Nvidia has unveiled a new AI data center system called DSX POD that aims to help companies start using their artificial intelligence (AI) systems and earn revenue faster. The company wants to reduce the time customers spend waiting for their AI infrastructure to become operational.
It is a ready-made design for an AI data center that can fit inside a structure about the size of a standard shipping container. The company introduced DSX POD at its GTC event in Berlin on Wednesday.
Nvidia AI system could save millions of litres of water
Nvidia says DSX POD can cut AI data center deployment time from six months to just 14 days. Nvidia says the system supports liquid cooling and uses dry-cooler designs that could save up to 54 million litres of water per day compared with a standard data center. The companies are expected to begin delivering the system in early 2027.
Because companies continue spending money while waiting for their facilities to become operational, delays can increase costs.
Nvidia CEO Jensen Huang said reducing the time needed to bring AI data centers online could be a major breakthrough. He said reducing the period between shipping the systems and making them operational from months to weeks, or even a month, would be a game-changer for AI factories . DSX POD units will be delivered to customer sites ready to be connected to the necessary power supply, cooling systems and networking infrastructure, according to Nvidia. Nvidia says DSX POD will work alongside traditional data centers rather than replace them. Huang said the system brings factory-built precision and speed to AI infrastructure construction. In a statement, he said the goal is to reduce deployment time from months to weeks so customers can expand their computing capacity and begin revenue-generating production sooner.
The system is designed to arrive at the site ready to connect to existing power, cooling and networking infrastructure. This could help companies start running their AI systems much sooner. Building a traditional data center can take months or even years. Companies cannot start earning revenue from AI services until their data centers are operational and begin processing AI tokens, which are units of data that AI models process to understand and generate information. DSX POD is designed as a faster and more flexible option for companies that need AI computing capacity quickly. Its modular design could allow businesses to bring additional computing infrastructure online without waiting for a complete traditional data center to be built. Speaking to Deirdre Bosa, Huang explained that installing these large AI facilities currently takes months. The new system is powered by Nvidia’s Vera Rubin platform. This approach is intended to reduce the amount of work required at the deployment stage. Companies could use the system to start running AI applications and generating revenue while their permanent, brick-and-mortar data center facilities are still being built. DSX POD is also designed to reduce water consumption. Faster data center construction could help Nvidia sell more chips. Data center builders face delays in getting their facilities ready for use. If companies can bring their infrastructure online sooner, they may be able to install and use AI processors more quickly, helping ease one of the bottlenecks in the industry’s expansion. A shortage of electricity is another major challenge for the AI industry. Data centers need large amounts of power to operate their processors, and building the power plants required to supply that electricity can take much longer than constructing the data centers themselves. This means faster deployment of data center systems will not solve every problem facing the industry. Nvidia is working with several companies to bring DSX POD to the market. Its partners include Anduril, Armada, Bull, Crusoe, Dell Technologies and Foxconn, among others.
