AI Demand Is Forcing a Rethink of Data Center Power, Cooling
AI's voracious demand is causing 50% mid-project power hikes and 5X growth in liquid cooling by 2030.
The explosive growth of generative AI and agentic AI is creating a domino effect across data center infrastructure, forcing a fundamental redesign of power, cooling, and construction. At the Data Center World conference, Aligned Data Centers CEO Phill Lawson-Shanks revealed that a single project's power requirements jumped 50% mid-construction due to AI and GPU demands, necessitating a complete restructuring for cooling and grid power. Major components like transformers face severe backlogs, pushing operators like Aligned to buy equipment 2.5 years in advance and secure land based on the availability of 'strand power.' Despite these constraints, Omdia analyst Shen Wang asserts projects are delayed by only months, not derailed, as the relentless demand for AI compute ensures 'AI factories will move forward somehow.'
Cooling technology is undergoing a parallel revolution. Omdia's Shen Wang projects shipments of liquid-cooling cold plates will skyrocket from 8 million in 2025 to 356 million by 2030, with liquid cooling capacity set to double air cooling capacity by the end of 2026. However, air cooling remains essential for components like memory and power shelves. Vertiv's Scott Armul emphasized the need for a holistic, modular approach, integrating power, compute, and cooling into standardized building blocks. This allows for full-environment simulation to optimize efficiency, a critical capability as liquid cooling is deployed at an unprecedented gigawatt scale across massive AI campuses.
- AI demand causes 50% mid-construction power requirement hikes, forcing infrastructure redesigns.
- Liquid-cooled chip shipments projected to surge 5X, from 8M in 2025 to 356M by 2030.
- Operators combat 2.5-year equipment backlogs by bulk-buying in advance and securing land with 'stranded power.'
Why It Matters
The physical limits of power and cooling are becoming the primary bottleneck for AI progress, impacting costs and deployment timelines.