Amazon Accelerates AI Data Center Builds With Liquid Cooling and Faster Timelines

The effort focuses on shortening the timeline from the start of building the outer shell of a data center to the first operational server room to under 35 weeks. That is well below typical industry standards. Titus also increases total compute capacity per site to about 68 megawatts from about 58 megawatts.

The company is expanding use of liquid-cooling systems that can be deployed more broadly than in previous designs. One internally developed technology called the In-Row Heat Exchanger allows support for both traditional workloads and demanding AI applications within the same facilities.

Internal targets call for a 15 percent reduction in cooling power consumption compared with current designs.

Titus is preparing data centers for new generations of AI servers including Nvidia GB200 systems and beyond. Newer versions will feature wider aisles to accommodate larger racks and cabling that extends from the front of servers. A newer iteration of the Titus design is scheduled to roll out in the first half of 2027 in time for even more powerful platforms with higher memory capacity and faster networking.

The project also seeks to reduce costs and environmental impact. Internal goals include a 10 percent reduction in cost per kilowatt of IT capacity while meeting 2028 carbon-emissions intensity targets and stricter noise standards. Facilities are being made more adaptable so that blocks of servers known as PODs can be scaled quickly without custom redesigns.

Power and Increasing Flexibility

Engineers are working to minimize unused electrical capacity referred to as stranded power by lowering minimum rack power requirements. This allows data centers to mix servers of varying power demands so that available electricity is used more fully rather than sitting idle when only high-intensity AI workloads are running.