On October 21, during the GTC Berlin conference, NVIDIA introduced the DSX POD, a modular reference design for AI factories. This innovative solution integrates the Vera Rubin system and its supporting infrastructure into modular containers, with thorough validation conducted prior to shipment. NVIDIA has announced that, assuming the site is prepared, the DSX POD can significantly reduce the deployment time of AI factories from six months down to as little as 14 days. This accelerated deployment enables customers to commence model training and inference operations much more swiftly. Moreover, the DSX POD incorporates a closed-loop liquid cooling and dry cooler system, virtually eliminating water usage for facility cooling and cutting energy consumption by up to 30%. NVIDIA intends to partner with global industry leaders such as Dell Technologies, HPE, and Lenovo to roll out these modular AI factories worldwide by early 2027.