitalBridge gives SoftBank a way to influence how and where that capacity comes online without turning itself into a hyperscale data center operator.
Rather than single companies funding multibillion-dollar data center campuses outright, much of the new capacity is being developed through partnerships between asset managers, infrastructure funds, and tech firms. DigitalBridge’s model fits into that ecosystem, allowing SoftBank to co-invest and align infrastructure development with the needs of its AI portfolio companies.
What isn’t so clear is how directly SoftBank intends to steer DigitalBridge’s assets toward its own AI projects. The firm could use the platform primarily as a capacity provider for partners, or pursue tighter integration, aligning specific developments with known AI workloads and accelerator roadmaps. There is also the question of geography. DigitalBridge’s portfolio spans North America, Europe, and parts of Asia, giving SoftBank optionality in where it anchors future AI clusters based on considerations such as regulatory environments, access to power, and connectivity.
In real terms, AI at scale is constrained just as much by civil engineering and electrical infrastructure as it is by semiconductor manufacturing. SoftBank’s decision to acquire DigitalBridge reflects a recognition that controlling the pipeline of land and megawatt potential might be just as important as securing access to next-gen silicon.
Alphabet's Intersect deal makes the same point from the hyperscaler perspective: power now plays a big role in dictating deployment schedules, so owning or controlling energy developments is naturally becoming part of broader compute strategies.
