SoftBank is reportedly in discussions to invest up to $25 billion in OpenAI, which would make it the startup’s largest backer.
In November last year, OpenAI allowed employees to sell $1.5 billion in shares to SoftBank, as per reports. A CNBC report stated that SoftBank founder Masayoshi Son has been pushing for a larger stake in OpenAI after having previously invested in The Stargate Project – a joint venture with OpenAI.
“$40 billion in equity plus MGX? Oracle? Then using the equity to raise debt gets you to $100 billion of dry powder for Stargate’s first year. Meta is gunning for 1.3 million GPUs by year-end, and people are seriously talking about an H20 downtick?” Tae Kim, author of the book ‘The NVIDIA Way’, said.
The $500 billion AI infrastructure initiative is backed by tech titans such as Oracle CTO Larry Ellison, SoftBank CEO and The Stargate Project chairman Masayoshi Son, and OpenAI CEO Sam Altman. In addition to this, US President Trump secured $3 trillion in investments – a move aimed at shifting the balance of technological dominance away from China and into the US.
OpenAI is also shifting to a for-profit model, allowing for greater investment.
The company, valued at $157 billion, raised $6.6 billion in October last year to compete with xAI, Microsoft, Google, Amazon, and Anthropic.
Rising Competition from China
Chinese AI startup DeepSeek has gained traction in the US, topping Apple’s App Store and Google Play Store rankings. Its technical paper claims the AI model was built at a fraction of its US rivals. However, experts dispute this.
A recent voice in the debate, Anthropic CEO Dario Amodei, stated in his blog ‘On DeepSeek and Export Controls’ that Claude 3.5 Sonnet was trained for tens of millions of dollars and developed 9–12 months before DeepSeek’s model, disputing claims about DeepSeek’s cost efficiency.
Amodei noted that DeepSeek’s model aligns with the US-based models from 7–10 months earlier and was trained at a lower cost but “not anywhere near the ratios people have suggested”. He attributed the cost reduction to the expected annual decrease in AI training costs by four times rather than a technological breakthrough.
