At the OCP Summit, NVIDIA announced the open-sourcing of the design specifications for its next-generation Vera Rubin platform servers and unveiled a dedicated rack system equipped with up to 256 self-developed CPUs. The company has introduced an 800-volt high-voltage DC solution to overcome the power consumption bottleneck of server racks, aiming to establish an industry benchmark early in the AI agent era and secure a leading position in the hundred-billion-dollar data center market.
$NVIDIA (NVDA.US)$It has opened the server design specifications of its next-generation AI platform, Vera Rubin, to external vendors and unveiled a CPU rack system specifically engineered for AI agents, aiming to embed its own technical standards across the entire AI data center ecosystem and further strengthen its industry influence beyond the GPU market.
This disclosure took place at the Open Compute Project (OCP) Global Summit held in San Jose, California, on the 13th. In his speech, NVIDIA Vice President Ian Buck announced that the company will open up the design blueprints for server racks and compute trays of the Vera Rubin platform to external partners, covering universal specifications for power supply, liquid cooling, and network connectivity. NVIDIA stated that this move will lower the R&D barriers for server manufacturers and shorten the construction timeline and costs of data centers.
Industry analysts believe this move reflects a clear strategic intent: as the era of AI agents unfolds, infrastructure demand is shifting from GPUs to CPUs and networking equipment, and NVIDIA is leveraging the openness of its design specifications to proactively build an AI infrastructure ecosystem centered around itself.
800V DC solution addresses the rack-level power consumption bottleneck.
In terms of the power supply architecture,$NVIDIA (NVDA.US)$The new design proposal suggests increasing the rack‑level DC voltage to 800 volts. Under this blueprint, an in‑rack converter will transform the external AC input into 800‑V DC before delivering it to the server rack's power inlet.
Previously, data center racks typically operated at a DC voltage of 48 to 54 volts. However, as the power consumption of a single rack rises to around 1 megawatt, the current required at such low voltages becomes excessively high, creating a physical bottleneck that limits scalability. A high‑voltage 800‑volt DC solution can effectively address this issue and is regarded as a key technological pathway for AI data centers to cope with the ongoing increase in energy density.
Dedicated CPU rack optimized for AI agent workloads
$NVIDIA (NVDA.US)$A dedicated rack design specification based on the in-house Vera CPU was also unveiled. The existing Vera Rubin NVL72 system integrates 72 GPUs with 36 CPUs, primarily targeting large-scale computing workloads. Meanwhile, the newly released Vera CPU rack can accommodate up to 256 Vera CPUs, designed for scenarios involving concurrent execution and management of multiple applications, with an emphasis on pure CPU‑centric processing capabilities.
Joe DeLaere, who oversees AI infrastructure solutions at NVIDIA, stated that the industry's real bottleneck today lies in the ability to build and deploy AI factories quickly enough. The release of open design specifications has been interpreted as NVIDIA's proactive move to accelerate the construction of AI data centers, as well as a strategic step to secure influence ahead of the establishment of industry standards.
By encouraging data center equipment vendors to widely adopt its technologies and design specifications, NVIDIA is seeking to extend its influence from GPUs to the broader AI infrastructure landscape.
The hundred-billion-yuan market is accelerating its expansion.
The size and outlook of the data center equipment market also provide strong competitive momentum in this standards battle. Research firm IDC projects that spending on equipment compliant with OCP technical standards will reach approximately $800 billion by 2029—about 2.7 times the $295 billion forecast it issued at the same time a year earlier. By 2030, related expenditures are expected to climb further to $892 billion, roughly 5.8 times the $154 billion actually spent in 2025.
In this rapidly expanding market, whoever sets the industry's design standards will gain a first-mover advantage in the supply chain and ecosystem. NVIDIA's latest move is a direct response to this logic.
Edited by melody