Nvidia has opened the server designs for its next-generation artificial intelligence platform, Vera Rubin, to outside manufacturers and released specifications for a dedicated central processing unit system tailored to AI agents. The move gives data center equipment makers a framework for building hardware that conforms to Nvidia's designs. Analysts read it as an attempt by the company, which has dominated the graphics processing unit market, to extend its influence across the data center as a whole.
Nvidia said on the 13th that it disclosed the server rack and compute tray designs for Vera Rubin at the Open Compute Project Global Summit in San Jose, California.
Ian Buck, a vice president at Nvidia, said in a presentation that the company would provide outside parties with designs for the rack, the structure that houses multiple servers and related equipment, and for the compute tray, the computing unit installed inside it. The blueprints he presented include common specifications covering power delivery, liquid cooling and network connections. Nvidia said the move will make it easier for server manufacturers to produce related equipment and will cut the time and cost of building data centers.
To address power, the new bottleneck in AI data centers, Nvidia proposed in the designs a direct-current system that raises voltage to 800 volts. Under the blueprints, a conversion device inside the data center takes incoming alternating current from outside, converts it to 800-volt direct current and delivers it to the entrance of the server rack. Data centers have previously used 48 to 54 volts of direct current inside racks, but once rack power rises to around 1 megawatt, current becomes excessive at voltages in the 50-volt range. High-voltage 800-volt direct current offers an alternative that eases the problem.
Nvidia also released design specifications for a dedicated rack built entirely around its in-house Vera CPU. While the existing Vera Rubin NVL72 combines 72 GPUs with 36 CPUs to handle large-scale computation, the Vera CPU rack carries up to 256 Vera CPUs and is used for running and managing multiple programs at once. It is designed to concentrate on CPU-based processing.
The decision to open the designs is seen as a step to accelerate the build-out of AI data centers. Joe DeLaere, who leads AI infrastructure solutions at Nvidia, said the industry's real bottleneck now depends on how quickly the world can manufacture and deploy AI factories.
Industry watchers see a strategic calculation behind the move as well: responding to shifts in the AI industry while expanding an ecosystem centered on Nvidia. The spread of agentic AI is expected to broaden infrastructure demand beyond GPUs to CPUs and network equipment. AI agents draw on CPUs to execute programs and process data as they carry out actual tasks, such as calling external programs and coordinating multiple jobs. By encouraging data center equipment makers to adopt its technology and design specifications more widely in line with that shift, Nvidia is seen as seeking to extend its influence across the AI infrastructure market.
The market for data center equipment is also projected to expand rapidly. Research firm IDC estimates that spending on equipment built to technical standards recognized by the OCP will reach about $800 billion in 2029. That is roughly 2.7 times the $295 billion the firm forecast for the same year a year earlier. Related spending will reach $892 billion in 2030, about 5.8 times the $154 billion recorded in 2025, IDC said.