Mark RadarMARK RADAR
中 EN
Event File AI AI Data Centers

AI Data Centers Confront Power and Cooling Constraints as Infrastructure Overhaul Accelerates

1 reports · First detected 2026-04-29 · Last active 2026-04-29

As generative AI models grow and GPU power consumption continues to rise, the chief data center bottleneck is shifting from computing capacity to power availability and cooling efficiency. Taiwan's Market Intelligence & Consulting Institute (MIC) said power demand from a single rack could reach hundreds of kilowatts, beyond what conventional power delivery and air-cooling systems can support. This is directly affecting the pace of data center construction and operating costs.

MIC's latest analysis shows the industry is accelerating the adoption of high-voltage direct current power systems (HVDC), battery backup units (BBUs) and liquid cooling, while evaluating new energy sources such as nuclear power to improve supply reliability and energy efficiency. The information did not disclose the report's publication date, investment amounts or deployment timelines, but demand for racks drawing hundreds of kilowatts has become a key threshold for infrastructure transformation.

All Coverage

1 original reports
ITHOME.COM.TW 2026-04-29
資策會MIC：AI資料中心擴展瓶頸從算力轉向電力與散熱，基礎設施正面臨轉型變革

The Backstory

The history behind this event
Liquid Cooling Becomes Key to Improving AI Data Center Compute Efficiency2026-02-24 · 1 reports · similarity 0.81

The Green Grid uses power usage effectiveness, or PUE, to measure the ratio of a data center’s total electricity consumption to the power used by its IT equipment. Generative AI is rapidly driving up the power density of GPU racks, pushing conventional air cooling closer to its thermal limits. If too much electricity is consumed by cooling, additional power cannot be converted proportionately into computing capacity, making liquid cooling a core component of AI data centers rather than an optional feature.

As of July 20, 2026, the latest reporting showed the industry’s focus shifting from securing more electricity to conserving it. Liquid cooling’s greater heat-transfer efficiency can support high-power racks while reducing cooling energy use, improving PUE and increasing effective computing capacity. The report identified no specific companies and disclosed no construction costs or project launch dates. For now, the key development is that cooling efficiency has become a direct constraint on expanding AI computing capacity.

Mark Radar｜MARK RADAR
Past Issues・by MarkReadFintech