As AI servers push rack power toward and beyond the 1 MW threshold, direct-to-chip (DTC) liquid cooling has become unavoidable — but not all approaches are equal.

This session demystifies the design, cost, and maintenance differences between single-phase and two-phase DTC cooling systems for both greenfield AI factories and brownfield retrofits.

Panelists will tackle:

• System Architecture: How 1P and 2P loops differ from CDU to cold plate, and how each integrates with existing facility water systems.
• Performance & Reliability: Heat flux thresholds, pump reliability, fluid management from water-based 1P loops to dielectric refrigerant handling in 2P systems, leak risk, and serviceability in production environments.
• Installation & Maintenance: Lessons from hyperscale and OEM pilots — the practical realities of install labor, fluid management, and long-term O&M.
• Retrofit Viability: When two-phase pays off, and when single-phase remains the smarter path.
• NVIDIA’s Role: How NVIDIA’s reference designs, qualified vendors, and rack standards are shaping coolant selection, manifold design, and CDU integration.

This roundtable will provide an unvarnished view into the operational trade-offs behind liquid cooling architectures — helping operators, engineers, and investors decide which system aligns with their density roadmap, risk tolerance, and retrofit constraints.