Son has slowed most other dealmaking at SoftBank’s Vision Fund to a crawl, and any deal above $50 million now requires his explicit approval, two of the sources told Reuters. Son’s firm is working to take public its payments app operator, PayPay. The initial public offering, originally expected this month, was pushed back due to the 43-day-long U.S. government shutdown, which ended in November. PayPay’s market debut, likely to raise more than $20 billion, is now expected in the first quarter of next year, according to one direct source and another person familiar with the efforts.
The Japanese conglomerate is also looking to cash out some of its holdings in Didi Global, the operator of China’s dominant ride-hailing platform, which is looking to list its shares in Hong Kong after a regulatory crackdown forced it to delist in the U.S. in 2021, a source with direct knowledge said. Investment managers at SoftBank’s Vision Fund are being directed toward the OpenAI deal, two of the above sources said.
SoftBank’s scramble to marshal funds offers a window into the strain faced even by the world’s biggest dealmakers as they scramble to finance ambitious AI data center projects worth hundreds of billions of dollars.
SoftBank declined to comment.
SoftBank has options
OpenAI has not yet received the remaining funding, but expects the money to come in by the end of 2025, as stipulated in the contract, sources said.
SoftBank has multiple sources of capital it could tap, including margin loans, cash on its balance sheet, stakes in listed companies, and corporate bonds or bridge loans, sources said. Son has strong reasons to draw on a range of funding mechanisms to fulfill those obligations. SoftBank secured a deal to invest in OpenAI at a $300 billion valuation in April. Since then, the valuation of OpenAI has risen dramatically and the company is in talks to raise additional funding from investors, including Amazon, tripling its valuation to close to $900 billion, one of the sources added, which would give SoftBank a significant paper gain once the transaction is completed.
A major pool of capital for SoftBank is its undrawn capacity of margin loans borrowed against its ownership of British semiconductor and software design company Arm Holdings. SoftBank recently expanded its margin loan capacity by $6.5 billion, bringing the total undrawn capacity to $11.5 billion. Arm’s stock has since tripled from its IPO price, providing SoftBank with additional collateral headroom to expand its borrowing capacity.
SoftBank reported parent-level cash of 4.2 trillion yen ($27.16 billion) as of September 30. The group still owns about 4% of T-Mobile US, remaining the wireless carrier’s second-largest shareholder, a stake worth roughly $11 billion at the end of September, according to LSEG data. Despite investing at a less active pace, it has continued to back AI startups such as Sierra and Skild AI.
OpenAI needs the money
Both OpenAI and SoftBank are investors in Stargate, a $500 billion initiative to build AI data centers for training and inference that executives say is crucial to the U.S. government’s ambitions to keep ahead of China in AI.
The rush to build data centres has also prompted tech giants including Meta Platforms to commit unprecedented sums to these buildouts – which need chips, power, cooling, and servers – and they have brought in deep-pocketed partners to spread the risk. Their hefty capital outlays have sparked concerns about what happens if the investments fail to bring commensurate returns, raising the specter of an “AI bubble” bursting. SoftBank promised in April to invest up to $30 billion in OpenAI – $10 billion of which the startup would receive the same month. The rest of the payment was contingent on the AI startup transitioning to a for-profit corporation by the end of the year, an ambitious feat that OpenAI achieved in October.
The new funding is crucial for covering OpenAI’s rising costs to train and run its AI models as competition from Alphabet’s Google ratchets up. OpenAI CEO Sam Altman told employees recently that the company is now entering a “code red” phase to improve ChatGPT – delaying other product rollouts to fend off the momentum behind Google’s Gemini.
In October, Altman said OpenAI aimed to build 30 gigawatts of computing capacity for $1.4 trillion. He said he ultimately wants OpenAI to add 1 gigawatt of compute every week – an enormous target given that each gigawatt currently comes with a capital cost of more than $40 billion.
