Companies like Duos Edge AI and LG CNS are offering pre-fabricated, modular data centers that can be deployed in about six months versus the two to three years required for traditional builds. These self-contained units hold up to 576 Nvidia GPUs each, use liquid cooling for AI workloads, and can be networked together on a simple concrete pad. Duos has already signed a deal to deploy 2,304 GPUs for Hydra Host, while LG plans a 50-unit campus in Busan with over 28,000 GPUs. Cost advantages are also notable, with Duos claiming roughly half the cost per megawatt of larger facilities. HPE, Vertiv, and Schneider Electric are also entering the space, and the modular data center market is projected to more than double by 2030.

5m read timeFrom spectrum.ieee.org
Post cover image
Table of contents
GPUs shipped straight to youOn the grid, but under the radar

Sort: