SoftBank borrows $20 billion to double down on $45 billion OpenAI gain

SoftBank Group disclosed a $45 billion unrealized gain on its OpenAI holdings through March and has taken on $20 billion in new debt to deepen its investment, a move that increases its financial risk as the artificial intelligence leader pivots from research to enterprise deployment.

S&P Global Ratings revised SoftBank's credit outlook to negative after its last funding round, judging that the "huge investment in OpenAI" would likely deteriorate the quality of its portfolio and its financial capacity.

The Japanese conglomerate revealed the $20 billion in new borrowing occurred in April, with $2.5 billion already repaid, according to a filing. This follows a separate $40 billion bridge loan secured in March to fund its commitments to the ChatGPT creator. However, reports last week indicated the company had to downsize an additional margin loan backed by its OpenAI stake after facing hesitation from some creditors, a sign of growing concern over its leverage.

The new capital is set to fund OpenAI’s aggressive push into corporate services, a strategy crystallized by its acquisition of consulting firm Tomoro. The deal makes Tomoro the foundation of a new $14 billion subsidiary, the OpenAI Deployment Company, which aims to embed engineers within enterprises to build AI solutions, directly competing with firms like Accenture and Cognizant.

The Services Pivot

OpenAI’s acquisition of Edinburgh-based Tomoro signals a strategic shift from simply building foundation models to installing them. The new Deployment Company, launched with $4 billion in initial capital from a syndicate including TPG, SoftBank, and Bain Capital, will copy the “forward-deployed engineer” model pioneered by Palantir Technologies. This model places engineers directly inside client organizations to build custom, production-grade systems.

The move is a response to a market-wide bottleneck where enterprise AI adoption has been slowed by the complexities of integration, security, and change management. Other AI labs are making similar moves. Anthropic has formed a $1.5 billion joint venture with Blackstone and Goldman Sachs to act as its deployment arm, while Google has committed $750 million to fund partners deploying its AI. The industry is concluding that the most durable value lies not in selling access to models, but in providing the services to make them work.

A Full-Stack AI Bet

For SoftBank, the massive OpenAI stake is the centerpiece of a broader strategy to own key parts of the AI value chain. The company has been assembling a portfolio of hardware and infrastructure assets, including a majority stake in chip designer Arm, the 2024 acquisition of AI chip developer Graphcore, and the 2025 purchase of silicon design firm Ampere Computing. SoftBank recently injected an additional $457 million into Graphcore, according to company filings.

This full-stack approach—spanning from chip design with Arm and Graphcore to enterprise deployment with OpenAI’s new subsidiary—shows a conviction that vertical integration is necessary to win the AI race. By funding OpenAI’s expansion into services, SoftBank is betting it can build a moat that competitors cannot erode simply by creating a better model. The risk is the immense capital required and the debt load it places on SoftBank’s balance sheet, drawing parallels to its previous concentrated bets.

This article is for informational purposes only and does not constitute investment advice.