AI accelerators at extreme heat fluxes demand radical improvements in junction-to-coolant pathways. This talk distils recent advances in two-phase direct-to-chip (D2C) into deployable design playbooks: engineered microstructures and topology-derived fins that advance the onset of nucleate boiling (ONB), widen thermal-reliability windows, and raise heat-transfer coefficients at practical pressure drops. We compare structured fins (e.g., zipper-inspired geometries) with straight channels, discuss as-printed AM surfaces and porous architectures that delay CHF, and outline facility-level implications—warm-water outlets (≈40–50 °C), reduced WUE, and compatibility with non-evaporative heat rejection. The session connects device-scale boiling physics to facility-scale outcomes, clarifying how two-phase D2C can enable stable, efficient thermal management for next-generation AI deployments.