NVIDIA unveils modular AI factories to cut data centre deployment to 14 days - Proactive financial news

NVIDIA is no longer selling silicon—it is selling deployment velocity as a product. The industry norm for a hyperscale AI cluster runs six to twelve months from site selection to first token. Compressing that to fourteen days means liquid cooling, power distribution, and network topology are factory-validated, leaving only crane work and cable termination on-site. This strips the value-add layer from system integrators like Dell, HPE, and Supermicro, and forces upstream cooling and switchgear vendors to adopt plug-and-play specifications. On compliance, a fourteen-day window collides structurally with EU building-energy codes and local grid-interconnection permitting. Concentrating the BOM in a single vendor creates a single-point-of-failure risk for sovereign AI programs. Competitively, AMD's EPYC-plus-Instinct and Intel's Gaudi can match raw FLOPS but cannot replicate the CUDA-and-NIM orchestration stack. The moat is no longer compute density; it is time-to-production. TPU and Trainium remain locked inside their parent clouds and cannot be delivered as standalone infrastructure. Within eighteen months, expect AI-factory-as-a-service pricing, accelerated sovereign deployments across the Gulf and Southeast Asia exploiting the fourteen-day window, and a repricing of data-center real estate as the bottleneck migrates from construction to land and power.