In September 2025, the technology world was intrigued by the news that NVIDIA and OpenAI jointly announced a long-planned collaboration through a letter of intent under which NVIDIA could invest up to 100 billion dollars in OpenAI’s artificial intelligence infrastructure, including the construction of huge data centers with a capacity of at least 10 gigawatts of power, which would require resources comparable to the electricity of dozens of nuclear reactors.
However, until the beginning of February 2026, that monumental agreement practically did not come to life. The previous announcement never turned into a firm, binding contract, and NVIDIA CEO Jensen Huang said in February that the $100 billion figure was “never a commitment” but just a maximum that both companies were targeting as a framework for potential investment.
According to reports, the negotiations never resulted in concrete agreements on the investment of the entire announced sum, so the market actors were also surprised, even the price of NVIDIA shares fell slightly after it became clear that the agreement from September was not concluded.
A partial reason for this development is internal reservations about the scope and structure of the investment, but also the strategic positioning of OpenAI itself. Reuters reported that multiple sources indicated that OpenAI was looking for alternative hardware solutions in addition to NVIDIA GPUs, as some within its infrastructure were dissatisfied with the speed of NVIDIA’s so-called chips. inference tasks, the phase when the model generates responses to users.
While fears about the collapse of cooperation reached the media, OpenAI’s CEO Sam Altman stated on the X social network that OpenAI still “loves to work with NVIDIA” and that the company remains one of their key partners, but also that a new round of financing in which NVIDIA would participate, although not necessarily with the full 100 billion USD, is being negotiated.
Meanwhile, OpenAI has entered into other partnerships, including a $10 billion investment with Cerebras Systems, intended to provide computing power for inference until 2028, as well as agreements with AMD and Broadcom to diversify its approach to hardware and reduce dependence on a single company.
These developments indicate that big announcements in the AI industry, especially those with astronomical figures, do not automatically translate into contractual obligations, but sometimes serve as a signal to the market or a starting point for negotiations that are later redefined. In this case, the announced “mega-deal” remained a non-binding intention without final realization in the intended scope, although both companies are still planning serious investments in cooperation which, as they say, will continue to play an important role in the construction of AI infrastructure, reports Ars Technica.