Who makes two-phase and microfluidic for AI data centers

A single-phase cold plate takes heat away by warming the water flowing through it. The two architectures in this part change that in different ways. Two-phase plates let the coolant boil at the hot surface, using the latent heat of vaporisation instead of a temperature rise, which moves much more heat per unit of flow. Microfluidic plates keep the coolant liquid but route far finer channels much closer to the die, shrinking the thermal path between silicon and fluid. Both exist for the same reason: heat flux at the die is rising faster than the area available to remove it, and a conventional plate eventually runs out of margin no matter how much water is pushed through it. Two-phase buys headroom in the fluid; microfluidic buys it in the geometry, sometimes by moving the channels into the package itself. The trade-offs are what to look for. Two-phase needs a fluid that boils in the right range, which ties it to the fluid supply in this layer and to the regulatory pressure on fluorinated chemistry; microfluidic channels are narrow enough that filtration and water chemistry become gating requirements rather than good practice. The suppliers here are mostly venture-stage specialists rather than the incumbents on the parent part, so the question on any given placement is whether the design has shipped at volume or is still in qualification.