Last year, Nvidia made an unusual proposal to Amazon Web Services and other cloud providers that have long been the biggest buyers of Nvidia’s specialized artificial intelligence server chips. Nvidia wanted to lease Nvidia-powered servers in the cloud providers’ data centers so it could turn around and rent the same servers to AI software developers. Those developers included some of the biggest cloud customers in the world.
As the discussions progressed, Nvidia’s leverage increased. Demand for Nvidia-powered servers exploded among AI software developers following the launch of OpenAI’s ChatGPT in November, and the cloud providers soon couldn’t keep up. In that delicate moment, Nvidia saw a way to essentially compete with the cloud providers for customers. Nvidia’s trump card? It was about to release a much anticipated new AI chip, the H100, which the traditional cloud providers needed.
Microsoft, Google and Oracle agreed to Nvidia’s proposal but AWS did not, according to a person with direct knowledge of the decision.