NVIDIA is opening the server design specifications for its next-generation AI platform, Vera Rubin, to outside manufacturers, and has released a CPU rack system built specifically for AI agents, aiming to embed its own technical standards across the entire AI data center ecosystem and further cement its industry influence beyond the GPU market.
The disclosure took place on the 13th at the Open Compute Project (OCP) Global Summit in San Jose, California. NVIDIA Vice President Ian Buck announced during his speech that the company will open the server rack and compute tray design blueprints for the Vera Rubin platform to external partners, covering common specifications for power delivery, liquid cooling, and network connectivity. NVIDIA said the move will lower the development threshold for server manufacturers and shorten the construction timeline and cost of data centers.
Industry analysts believe there is a clear strategic intent behind the move 鈥?as the era of AI agents arrives, infrastructure demand is extending from GPUs to CPUs and networking equipment, and NVIDIA is using the openness of its design specifications to build an AI infrastructure ecosystem with itself at the center ahead of rivals.
800-Volt DC Solution Tackles Rack Power Bottleneck
On the power architecture side, NVIDIA proposes raising rack DC voltage to 800 volts in its new design. Under the blueprint, conversion equipment inside the data center converts incoming alternating current into 800-volt direct current, which is then delivered to the server rack inlet.
Previously, data center racks typically used DC voltages of 48 to 54 volts. However, once single-rack power consumption climbs to around 1 megawatt, the current intensity at low voltage becomes excessively high, turning into a physical bottleneck that restricts expansion. The high-voltage 800-volt DC approach can effectively alleviate this problem and is seen as a key technological path for AI data centers to cope with continuously rising energy density.
Dedicated CPU Rack Targets AI Agent Workloads
NVIDIA also released a dedicated rack design specification based on its in-house Vera CPU. The existing Vera Rubin NVL72 system combines 72 GPUs with 36 CPUs and is mainly aimed at large-scale computing tasks; the newly released Vera CPU rack can carry up to 256 Vera CPUs, positioning it for scenarios involving concurrent operation and management of multiple programs, with a focus on pure CPU processing capability.
Joe DeLaere, who heads AI infrastructure solutions at NVIDIA, said the real bottleneck in the industry today is whether AI factories can be built and deployed fast enough. Opening the design specifications is interpreted by outsiders as NVIDIA making a proactive move to accelerate AI data center construction, and as a strategic step to lock in a voice early during the period when industry standards are still forming.
By pushing data center equipment vendors to widely adopt its technology and design specifications, NVIDIA is seeking to extend its influence from GPUs to a broader range of AI infrastructure layers.
Hundred-Billion-Dollar Market Accelerates Expansion
The scale outlook for the data center equipment market also provides ample competitive motivation for this standards battle. Research firm IDC estimates that spending on equipment compliant with OCP technical standards will reach about $800 billion by 2029, roughly 2.7 times the $295 billion it forecast for the same period a year earlier. Related spending will rise further to $892 billion by 2030, about 5.8 times actual spending of $154 billion in 2025.
In this rapidly expanding market, whoever's design specifications become the industry benchmark will gain a first-mover advantage in the supply chain and ecosystem. NVIDIA's latest move is a direct response to this logic.
Comments