The AI Data Centers That Fit on a Truck

Companies like Duos Edge AI and LG CNS are developing modular, pre-fabricated data centers that can be rapidly deployed for AI workloads, in contrast to traditional data centers that take years to build.

Details

Traditional data centers are built with a protective shell of steel and concrete, which is expensive but necessary to house the expensive hardware and infrastructure inside. However, this construction process can take years, which is a problem for companies looking to quickly deploy AI hardware. To address this, companies like Duos Edge AI and LG CNS have developed modular, pre-fabricated data centers that can be deployed in just months. These self-contained boxes are about the size of a shipping container and can be transported by truck. They contain racks of GPUs and supporting hardware, similar to what would be found in a traditional data center. The modular approach allows the data centers to be quickly assembled on-site, without the need for extensive construction. Duos Edge AI's pods can contain up to 576 GPUs each, with the option to scale up to over 4,600 GPUs across multiple pods. LG CNS is also developing a similar modular data center that can support over 4,600 GPUs in a single unit. These modular data centers can be placed on a simple concrete pad, rather than requiring a large, purpose-built facility. This greatly simplifies and speeds up the deployment process compared to traditional data centers, which can take 2-3 years to build. The modular approach provides flexibility and speed, allowing companies to quickly scale their AI infrastructure as needed. It also helps address the challenge of finding available data center space and getting the necessary permits, which can often delay AI deployments.