NVIDIA announced the opening of the design specifications for the next generation of Vera Rubin platform servers at the OCP summit, and released a dedicated rack system equipped with up to 256 self-developed CPU. The company introduced an 800-volt high-voltage direct current solution to break through the bottleneck of rack power consumption, aiming to establish an ecological benchmark in the era of AI intelligent entities and secure a dominant position in the market for data centers with hundreds of billions of units.

NVIDIA ($NVDA.US) has opened up the server design specifications of its next-generation (AI) platform (Vera Rubin) to external manufacturers, and released a (CPU) rack system specifically designed for (AI) intelligent agents. The aim is to implant its own technical standards throughout the (AI) data center ecosystem, further consolidating its industry influence beyond the (GPU) market.

The disclosure occurred on the 13th during the Global Summit of the Open Compute Project ( OCP ) held in San Jose, California. NVIDIA's Vice President Ian Buck announced in his speech that the company would make available to external partners the design blueprints for the server racks and computing trays of the Vera Rubin platform, which include common specifications for power supply, liquid cooling, and network connections. NVIDIA stated that this move will lower the research and development barriers for server manufacturers and reduce the construction time and costs of data centers.

Industry analysts believe that there is a clear strategic intention behind this move: with the advent of the AI era of intelligent entities, the demand for infrastructure is shifting from GPU to CPU and network devices. NVIDIA is taking advantage of the openness of design standards to proactively build an infrastructure ecosystem centered around itself, known as AI.

800V DC solution to address rack power consumption bottlenecks

In terms of the power supply architecture, $NVIDIA ( NVDA.US )$ proposes in the new design scheme to increase the DC voltage in the rack to 800 volts. According to this blueprint, the conversion devices inside the data center will convert the externally inputted AC power into 800-volt DC power, which is then delivered to the server rack entrances.

Previously, data center racks typically used a direct current voltage of 48 to 54 volts. However, when the power consumption of a single rack reached around 1 megawatt, the current intensity at lower voltages became too high, posing a physical bottleneck that restricted capacity expansion. The high-voltage 800-volt DC solution can effectively alleviate this issue and is considered a key technical approach for AI data centers to cope with the continuously increasing energy consumption density.

Dedicated to CPU rack, targeting AI intelligent entity workloads

NVIDIA ($ NVDA.US ) has simultaneously released exclusive rack design specifications based on its own-developed $ Vera CPU . The existing $ Vera Rubin NVL72 system combines 72 $ GPU and 36 $ CPU chips, mainly targeting large-scale computing tasks; whereas the newly released $ Vera CPU rack can accommodate up to 256 $ Vera CPU chips, focusing on scenarios that require concurrent execution and management of multiple programs, with an emphasis on pure $ CPU processing capabilities.

NVIDIA, responsible for the Joe DeLaere of AI infrastructure solutions, indicates that the real bottleneck in the industry currently lies in the ability to manufacture and deploy AI factories fast enough. The open design specifications have been interpreted by the outside world as NVIDIA's proactive move to accelerate the construction of AI data centers, and it is also a strategic measure to lock in a voice during the formation period of industry standards.

By promoting the widespread adoption of its technologies and design specifications by data center equipment manufacturers, NVIDIA is seeking to extend its influence from GPU to a broader range of AI infrastructure.

The market worth hundreds of billions is accelerating its expansion.

The scale and prospects of the data center equipment market also provide ample competitive impetus for this standard dispute. Research institution IDC predicts that expenditures on equipment that meet OCP technical standards will reach approximately $800 billion in 2029, which is 2.7 times the institution's forecast from a year ago of $295 billion. By 2030, related expenditures are expected to further rise to $892 billion, about 5.8 times the actual expenditures of $154 billion in 2025.

In this rapidly expanding market, whoever establishes their design standards as the industry benchmark will gain a first-mover advantage in both the supply chain and ecosystem. NVIDIA's recent move is a direct response to this logic.

Edit / melody

Risk and Disclaimer Notice