Yes, it looks like cracks are starting to appear in the relationship between Nvidia and OpenAI. Two AI giants that until recently were seen as a model partnership are now beginning to keep their distance. What happened? And is the $100 billion deal really in danger?
THE $100 BILLION DEAL
Let’s take things from the beginning. In September 2025, Nvidia announced an enormous investment intention: up to $100 billion for OpenAI, aimed at building and supporting at least 10 gigawatts of computing power in AI data centers based on Nvidia infrastructure.
source
The agreement envisioned OpenAI leasing the chips and systems, with the first phase of 1 GW expected to go live in mid-2026. To put the scale into perspective, 10 GW is astronomical. We are talking about dozens of ultra modern data centers.
Investors were thrilled. AI stocks surged. Everything pointed to a future where OpenAI and Nvidia would build the next era of artificial intelligence together.
Or at least that was the assumption.
THE FIRST CLOUDS
A few months after the announcement, rumors began to surface that something was not right. According to reports from the Wall Street Journal and Bloomberg, Nvidia executives started expressing serious doubts about the viability of the deal.
Jensen Huang, Nvidia’s powerful CEO, reportedly stated that there was never a real commitment and that the company views any investment one step at a time.
Additionally, Huang allegedly commented privately that OpenAI has issues with strategic discipline and that competition is intensifying, especially from players like Google, Amazon, and the Anthropic ecosystem backed by them.
OPENAI STRIKES BACK
The response from OpenAI came quickly. CEO Sam Altman posted on X: “We love working with Nvidia. They make the best AI chips in the world. We hope to remain their customer for many years.”
Sachin Katti, head of infrastructure, went even further: “Our relationship with Nvidia is foundational. This is not a simple supplier relationship. It is a deep, ongoing co production partnership.”
At the same time, however, OpenAI has been actively searching for alternatives for specific workloads, particularly inference, the stage where AI models respond in real time.
It has signed agreements with AMD, Cerebras, and Groq, companies that design inference focused chips based on SRAM, high speed on chip memory that significantly accelerates real time applications such as Codex, which generates code on the fly.
According to reports, OpenAI considers Nvidia’s performance in this area insufficient for some of its applications. That said, the company admits that for model training it remains fully dependent on Nvidia chips.
JENSEN’S VIEW
Jensen Huang, while publicly remaining polite, is showing signs of distancing. In public statements, he clarified: “It was simply an invitation to invest. It was never a commitment. We will evaluate it step by step.”
When asked whether it was true that he was unhappy with OpenAI, he replied: “That’s nonsense. We will invest heavily, possibly the largest investment we have ever made.”
Yet shortly afterward, Nvidia signed a $20 billion licensing deal with Groq, a startup that had been in talks with OpenAI, effectively blocking cooperation between the two. Many analysts described this move as a clear counterattack by Nvidia.
INVESTMENT IMPLICATIONS
As you can imagine, all of this has created uncertainty in the investment world. Why?
Because OpenAI has signed major agreements with companies backed by large institutional investors, and those companies rely on Nvidia infrastructure. If this chain breaks, the entire AI narrative loses part of its momentum, at least in the short term.
