Published: 22 October 2026

Nvidia has introduced a modular AI factory design that it says can reduce on-site deployment from six months to as little as 14 days. Founder and CEO Jensen Huang confirmed it in his keynote at GTC Berlin on 21 October.

Key points

• Nvidia says this brings facility cooling water use close to zero and could save 54 million litres a day at a 2GW deployment, roughly the daily use of 180,000 households.

• - Nvidia says its DSX MaxLPS software can produce the same number of tokens with up to 30% less power.

• - In an evaluation at its Icelandic AI factory, Nscale reported 49% higher inference throughput within the same 264kW power budget.

• - Huang cited 100,000 sites and about 100GW of distributed data centres, and noted cell towers would become distributed AI infrastructure.

• - Amkor aims to grow advanced packaging capacity by more than 300% by the end of 2028.

• Background and analysis. The story fits a broader pattern: AI infrastructure decisions made in Europe and other overseas markets are reshaping supply chains, procurement practices and competitive dynamics worldwide. With power equipment, land and grid connections all in short supply, the playbook for building at scale is being rewritten in real time — and the lessons travel quickly across regions.

• Our take. For anyone building or buying AI infrastructure, the practical lesson is to plan further ahead than instinct suggests. Whether the bottleneck is silicon, transformers or optical links, the winners are securing their constraints years before they need them.

• Source: Capacity, "Nvidia cuts AI factory deployment from six months to 14 days as Huang urges Europe to build its own AI", 21 October 2026. https://capacityglobal.com/news/nvidia-cuts-ai-factory-deployment/