Nvidia has introduced a container-type AI system design that reduces the time required to build artificial intelligence (AI) data centers from six months to just 14 days. This approach involves pre-assembling computing equipment, cables, and cooling systems in the factory before transporting them in containers to the installation site.
According to CNBC, Nvidia announced the new container-type AI system, called the 'DSX Pod AI Factory,' during the Nvidia Developer Conference 'GTC Berlin 2026' held on October 21 in Germany.
The DSX Pod integrates seven racks of the next-generation AI system, 'Vera Rubin,' into a single container. Similar in size to a standard cargo container, it weighs approximately 28 tons. The design allows for cable connections and liquid cooling systems to be installed in the factory, enabling transportation by truck on regular roads.
This contrasts with the traditional method of constructing data centers, which requires significant time for on-site installation and connection of servers and cooling equipment. Nvidia stated that its new system can cut the installation time for AI data centers from six months to just 14 days.
Jensen Huang, Nvidia's CEO, emphasized the importance of reducing the construction time for AI data centers during his keynote address. He pointed out that tokens cannot be produced until the AI factory is operational, explaining that completing equipment assembly and validation in the supply chain can significantly shorten the time from arrival on-site to token production. This aims to enable customers to activate their AI production facilities more quickly and generate revenue sooner.
The product pricing was not disclosed during the announcement. Nvidia plans to collaborate with manufacturing partners, including Foxconn, Gigabyte, Lenovo, Hewlett Packard Enterprise (HPE), Quanta, and Pegatron, to produce the product, with shipments expected to begin in early 2027.
CNBC noted that this announcement reflects Nvidia's expansion from selling individual semiconductors, such as graphics processing units (GPUs), to supplying entire systems that make up AI data centers.
In a recent podcast, CEO Huang also revealed that the value of products supplied by Nvidia for data center power capacity of one gigawatt (GW) has increased from approximately $18 billion to $20 billion in the past, and is now over $40 billion. He anticipates this trend will continue.
Additionally, Nvidia announced plans to expand production capacity in collaboration with six partners, including Samsung Electro-Mechanics, to alleviate bottlenecks in AI chip supply. Ibiden, Samsung Electro-Mechanics, and Unimicron will assist in securing substrates, while Amkor Technology aims to quadruple chip packaging supply by the end of 2028. The semiconductor testing capacity is set to double next year compared to this year.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.