Nvidia may have just built its own hyperscaler without owning a single data center

Nvidia has quietly assembled the components of a cloud giant, a theory suggesting the company is building its own hyperscaler without owning any data centers. This development was highlighted by tech analyst Ryan Shrout, who pointed to a post by Clark Tang, a partner at Altimeter Capital. According to Tang, Nvidia is creating a synthetic hyperscaler by leveraging its expertise in AI training and inference, and its proprietary hardware.

The financing arrangement signed by Nvidia with BlackRock, Goldman Sachs, and other Wall Street heavyweights is the last piece of the puzzle. This $500 billion capital infusion will be used to build out Nvidia's hyperscaler capabilities. Traditionally, cloud giants like Amazon AWS, Microsoft Azure, and Google Cloud have earned their 35% to 40% operating margin by commoditizing hardware and optimizing software.

However, the hyperscalers' interest in integrating Nvidia's hardware, which carries much higher gross margins, created an opportunity for a new breed of companies.

These neoclouds, such as CoreWeave, Nebius, Lambda, Crusoe, and Nscale, have developed software specifically for Nvidia's hardware, eliminating the need for virtualization overlays and operating at roughly 20% margins. This allows them to win workloads without the heavy investment required by the established cloud giants. Nvidia's hardware and software playbooks, such as DSX OS, Mission Control, and Dynamo, provide the necessary operational infrastructure for these neoclouds to build data centers and serve inference demands.

In summary, Nvidia's synthetic hyperscaler is a combination of its proprietary hardware and software capabilities, now made available through third-party financing. This innovative approach could disrupt the traditional cloud computing landscape and reshape the AI infrastructure market.