SoftBank in talks to invest up to US $25 billion in OpenAI

JAPAN – SoftBank is reportedly in discussions to invest up to US $25 billion in OpenAI, which could make the Japanese conglomerate the largest investor in the company, surpassing Microsoft.

This news, shared by the Financial Times on January 29, follows an announcement that OpenAI and SoftBank are joining Oracle on an AI data center project called “Stargate,” which could cost as much as US $500 billion over the next four years.

Sources close to the matter say that SoftBank is in talks to invest between US $15 billion and US $25 billion in OpenAI, in addition to the over US $15 billion it has already committed to Stargate.

If the deal goes through, SoftBank’s total investment in OpenAI and its related projects could exceed US $40 billion.

The company has previously invested US $500 million in OpenAI and has plans to invest US $100 billion in AI infrastructure in the U.S. over the next four years.

However, the investment amount is not yet final. One source noted that “the talks are ongoing and the amount that SoftBank could invest in primary equity into OpenAI is a moving target.”

OpenAI also plans to invest about US $15 billion in the Stargate project, with some sources indicating that SoftBank’s equity investment in OpenAI might cover this amount.

AI data centers, like the one being developed for Stargate, are unique in that they require specialized hardware and infrastructure to support the heavy computational demands of AI workloads.

Deborah Perry Piscione, co-founder of the AI/Web3 advisory firm Work3 Institute, explained that “traditional data centers focus on storage and basic compute, while AI facilities need dense configurations of GPUs and AI accelerators, like Nvidia’s H100s, designed specifically for the complex matrix calculations that power AI models.”

At the same time, there is ongoing debate in the AI community about the level of investment needed for AI projects.

Chinese company DeepSeek recently introduced a model that reportedly trained on just $5.58 million using 2,048 Nvidia H800 chips.

Roy Benesh, CTO at eSIMple, pointed out that this development shows how “small companies, individual developers, and even researchers are now able to harness the power of AI without breaking the bank.”

However, analysts at Bank of America have argued that the US $5.58 million figure is misleading, as it doesn’t account for the costs of research, experiments, architectures, algorithms, and data.

Despite this, they acknowledged that DeepSeek’s model demonstrated that less costly AI training might be possible in the future.