AI Data Centers Embrace Liquid Cooling and High‑Efficiency Optical Modulators

Swagelok highlighted that fluid‑control reliability is a decisive factor for AI data‑center uptime. As GPU power rises to 1,200 W per chip and rack consumption can exceed 30 kW, liquid‑cooling systems must manage flow, pressure, material compatibility and sealing to prevent leaks that could disrupt services. Swagelok’s Taiwan representative Li Hong‑fu noted that many customers only seek help after a leak occurs, underscoring the need for comprehensive system design.

In South Korea, a KAIST research team together with KIST, KANC and Samsung Electronics developed a next‑generation Mach‑Zehnder optical modulator that combines InGaAsP semiconductor layers on silicon. The 500 µm device achieves world‑leading modulation efficiency and high‑speed bandwidth, enabling substantially more data to be transmitted per unit of power. Professor Kim said the technology “provides a realistic solution to the power‑consumption increase and data‑throughput bottleneck faced by AI data centers.” Both advances aim to improve energy efficiency and reliability of AI‑driven computing facilities.