Last September, OpenAI and Nvidia announced a US$100 billion (HK$780 billion) partnership, but the deal is now “frozen.” Under the original proposal, Nvidia would commit up to US$100 billion to OpenAI in the form of compute resources, including 10 gigawatts of capacity powered by next-generation GPUs such as the Vera Rubin series. In return, Nvidia would receive significant equity in OpenAI.
The tricky part is that the structure was almost circular. Most of the invested capital would flow back to Nvidia through GPU purchases. Revenue swapping, which was common during the Internet bubble, would likely fail modern accounting compliance tests. However, these issues could be circumvented if OpenAI receives investment capital that is then used to purchase GPUs, with Oracle acting as the infrastructure builder in between, could circumvent those issues.
But why did the deal stall? Reports indicate that Nvidia executives were uneasy about the deal’s mega-size, equivalent to several years of Nvidia’s revenue. Recent figures from Microsoft’s quarterly reports suggest that OpenAI incurred quarterly losses approaching US$12 billion, with no clear path to profitability. Nvidia chief executive Jensen Huang reportedly described OpenAI’s spending habits as lacking discipline in private conversations. Though Huang publicly dismissed those reports as nonsense, he did confirm that Nvidia’s participation in OpenAI’s funding round would be substantially smaller than the original US$100 billion, with the structure likely shifting toward equity participation rather than a direct infrastructure commitment.
A similar dynamic played out last year when Apple entered talks to invest in OpenAI. Both Apple and Nvidia appear to have uncovered red flags during the due diligence stage, prompting Apple to withdraw and Nvidia to scale back dramatically rather than proceed with their original commitments.
This is not an isolated case. Oracle’s US$300 billion agreement with OpenAI and CoreWeave’s US$6.3 billion deal with Nvidia, which have reportedly gone underwater and been scaled down, respectively, suggest that all these multi-billion-dollar ambitions may now be colliding with financial reality.
As OpenAI eyes a potential 2026 IPO, possibly targeting valuations north of US$800 billion, the paused Nvidia deal underscores a truth: hype can propel industries forward, but bubbles form when promises outrun fundamentals.
