ZERO-WATER CLOSED-LOOP DIRECT-TO-CHIP COOLING ARCHITECTURES FOR HIGH-DENSITY AI DATA CENTERS USING ELEVATED COOLANT TEMPERATURES

High-density AI data centers generate large heat loads while facing growing pressure to limit water use. Traditional evaporative cooling commonly requires 1.5-2.6 million gallons of water per megawatt each year. This paper presents the Multi-Stage Elevated Temperature Closed-Loop Architecture (MET-CLA), which combines direct-to-chip cold plates, dual-stage coolant distribution units, staged dry coolers, and predictive control at coolant temperatures of 40-45 C. A hybrid steady-state and transient thermal-fluid model, cross-checked against published elevated-temperature benchmarks, indicates that on-site Water Usage Effectiveness can be reduced to approximately 0.02-0.05 L/kWh in temperate and hot-dry climates, while Power Usage Effectiveness remains in the range 1.06-1.12. The architecture is compatible with current high-power GPU platforms and is offered as a practical systems design for lower water intensity.